Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

A Hybrid Method for Recommendation Systems based on Tourism with an Evolutionary Algorithm and Topsis Model

A Hybrid Method for Recommendation Systems based on Tourism with an Evolutionary Algorithm and... FUZZY INFORMATION AND ENGINEERING 2022, VOL. 14, NO. 1, 26–50 https://doi.org/10.1080/16168658.2021.2019430 A Hybrid Method for Recommendation Systems based on Tourism with an Evolutionary Algorithm and Topsis Model a b c Saman Forouzandeh , Mehrdad Rostami and Kamal Berahmand School of Mathematics and Statistics, Faculty of Science, University of New South Wales, Sydney, NSW, Australia; Centre of Machine Vision and Signal Processing, Faculty of Information Technology, University of Oulu, Oulu, Finland; School of Computer Sciences, Science and Engineering Faculty, Queensland University of Technology (QUT), Brisbane, Australia ABSTRACT ARTICLE HISTORY Received 16 September 2020 Recommender systems have been pervasively applied as a technique Revised 11 July 2021 of suggesting travel recommendations to tourists. Actually, recom- Accepted 14 December 2021 mendation systems significantly contribute to the decision-making process of tourists. A new approach of recommendation systems in KEYWORDS the tourism industry by a combination of the Artificial Bee Colony Recommender systems; ABC (ABC) algorithm and Fuzzy TOPSIS is proposed in the present paper. algorithm; topsis model; A multi-criteria decision-making method called the Techniques for tourism; Facebook social network Order of Preference by Similarity to Ideal Solution (TOPSIS) has been applied for the purpose of optimizing the system. Data were gath- ered through a 1015 online questionnaire on the Facebook social media site. In the first stage, the TOPSIS model defines a positive ideal solution in the form of a matrix with four columns, which indicates factors that get involved in this study. In the second stage, the ABC algorithm starts to search amongst destinations and recommends the best tourist spot to users. 1. Introduction One of the inevitable aspects of human life is decision making. Faced with many options, lack of proper expertise, and time constraints resulted in hectic human judgments. Recom- mendation systems are therefore provided to resolve these issues [1–3]. The knowledge discovery methods and statistical techniques are applied by recommendation systems for suggesting suitable items to clients [4–6]. These systems operate on the basis of the pref- erences and behavior of the user to recommend an option that is strongly adapted to the preferences of the client to make a proper decision [7–9]. In several different domains, rec- ommendation systems are utilized and recently implemented in the tourism industry to suggest a suitable destination for visitors [10,11]. The objective of travel recommendation systems is to match consumer requirements with the characteristics of tourism and leisure resources or attractions [2]. For recommending the best place, this system needs some information about tourist attractions and destinations, tourists visiting schedules, rating score by other users for each destination, priorities, and preferences of travelers, the length CONTACT Saman Forouzandeh s.forouzandeh@unsw.edu.au © 2022 The Authors. Published by Taylor & Francis Group on behalf of the Fuzzy Information and Engineering Branch of the Operations Research Society, Guangdong Province Operations Research Society of China. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons. org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. FUZZY INFORMATION AND ENGINEERING 27 of travel days, starting and the ending time of visiting tourist attractions etc. Through this type of information, the itinerary is offered to customers. Therefore, in suggesting the best travel to users, different criteria should be taken into account. The TOPSIS model can cover various criteria in providing recommendations to users, and it is used in this research to solve the problem. In order to cover users’ multi-criteria decisions, the authors use the TOPSIS model, and their preferred criteria are allocated impact factors depending on their preferences. Afterward, according to client pri- orities, the ABC algorithm is applied to search for qualified tourist places. The objective of applying the ABC algorithm is to discover different places in the selected range of the user, following their preferences with high ratings according to their criteria. Applying the ABC algorithm does the search for tourist places for users in three stages, and also the search for places is based on defined criteria in the TOPSIS model. In essence, the recommendation system offers personalized information for users regarding the trip plan [12,13]. Also, a couple of factors play an important role in the itinerary, including time managing and selecting the best option based on tourist prefer- ences [14,15]. There are plenty of tourist destinations, activities, and transportations in every tourist spot, and there are various types of travel plans as well. Since travelers have different preferences and flavors, the uniform travel plan would not be proper for all tourists. There- fore, a flexible and customized system is required to input some parameters by the user, then an excellent choice is offered to tourists to provide them a marvelous time during their vacation [16]. In this regard, the significant role of social media can be drawn con- cerning tourism development. For instance, social media users have a chance to check the travel experiences of their friends [17]. Moreover, mobile phones significantly contribute to the recommendation for tourists through discovering a current location of users, rec- ommending a particular near destination through the use of recommendation techniques [14]. Due to the growing number of options for users to find a tourist destination, Fuzzy logic systems were produced in this field. All possible choices were categorized by a multi- criteria decision-making system, providing a situation for users to select an appropriate solution [18,19]. The objective of the present research is to apply the TOPSIS model and ABC algorithm for optimizing tourist destinations according to specific principles and finally recommend these options to visitors. The article is structured as follows. Section 2 provides a literature review on recommen- dation systems and methods in the tourism, Section 3 addresses the multi-criteria decision method, and Section 4 outlines the ABC algorithm. Section 5 discusses the methodology of the research, Section 6 includes analysis and results, and the conclusion is presented in Section 7. 2. Literature Review One of the purposes of recommendation systems is detecting the user preferences and suggesting items or services for meeting users’ requirements and favors [20–22]. These systems are being developed as an efficient research area and possess applications in var- ious fields, including medical science, business, education, e-commerce, etc. [23–28]. The primary objective of recommendation systems is proposing a considerable suggestion of goodsorservicestousers [28–30]. There are several types of research in this area that are carried out in the tourism industry. For instance, Wang and Lee studied a methodological 28 S. FOROUZANDEH ET AL. technique for assessing smart tourist attraction (STA) tourist preference; afterward, a fuzzy comprehensive calculation strategy and analytical hierarchy procedure were introduced to the Zoo’s STA assessment, which is one of China’s tourist attractions. After evaluating the data and conducting the fuzzy assessment, they distributed 680 questionnaires to col- lect information from zoo visitors, and findings derived primary STA evaluation factors, including intelligent tourism management, smart information systems, smart forecasting, etc. [31]. Some researchers have provided new technologies, including Augmented Reality (AR), to satisfy visitors with smart recommendations. They created an AR tourist application for smartphones. In order to evaluate the results, 26 in-depth interviews with 26 international and local tourism and thematic study were used. The results revealed that the knowledge generated by social networks containing feedback, attractions, and destinations was com- mon with travelers and should be taken into consideration in AR applications to encourage visitors to use this modern technology and recommend tourist attractions and urban her- itage to them [32]. Some considerations could play an essential role in tourist choices. For example, one research suggested a recommendation framework focused on the users’ flavor, the destinations they have visited previously, and also their demographic catego- rization. The authors presented a recommendation framework that could be provided to a single user or a group of users. The questionnaires were distributed among 60 people, asking about overall preferences, demographic information, and the places they visited. The findings revealed that the combination of three basic recommendation approaches: demographic, content-based, and general like filtering, was conclusive in recommending the alternative for a single or group of users [33]. A recommender system was developed to recommend hotels to consumers in another study performed by Nilashi and Ibrahim. In order to fulfill the needs of a consumer on data originating from a similar user, this system used recommendation techniques. Through dimensional reduction and estimation tech- niques, they suggested a novel hybrid approach. In order to estimate tourist preferences, the proposed model was focused on data mining techniques, EM algorithms, cluster- ing techniques, and ranking items in user profiles, recommending a suitable option for them [34]. Another research used opinion mining techniques to recommend a particular hotel based on information from clients. Opinion mining methods were applied to data by har- vesting and then classifying hotel reviews from the Internet. The collection of data consists of 3000 reviews posted on tripadvisor.com by users. The findings of this analysis showed that both visitors looking for accommodation and hotels are seeking to develop their ser- vices by utilizing these strategies [35]. In the research study conducted by Braunhofer and Lamche, a prototype of a recommender system for a mobile point of interest (POI) was presented. This system operated on the basis of text mining and user location and five fac- tors that might touch a user-perceived, including travel time accessible to the user, visit time, weather, and the POI visit history of the user. The findings of this analysis indicate that user’s preferences and various forms of heuristics were derived by the proposed model to estimate the total usefulness of the POI recommendation to the target user in the current contextual position [36]. Multi-criteria TOPSIS decision model and fuzzy logic were applied in recent years to evaluate the quality of travel websites from the consumer’s viewpoint [37]. The deci- sion matrix has been created in this method, and certain parameters are regarded in FUZZY INFORMATION AND ENGINEERING 29 terms of the purpose. In this technique, the ranking procedure is centered on the short- est distance to the best positive answer and the farthest distance from the negative answer [38]. A study indicates a TOPSIS-based combination of genetic and fuzzy for a real-time route recommendation system for tourists in tourist hot spots in this regard. This analysis carried out three phases: data collection, TOPSIS-based data phase regard- ing route optimization, and the generation of new routes on the genetic algorithm to choose the best route. The data was gathered through questionnaires provided to self- driven travelers. The results of this study showed that this personalized route recom- mendation system in real-time could enable the self-drive tourists to finish their trip more effectively, leading to reduce the congestion of traffic in the tourist hotspot. In addition, this proposed model saves average visiting time and satisfies visitors’ specific preferences [39]. Another analysis applied a fuzzy number and a TOPSIS model to quantify service quality and rank each hotel’s service performance, respectively. The authors distributed a survey to hotel visitors to gather data on the level of service at three Grand Canaria Island hotels. After implementing the fuzzy set theory on data, they concluded that fuzzy numbers act as an appropriate method to resolve the vagueness of views relevant to the subjective judgments of human beings. Also, detecting the customer perception and preferences about service quality in the hotel are two key points to boost marketing effort and improve the perfor- mance of the recommendation systems [40]. In hotel location selection for visitors, fuzzy logic is also applied. Fuzzy multi-criteria decision making (FMCDM) has been performed in research conducted by Clou to help travelers in selecting the appropriate hotel. Twenty- one criteria that play an important role in consumer satisfaction in hotel selection were derived by the authors. The results of this study indicate that FMCDM would enable tourists to choose a suitable hotel location [41]. The Artificial Bee Colony algorithm is another feature used in the recommendation system (ABC). The paper offered a modern recommendation system centered on a col- laborative filtering recommendation system on the basis of the ABC algorithm and the K-means clustering algorithm. It actually contributes to better clustering accuracy. The ABC algorithm is often used for the purpose of solving this clustering technique’s local optimal problem. The results revealed that this collaborative filtering system focused on the algorithm of tourist clustering showed better performance than several other approaches of recommendation [42]. A recommendation system based on the ABC algorithm on Facebook, which is the most prominent social network worldwide, was pro- posed by researchers in another study. The data gathered by questionnaires, the explo- ration of popular topics, and the difficulty of materials were valued. In the next level, a system optimized its performance using the ABC algorithm. The results showed that this approach enhanced the implementation time and made the recommendation sys- tem suggestions accurate in an effective way [43]. Furthermore, the ABC algorithm can boost the effectiveness of the recommendation system. Arup Roy, in a research study, developed a recommender system using the ABC algorithm and the Simulated Anneal- ing (SA) technique. Arup Roy applied the Irish TripAdvisor datasets, including 29,799 reviews of Irish hotels by 21,851 users. Results argued that it might be useful to use the ABC algorithm to identify an optimal or near-optimal choice for user recommendation advice [44]. 30 S. FOROUZANDEH ET AL. 3. Multi-Criteria Decision Method Multi-criteria decision analysis is a method in decision-making that estimates multi- incompatible criteria. There are two key aspects of this method; multi-criteria optimization and multi-criteria decision analysis. Although multi-criteria decision-making concentrates on multi-condition disciplines with few solutions and in an uncertain situation, multi- criteria optimization covers more than one goal [45]. It is necessary to compare all options comprehensively in the multi-criteria decision process and take into consideration all cri- teria. All responses and ideal solutions have been described in the numeric format in this approach [46]. The TOPSIS model was offered in 1981 [47], which is one of the best multi- criteria decision-making models. This model is on the basis of selecting the farthest negative ideal solution and the nearest positive ideal solution between the alternatives that is the best solution [41,48]. TOPSIS model includes the m∗n matrix in which ‘m’ implies the solu- tions and ‘n’ refers to criteria [41,48]. All qualitative measures should be converted into quantitative criteria for calculating mathematic techniques, as presented in Table 1. TOPSIS consists of several steps, which are expressed as follows: In the first step, a decision matrix is formed according to the desired indicators and cri- teria that the weight of each of these indicators may differ from each other. The decision matrix is m × n, in which m is equal to the desired indicators and criteria and is equal to m = 4 in the present research because four criteria are considered to evaluate tourist loca- tions. n is also equal to the number of tourist places in each city; the amount of which depends on the considered city. Matrix (1) is formed based on m×n matrix as follows (decision matrix): ⎡ ⎤ a a ... a 11 12 1n ⎢ ⎥ a a ... a 21 22 2n ⎢ ⎥ ⎢ ⎥ .. . . ⎢ ⎥ A = (1) ⎢ ⎥ ij ⎢ .. . . ⎥ ⎢ ⎥ ⎣ ⎦ .. . . a a ... a m1 m2 mn unscaled matrix is one of the most important steps in multi-criteria decision-making and is used to unscaling the decision matrix. In the present research, the Euclidean method is used to unscale the decision matrix. For normalization and creating the standard matrix, Equation (2) is defined as follows [49,50]: ij r =  (2) ij kj k=1 In Equation (3), positive and negative ideal values are specified so that the distance of other points from these positive and negative ideal points can be determined. Where ω is the Table 1. Converting qualitative criteria to quantitative criteria. Qualitative Criteria Too Low Low Medium Much Too Much Quantitative Criteria 1 3 5 7 10 FUZZY INFORMATION AND ENGINEERING 31 weight of the j-th criterion; w =1(3) i=1 The criteria and measures do not possess the same prioritized index, and the index crite- ria table solves this impediment. In this regard, a high indexed value has a high weight. The matrix, presented in the following, illustrates the value of the standard index for each criterion: ⎡ ⎤ w r w r ... w r 1 11 2 12 2 1n ⎢ ⎥ w r w r ... w r ⎢ 1 21 2 22 n 2n ⎥ ⎢ ⎥ .. . . ⎢ ⎥ v = ⎢ ⎥ (4) ij ⎢ .. . . ⎥ ⎢ ⎥ ⎣ .. . . ⎦ w r w r ... w r 1 m1 2 m2 n mn ∗ − ∗ Determining a value for S and S calculated based on (5) and (6) formula. S represents a i i i distance of solution from ideal positive criteria and S represents a distance from negative solution criteria. There are two methods for calculating the distance of other points from the positive and negative ideal points: The euclidean method and the block method. The Euclidean method is used in the present research, Equation (5) calculates the distance of option i from the positive ideal, and Equation (6) calculates the distance of option i from the negative ideal. ∗ ∗ S = (v − v ) (5) ij j=1 Figure 1. TOPSIS model implementation steps. 32 S. FOROUZANDEH ET AL. − − S = (v − v ) (6) ij i j j=1 Where, i and j belong to [1..m] and [1..n], respectively. The relative closeness of i-th alterna- tive S with respect to S∗ described as follows: i i ∗ i C = (7) − ∗ S − S i i It is evident that the rating of options was calculated by C ∗. Equation (7) is used to calculate the relative closeness of options to the ideal solution. Finally, the options are ranked based on their closeness to the ideal solution. Accordingly, the TOPSIS model consists of six steps, as shown in Figure 1: Algorithm 1 TOPSIS Method in the 6 steps as below: Algorithm 1: TOPSIS method Step 1: Calculate the normalized decision matrix (nij) as: ij N = , i = 1, ... , m; j = 1, ... , n. ij i−1 ij Step 2: Calculate the weighted normalized fuzzy decision matrix (r )as: ij x = w × n , i = 1, ... , m ; j = 1, ... , n. ij j ij Step 3: Obtain the positive (PIS) and negative (NIS) ideal solutions as: max(r |i = 1, ... , m), IFj ∈ B ij + + + PIS ={r , ... , r , ... , r }, where v = PIS = 1 j n j min(r |i = 1, ... , m), IFj ∈ C ij max(r |i = 1, ... , m), IFj ∈ B ij − − − − {r , ... , r , ... , r }, where v = 1 j n j min(r |i = 1, ... , m), IFj ∈ C ij Step 4: Calculate  the distances of each alternative i in relation to the ideal solutions as: + n + S = (r − r ) , i = 1, ... , m ij i j − − S = (r − r ) i = 1, ... , m ij i j j Step 5: Calculate the closeness coefficient of the alternatives (CCi) as: CC = i + − S +S i i Step 6: Sort the alternatives in descending order. The highest CC the value indicates the best performance in relation to the evaluation criteria. 4. ABC Algorithm The artificial Bee Colony (ABC) algorithm is an optimization method to simulate the foraging behavior of honey bees. Karaboga introduced this algorithm in 2005 [51–53]. Honey bees can spread far from their hive, searching for food. The colony of artificial bees is classified into three groups [54]: employed bees, onlooker bees, and scout bees. The search mecha- nism for food begins with the bees, which are employed. In their memory, the employed bees investigate food sources inside the neighborhood; in the meantime, they share the information from these sources of food to the onlooker bees. Onlookers choose a source of food from the food sources selected by them within the neighborhood. An employed bee from which the source has been discarded becomes a scout and begins spontaneously FUZZY INFORMATION AND ENGINEERING 33 to look for a new food source. The fulfillment of initializing a value in the ABC algorithm is done as the following [54–56]. 1. Employed bees: in this level, artificial bee X forms a new candidate solution V in the i i neighborhood of its current location. Selecting a location of new food source is carried out as follows: V = x + Ø ( x – x )j = 1, 2, ... ., D ; k = 1, 2, ... , SN (8) ij ij ij ij kj Here v = [v ,v ... ., v ] represents current bee’s location and x = [x ,x ... ., x ] indi- i i1 i2, in i i1 i2, in th cates a i location of bee’s. k(k#j) implies a random number index chosen from the set {1,2, ... ,n} and SN presents a number of artificial bees, and D represents the solutions. Ø ij refers to a random number within [−1, 1]. Choosing a random x applied as the following formula: x = L + rand(0, 1) × (U − L ) (9) ij j j j Here U and L are the upper bond and the lower bond of the X , respectively, and rand () j j i denotes a random number within [0,1]. Subsequent to selecting the location of a new food source, the optimization value should be measured by the following formula: f ≥ 0 1 + f fit = (10) 1 + abs(f ) f < 0 i i In this study, on the basis of increasing repetition of employed bees, their search range is increased, and each time they search for 3 points for a similar purpose improves the likelihood that they can reach the optimal points. Therefore, at each output, the most appropriate point is optimized compared to the other two points and closer to the aim of the user (Max Points). Figure 2. ABC algorithm implementation step. 34 S. FOROUZANDEH ET AL. 2. Onlooker bees: employer bees share the food source knowledge with onlooker bees via waggle dance after the search phase is completed by them. At this level, the informa- tion collected from the bees employed is estimated by each onlooker bee, selecting a food source with a likelihood correlated with food resources. The following formulas present this probability: fit p = (11) SN fit j=1 D = [v p , v p , ... , v p (12) i i1 i i2 i in i th Where fit implies the fitness value of the i solution in the swarm. It is clear that the th higher probability of i is obtained by the better solution i. After picking all food sources by onlooker bees, the location of the neighbors’ food source is measured by Equation (9) and this cycle will be continued until getting the required goal. In Equation (11), the enhance- ment of the function of the onlooker bees is relevant to the location of their movement, and their search for food commences from the location where the employer bees were in that position (the newest) last time (V = [v ,v , ... ., v ]). The onlooker bees are then around i i1 i2 in the locations where the employer bees seek and the neighborhood between reviewed points and this local search for improvement, and in this way, further points can be found. 3. Scout bees: In the ABC algorithm, some bees give up their solution method, becom- ing scout bees to search randomly in the domain if the target cannot be reached after repetitive default processes. Applying this type of method will increase search efficiency. S = S – (V and Di) (13) ij b b Equation (12) represents that the Sb discover the points that the employed bees and onlooker bees (Vij and Di) have not visited previously. Accordingly, the ABC algorithm performs the search in three steps according to Figure 2: The optimized ABC algorithm is as the Algorithm 2 [57]: 5. Methodology The following study tries to present a recommender system for recommending an attrac- tive tourist zone for travelers. For this purpose, the multi-criteria decision analysis TOPSIS model has been applied that works based on fuzzy logic as well as a tourist destination ABC algorithm has been utilized for searching. The method of this research consists of two main parts. In the first part, the TOPSIS model has been used to define and model the problem, and in the second part, the ABC algorithm has been applied to find tourist spots to suggest to the target user. The procedure is as follows: studying the number of five cities that have many tourist places, using the experi- ence of users in traveling to the desired tourist destinations through a questionnaire and four criteria are set to assess the quality of each of them and users are asked to rate each of the criteria in the form of a questionnaire, and qualitative criteria have been used to make the questionnaire simple for users. Subsequently, the proposed method is presented FUZZY INFORMATION AND ENGINEERING 35 Algorithm 2: Artificial bee colony algorithm 01. Randomly generate SN points in the search space to form an initial population. 02. Evaluate the objective function values of the population 03. cycle = 1, trial = 0, i = 1, 2, ... , SN 04. repeat {The employed bee phase} 05. for i = 1to SN do 06. Generate a candidate solution V by Equation 7 07. Evaluate f(V ) 08. If f(V ) < f(X ), set X = V , trial = trial + 1, otherwise trial = 0 i i i i i i i 09. end for 10. Calculate the probability values p by Equations 10, 11 {The onlookers phase} 11. t = 0, i = 1 12. while t < = SN do 13. if rand(0, 1) < p then 14. Generate a candidate solution V by Equation 7 15. Evaluate f(V ) 16. If f(V ) < f(X ),set X = V , trial = trial + 1, otherwise trial = 0 i i i i i i i 17. t = t + 1 18. end if 19. Set i = i + 1, if i = SN,set i = 1 {The scouts phase} 20. If max(trial ) > limit and X –(V and D )by then i i ij i 21. Replace X with a new randomly produced solution by Equations 8, 12 22. end if 23. Memorize the best solution found so far 24. cycle = cycle+1 25. until (cycle = Maximum Cycle Number) using the TOPSIS model. Since there are several tourist places to suggest to the user, sev- eral ideal points (places with the highest score) should be defined as the best options to offer to the target user, and finally, a search should be done (using the ABC algorithm) for places that have the shortest distance to these ideal locations. For this purpose, the TOPSIS model is used; the procedure is first to convert qualitative values  into quan- titative values and then create a decision matrix numerically and based on the desired criteria. The TOPSIS method is one of the Multiple Attribute Decision Making (MADM) meth- ods that rank options. In this method, two concepts of ‘ideal solution’ and ‘similarity to ideal solution’ are used. The ideal solution is the solution that is the best choice in every way that we trying to approach it. In order to measure the similarity of a design (or option) to an ideal (places with the highest score) and anti-ideal (places with the lowest score) solution, the distance of that points (places) from the ideal and anti-ideal solu- tion is measured. The points are then evaluated and ranked based on the distance them from the ideal solution and anti-ideal solution. Accordingly, in this research, tourist places ranked and in the ideal case, each feature has 10 scores and in the worst case, it has 1 score. Ideally, each place has a maximum score of 40 and we are looking for places that have the least distance with the number 40. The method is based on the concept that the found place should have the shortest distance from the positive-ideal solution and the longest distance from the negative ideal solution. TOPSIS defines an indicator called similarity with the ideal positive solution and distance from the ideal negative solu- tion. The ABC Algorithm then searches for places with maximum similarity to the ideal 36 S. FOROUZANDEH ET AL. Figure 3. Research methodology. positive solution. To measure the similarity of a place with an ideal and non-ideal state, we consider the distance of that point from the ideal and non-ideal solution. In this research, we also use the ranking matrix and the distance of each place from the ideal state. Given that multiple tourist destinations may be offered to the user, multi-criteria decision-making is used, which is made possible by TOPSIS. Hence, in each of the stud- ied cities, tourist locations are ranked based on the TOPSIS model in the form of decision matrices and by converting quality criteria into quantitative values (based on users’ com- ments), and then places with the highest points are considered as ideal points. Accordingly, the ranking of each tourist place is determined according to Table 1. Afterward and in the second step, the points that have the shortest distance with ideal points in the decision matrix are searched using the ABC algorithm and according to the selection of the desired city by the user, and finally, in each searching process with ABC, three tourist places with higher points are found according to the desired criteria and then suggested to the user. The case study in the present study is about some tourist cities in Iran, and the research methodology steps are depicted in Figure 3. Our research methodology is categorized into eight levels as follows: FUZZY INFORMATION AND ENGINEERING 37 1. Selecting some tourist cities: The methodology procedures begin with selecting some tourist cities that each city has a specific tourist attraction. The purpose of this level is that when travelers go to one of these cities, our system will recommend the best tourist destination in this city based on their preferences and experiences. The authors nominate five cities at this level that have a great number of tourist attractions. These cities are Tehran, Isfahan, Shiraz, Mashhad, and Orumieh. 2. Selecting tourist destinations for each city: Each city has several tourist spots. At this level, ideal destinations highly matched with tourists’ preferences are considered to recommend to the user in the subsequent stages. 3. Applying the TOPSIS model: After selecting touristic cities and destinations for each city, the TOPSIS model has been carried out to form a matrix for each city, consisting of four columns as follows: • Distance • Road Condition • Traffic Congestion on The Road • Tourist Attractions of the City Essentially, all responses and optimal solutions have been clarified in numeric format; thus, it is necessary to convert qualitative parameters to quantitative measures in the pre- vious section and based on Table 1. According to the TOPSIS model, the qualitative variables were converted into a quantitative element rating based on Table 1 from 1 (the lowest point) to 10 (the highest point). The ideal matrix is a matrix that produces 10 for all parts, so the nearest rate to the 10-band score implies an ideal solution, which is one of the good options to recommend to the user. 4. Distributing questionnaires on Facebook: At this level, each tourist attraction is ranked by the user through an online questionnaire on the Facebook website. The users give the mark for each destination based on four criteria mentioned in the previous level. The questionnaire is divided into four sections: distance, road condition, traffic conges- tion, and tourist attraction. The authors arranged an online questionnaire uploading it on the Facebook social media website through several accounts and asked respon- dents to rate each item based on their travel experience to those cities. Finally, 1015 questionnaires were gathered from respondents. 5. Creating a matrix for each city on the basis of TOPSIS outcome: At this level, matrices are created for every single city according to questionnaire outcomes taken from users. Matrices are formed based on the following formula: 5 4 M = C && L (14) n i=[A→1,3,5,7,10B→1,3,5,7,10C→1,3,5,7,10D→1,3,5,7,10] n=1 i=4 Where implies each city, and i = 1 means that a minimum of one city must be picked n=1 and the maximum number of cities is equal to five. Indicates four filed for the quality i=4 of tourist attractions. The upper bond of 4 means that all 4 fields must be customized. An AND operator exists between C cities and L tourist destinations since our function is only n i true when both values are true. Furthermore, L has four fields; A, B, C, D, presenting four i 38 S. FOROUZANDEH ET AL. criteria for destination quality, and every single field of the matrix can be customized by 1,3,5,7,10 values; the best quality is obtained by higher numbers. As mentioned, in relation to the weights intended for tourist places, we used the quality criteria according to Table 1, then we convert them to have quantitative criteria. Weighing and scoring for each tourist place is based on the dataset and Tripadvisor site. Also, quality criteria have been selected in such a way that users can easily evaluate the quality of the tourist places. Then, a score is considered for each quality criterion, and through this, the weighing of each tourist place is done. 6. Selecting a specific city by the user: The user chooses one of the five cities in this level, and then the recommender system starts to suggest a tourist destination to the user. 7. Applying the ABC algorithm for a searching destination with minimum distance and an ideal matrix: After choosing a city by the user, the recommender system starts to search and analyze tourist destinations in the selected city. The ABC algorithm is used to fulfill the searching process for the purpose of discovering an optimized point. This searching process is carried out in two general levels. First, the user selects preference criteria, including traffic congestion, then the system finds a solution, which includes the lowest traffic jam via employed bees. Second, onlooker bees find the nearest ideal location to the user’s preferences. 8. Recommendation to tourists: Finally, the recommender system finds three places based on the results obtained from level 7 and recommend these tourist destinations to the user. 6. Analysis and Findings In this paper, five tourist cities like Tehran, Isfahan, Shiraz, Mashhad, and Orumieh were selected for conducting a pilot study. Hence, these cities with some tourist attractions were recommended for new arrivals. Each of these cities converted to a matrix with specific val- ues on the basis of the rating system (Table 1) via the TOPSIS model. Table 2 demonstrates a matrix for each city: It is clear that every city has its specific matrix and each row of the matrix represents one tourist’s spot in this city, and all fields are rated according to Table 1. The user selects the area he/she needs to travel; then, the system suggests the best place to visit according to the importance of each factor. Secondly, the ABC algorithm is carried out, looking by employed bees for the best ideal matrix, then onlooker bees searching for other nearest ideal two matrices, proposing all to the user. The outcome of this algorithm for cities is described in Table 3 as follows: As shown in Table 3, the ABC algorithm was implemented for five cities with the outcome of different values for each city’s factors. Every city has three matrices that show three tourist destinations. At the top of the matrix, the cost of each matrix is specified, and the best matrix is defined in terms of the cost outlined in the figure. For example, with a cost of 37, the best matrix for Tehran is the second one, and it is the closest ideal matrix to the best positive ideal matrix. At this point, with 1000 repeats, the employed bee starts looking in the loop to find a first matrix based on the given value. Then, onlooker bees start searching and finding the other two matrices. Consequently, three matrices are found and recommend to the user. FUZZY INFORMATION AND ENGINEERING 39 The evaluation of a proposed model contains three sections. In the first section, the authors assumed and compared the accuracy of recommending items to a tourist based on the questionnaires. In the second section, the ABC algorithm’s accuracy about finding an optimized matrix is assessed by a specific function. In the third section, the proposed method in the present research has been implemented on the Tripadvisor dataset and compared with other algorithms. 6.1. Accuracy of the Recommender Systems For evaluating the TOPSIS model’s performance, one city is selected as an example, using an online questionnaire to assess the accuracy of this method. The outcome recommendation items are compared with users’ suggestions. If the similarity of these factors is high, it is concluded that the accuracy of our recommender system is acceptable. In order to calculate a difference between these two criteria, the Euclidean distance formula [58] was utilized as follows: d(x, y) = (X − Y ) (15) k k k=1 Table 2. Presenting tourism destinations in the form of a matrix using the TOSIS model. Tehran Esfahan Shiraz Mashhad Orumieh 110 3 710 135310 77510 53710 3 7 710 3 73710 10 3510 7 77310 5 10 10 7 7710 10 10 10 5 10 510 77710 553710 1 7510 37510 7 510 3710 5553155 10 10 75110 7 57575337510 10 1 3 10 10 510 7733510 10 5 17710 5 510 1 710 10 10 10 7773313710 57710 10 1 710 10 3335810 10 3337710 1 355 15710 75710 355710 10 10 10 10 5 57 710 10 711510 10 1355713310 10 3 55710 510 5 3577510 533110 7 10 77510 537110 55773310 3 5710 10 10 3710 17710 10 775710 3710 10 1 55710 357710 11357710 3 315 10 5317510 3 710 3315710 3 10 10 7 710 5 7573310 10 13735710 10 10 7 3 103 5 107 10 3 1 7 10107 5 10105 3 7 5 77710 73355137755710 3 510 73710 3510 1 7710 5 310 10 5310 7 5 5710 373510 3 10 355710 31715 7710 10 10 71310 310 1 17577710 7 ... ... ... ... ... Table 3. Performing the ABC algorithm and TOPSIS technique. Orumieh Mashhad Shiraz Esfahan Tehran 1 Cost 27 Best Ans Cost 25 Best Ans Cost 26 Best Ans Cost 22 Best Ans Cost 34 Best Ans 5 5 710 10 5 371 10 105 5 5 5710 710 7 2 Cost 34 Best Ans Cost 30 Best Ans Cost 32 Best Ans Cost 37 Best Ans Cost 31 Best Ans 7 7 10 10 10 7 3 10 5 7 10 10 10 10 7 10 10 7 7 7 3 Cost 37 Best Ans Cost 32 Best Ans Cost 30 Best Ans Cost 29 Best Ans Cost 29 Best Ans 10 10 10 7 10 7 5 10 3 10 7 10 10 7 7 5 7 10 7 5 40 S. FOROUZANDEH ET AL. Figure 4. Comparison of user choices and recommendation system results. Where n represents the number of characteristics, X and Y denote Euclidean vectors start- k k ing from the origin of the space, and their tips show two points. For example, ‘Esfahan’ is selected to evaluate in terms of the accuracy of the TOPSIS model. Esfahan consists of 20 tourist destinations that each destination formed into a matrix through the TOPSIS model. In this regard, a questionnaire was distributed on Facebook to ask respondents to select tourist spots based on four criteria: distance, road condition, traffic congestion, and tourist spot attractions. The numbers of 526 respondents were getting involved in this survey. They selected specific destinations according to their previous experiences of traveling to these tourist destinations. Afterward, the outcomes of online questionnaires were compared with the results of our recommender system. Results are presented in Figure 4 as follows: According to Figure 2, the blue color line indicates that user’s responses, including the places they selected, and the red color line, presents recommendation system out- comes. As can be seen, in most cases, the user’s answers and system recommendations are matched together, and it is concluded that the accuracy of our recommender system is high regarding offering places for tourists. The next comparison is related to quality. As mentioned in the previous section, a matrix with a maximum value of 10 points was formed for each column. Hence, in the best situ- ation, the sum of all columns in the ideal matrix would be 40 scores. Thus, an ideal option would be an option with the nearest distance to 40 scores in the ideal matrix. For exploring a similarity between recommender system results and users opinions, the following formula canbeused[59–62]: R = sim(C, C ) = argmax sim(C, C ) (16) c,c c,c ∈R By applying this method, the similarity between the user’s judgment and recommendation system suggestions, which is represented by C and C were examined by items such as sim (c,c ). Here, argmax picks a higher number between C and C , which means there is a com- parison between the user’s opinion and recommender system outcome for determining FUZZY INFORMATION AND ENGINEERING 41 Figure 5. Comparison of user choices and recommender system results. the nearest distance to the ideal purpose. Consequently, the outcome, which is near to 40, is selected as an ideal option. Figure 5 illustrates the results from this level. According to Figure 3, the blue color line presents users to review, and the red color line illustrates recommendation systems results. Besides, in some parts of the figure, there is a gap between the user’s suggestion and recommender system suggestions; the authors analyzed these differences based on their score and found that all suggestions by the rec- ommender system are outscored the user’s recommendation. Hence, in terms of scoring criteria, this is acceptable that the places were recommended by recommendation systems that have a near distance to the ideal matrix. 6.2. Accuracy of the ABC Algorithm In this section, the ABC algorithm’s accuracy to find the best places to recommend to users is evaluated by the function, which produces random numbers in a specific domain. This func- tion generates random numbers available in a matrix-domain, then compares these points with destinations offered by the recommender system. The random number generation function is defined as follows [42]: L = Rand(1, 10), i ∈{1, 3, 5, 7, 10} (17) i=1 This formula includes four digits ranging from 1 to 10 (TOPSIS model domain). In each cycle of performing this function, one random number has been generated, then all generated numbers sum together and calculate the cost function. The results of comparing the ABC algorithm and Random number generation function is presented in Figure 6 as follows: Both algorithms were applied five times for five cities, and the best results outcome was selected in each cycle of performing algorithms. Hence, 25 destinations were explored. The precision of the ABC algorithm was higher than the Random number generation algorithm because the outcomes of ABC have a minor distance with an ideal solution. The results of 42 S. FOROUZANDEH ET AL. Figure 6. Evaluating the accuracy of the ABC algorithm and Random number generation function. Table 4. Converting qualitative criteria to quantitative criteria on the Tripadvisor site Qualitative criteria Terrible Poor Average Very good Excellent Quantitative criteria 1 2 3 4 5 the ABC algorithm recommend appropriate options to users due to conclusive and highly valued outcomes. 6.3. Experimental Evaluation The method presented in this analysis was applied on the Tripadvisor dataset that contains the recommended hotel to the user and involves a review of 235,793 hotels collected by Crawler in one month [63–65]. The ratings provided by users are between 1 and 5, and according to the type of dataset studied by TripAdvisor, the criteria and rating associated with TOPSIS have changed in the research, and the selection of criteria for scoring is based on qualitative criteria on the Tripadvisor site and is considered as Table 4 as follows: Also, in this dataset to select each hotel has been intended for some criteria, and the users’ rating for hotels is according to Table 4 and these criteria. Hence, each hotel possesses an identification ID, and the criteria for hotels are as follows: 1. Business_Service 2. Service 3. Check in/front desk 4. Cleanliness 5. Location 6. Rooms 7. Value FUZZY INFORMATION AND ENGINEERING 43 On the basis of these seven criteria, the algorithm is then applied and tries to identify hotels that are based on each of these criteria and have an appropriate ranking in each of them to recommend to users. The accuracy of the algorithm provided in this study is also compared with other recommended system algorithms, such as Content-Based (CB) Systems, Collaborative Filtering (CF) Standard, and Collaborative (CL) Systems Algorithms. [66], that contains several algorithms (called Tourism Recommender Systems related algo- rithms). According to the following criteria, the accuracy of the recommender system in the present article was investigated by CF standard, CL, and CB algorithms on the Tripadvisor dataset. Ultimately, the authors evaluate two approaches and used precision, recall, and F1 mea- surements as recommended by Herlocker et al. to illustrate the efficacy of the proposed approach in recommending with the ABC & TOPSIS method and CF Standard and CL and CB for recommendation hotels (2004). The authors have therefore separated the ratings of each user on ABC & TOPSIS hotels into two main sets, the training set and the test set as used by Billsus and Pazzani (1998). On the training set, the model was trained, and then the test set was used to predict the Top-N hotels. In addition, the hotels were divided into two groups, irrelevant and relevant, to measure the accuracy and recall. The precision met- ric provides the probability that the hotel selected is relevant (Equation 18), and the recall metric provides the probability that the hotel selected would be selected (Equation 19). It is important to evaluate these two metrics together. The F1 measure that combines these two metrics can also be used as a single metric to evaluate the quality of recommendations, as suggested by Sarwar et al. [67]. Hotel relevant−selected Precision = (18) Hotel + Hotel relevant−selected relevant−selected Hotel relevant−selected Recall = (19) Total relevant(selected−notselected) 2 × Precision × Recall F1 = (20) Precision + Recall In order to assess the accuracy of the recommender system for the provision of hotels to users on the basis of the above criteria, a cost matrix has been established, and the mini- mum number of points associated with the created user matrix has been specified. For this purpose, it should be regarded that if all fields get an outstanding ranking, the cost of the matrix is equivalent to (7 ∗ 5 = 35), and this is the highest rating that can be earned by each matrix. The next high-quality criterion in the second of the maximum score is Very Good, with a score of 4, and the matrix cost is equal to (7 ∗ 4 = 28) if all the metrics have a Very Good rating. The recommendations presented in this research are therefore based on two appropriate parameters for users, and on this basis, there are appropriate matrices that achieve the lowest score (28), and the score achieved in the matrix must be at least equal to 4 for both of the criteria. Correspondingly, the following rules in Table 5 give an acceptable matrix: 6.4. Rules for Relevant Hotel As shown in Table 5, for providing recommendations hotels to users, there are accept- able matrices possessing all the features of that hotel or rated (‘Very good’ or ‘Excellent’) 44 S. FOROUZANDEH ET AL. Table 5. Rules for created acceptable matrixes in evaluating the accuracy of the recommender system in Tripadvisor 01 IF [ ‘Value’ = (‘ Very good’ or ‘Excellent’) OR ‘Value’ > = 4 AND Rating > = 28 ] 02 AND [‘Room’ = (‘ Very good’ or ‘Excellent’) OR ‘Room’ > = 4 AND Rating > = 28] 03 AND [‘Location’ = (‘ Very good’ or ‘Excellent’) OR ‘Location’ > = 4 AND Rating > = 28] 04 AND [‘Cleanliness’ = (‘ Very good’ or ‘Excellent’) OR ‘Cleanliness’ > = 4 AND Rating > = 28] 05 AND [‘Check in/front desk’ = (‘ Very good’ OR ‘Excellent’) OR ‘Check in/front desk’ > = 4 ] 06 AND [‘Service’ = (‘ Very good’ or ‘Excellent’) OR ‘Service’ > = 4 AND Rating > = 28] 07 AND [‘Business_Service’ = (‘ Very good’ OR ‘Excellent’) OR ‘Business_Service’ > = 4 AND Rating > = 28] 08 Then Generate Matrix: = Relevant Hotel; Figure 7. Accuracy of matrices produced in Precision and on the basis of user priorities for hotel-user recommendations on TripAdvisor. or at least possess 4 points (every user chooses a feature as a search priority among hotels). Therefore, seven times the algorithm is executed according to the seven prior- ities in selecting each hotel, and the archived matrices are determined by applying the desired approaches, and then the accuracy of each of them is achieved in different runs. Accordingly, the findings are presented in Figures 7–9. It is obvious that the diagram created in the vertical column includes Generate Matrix, and in the horizontal column involves Precision, Recall, and F1 and each of which was according to user priorities and the created rule (Every Priority > = Very Good or points 4). The number of matrix production has been made for every single diagram in ranges of 100, 200, 400, 600, 800, and 1000, which means that in the first stage, 100 matrices and the second stage 200 and finally 1000 matrices produced and specified the number of gener- ated matrices (based on percentage) with a bet (Every Priority > = Very Good or 4) and the precision of the matrices produced to provide the hotel to the users is measured and illustrated according to the users’ priorities. The important point regarding the increase of the production of matrices was that with the increase in their product ranges, the accu- racy of the matrices also increased, which shows the usefulness of the algorithm in offering recommendations to users in a large number. FUZZY INFORMATION AND ENGINEERING 45 Figure 8. Accuracy of matrices produced in Recall and on the basis of user priorities for hotel-user recommendations on TripAdvisor. Figure 9. Accuracy of matrices produced in F1 and on the basis of user priorities for hotel-user recom- mendations on TripAdvisor. 7. Conclusion Choosing a tourist destination to travel to is one of the hectic activities due to the numer- ous options travelers usually come across. There are various sorts of factors that should be considered to travel, ranging from activities to transportations, etc. Each traveler has special favors and preferences and usually faces many options. Thus, the uniform trip plan is not suitable for all kinds of tourists. A productive and customized system is essential to get some parameters via the user and offer an excellent choice for tourists to provide them a memo- rable trip. The present study provides a recommendation system to suggest an appropriate tourist destination for each type of visitors based on their preferences. This system works based on TOPSIS fuzzy model and the ABC algorithm. First of all, the TOPSIS model defines 46 S. FOROUZANDEH ET AL. a positive ideal solution in the form of a matrix with four columns, which indicates a factor that includes in this study. By distributing online questionnaires on the Facebook website, the respondent was asked to put a value of each factor out of 10 scores. Therefore, the desired indicators and criteria are evaluated using the TOPSIS model and decision matrix, and the ideal points are detected and by converting qualitative values into quantitative to solve the problem, then the distance of other points or the same tourist places in each city to the ideal points is calculated. After creating a matrix for every tourist’s destination, the ABC algorithm started to search amongst destinations and recommend the best tourist spot to users. Conclusive results of the present investigation indicate a high performance of these algorithms. The proposed recommendation system would help travelers select the best tourist attraction in the cities based on their preferences. 8. Limitation and Future Work In the present article, due to extensive tourist zones in Iran, the authors only selected the five most touristic cities; the number of cities can be increased in future research work since there are plenty of tourist cities and urban heritages in Iran that can be taken into account in future work. Another limitation of this research was about distributing the questionnaires. This task was done through online questionnaires within the Facebook website via several accounts for gathering data from users. Future works can be expanded through the sur- vey method by an in-depth interview with travelers and make face-to-face questionnaires. Besides, future studies can focus on the relationship between touristic cities and recom- mend other tourist attractions between these cities; when travelers want to travel between two cities, the recommender system can suggest some tourist destinations around these cities based on user location and routing system. Actually, through making location graphs and neighborhood matrix, and the related algorithms, more tourist hot spots can be recom- mended to travelers, particularly in far distances. Hence, future works can boost our novel recommender system to enhance the results. Disclosure statement No potential conflict of interest was reported by the author(s). Notes on contributors Saman Forouzandeh is currently a PhD student with the Faculty of Science, University of New South Wales (UNSW), Sydney, Australia. He received the M. Sc. degree in Software Engineering from Islamic Azad University, Science and Research Branch, Tehran, Iran in 2013 and B. Sc. degree of Software Engineering from University of Orumiyeh, Iran in 2009. His main research interests include Machine Learning, Graph Neural network, and Recommender Systems. s.forouzandeh@unsw.edu.au Mehrdad Rostami received the B. Sc. degree with honors in Computer Engineering from Razi Uni- versity, Kermanshah, Iran, in 2012, and the M. Sc. degree with honors in Artificial Intelligence from University of Kurdistan, Sanandaj, Iran, in 2014. He is currently a Ph. D. student and researcher at Uni- versity of Oulu, Oulu, Finland. His research interests include pattern recognition, medical diagnosis, machine learning, big data mining, and bioinformatics. FUZZY INFORMATION AND ENGINEERING 47 Kamal Berahmand received the M.S. degree in information technology engineering from Azarbai- jan Shahid Madani University, Tabriz, Iran, in 2017. He is currently a Ph. D. student with the Faculty of Computer Engineering and Information Technology, Queensland University of Technology, Brisbane, Australia. His current research interests include Machine learning, network science, graph neural network, and knowledge graph. ORCID Saman Forouzandeh http://orcid.org/0000-0002-5952-156X References [1] Tugaonkar PS, Chitre V. Survey on recent methodologies used for recommender system. 2016 International Conference on Electrical, Electronics, and Optimization Techniques (ICEEOT). 2016. IEEE. [2] Lu J, Wu D, Mao M, et al. Recommender system application developments: a survey. Decis Support Syst. 2015;74:12–32. [3] Esmaeili L, Mardani S, Golpayegani SAH, et al. A novel tourism recommender system in the context of social commerce. Expert Syst Appl. 2020;149:113301. [4] Pujahari A, Sisodia DS. Model-Based Collaborative Filtering for Recommender Systems: An Empirical Survey. 2020 First International Conference on Power, Control and Computing Tech- nologies (ICPC2T). 2020. IEEE. [5] Alhijawi B, Kilani Y. The recommender system: A survey. Int J Adv Intell Paradigms. 2020;15(3):229–251. [6] Ashley-Dejo E, Ngwira S, Zuva T. A survey of context-aware recommender system and services. 2015 International Conference on Computing, Communication and Security (ICCCS). 2015. IEEE. [7] Osadchiy T, Poliakov I, Olivier P, et al. Recommender system based on pairwise association rules. Expert Syst Appl. 2019;115:535–542. [8] Forouzandeh S, Aghdam AR. Health recommender system in social networks: A case of Facebook. Webology. 2019;16(1):40–54. [9] Forouzandeh S, Aghdam AR, Forouzandeh S, et al. Addressing the cold-start problem using data mining techniques and improving recommender systems by cuckoo algorithm: A case study of Facebook. Comput Sci Eng. 2020;22(4):62–73. [10] Braunhofer M, Ricci F. Contextual information elicitation in travel recommender systems. In: Inversini A, Schegg R, editors. Information and Communication Technologies in Tourism 2016. Cham: Springer; 2016. p. 579–592. DOI:10.1007/978-3-319-28231-2_42 [11] Renjith S, Sreekumar A, Jathavedan M. An extensive study on the evolution of context-aware personalized travel recommender systems. Inf Process Manag. 2020;57(1):102078. [12] Kenteris M, Gavalas D, Mpitziopoulos A. A mobile tourism recommender system. The IEEE symposium on Computers and communications. 2010. IEEE. [13] Logesh R, Subramaniyaswamy V. Exploring hybrid recommender systems for personalized travel applications. Cognitive informatics and soft computing. 2019, Springer. p. 535-544. [14] Gavalas D, Kasapakis V, Konstantopoulos C, et al. A survey on mobile tourism recommender systems. 2013 Third International Conference on Communications and Information Technology (ICCIT). 2013. IEEE. [15] Ricci F. Recommender systems in tourism. Handbook of e-Tourism, 2020: p. 1-18. [16] Gavalas D, Konstantopoulos C, Mastakas K, et al. Mobile recommender systems in tourism. J Network Comput Appl. 2014;39:319–333. [17] Zeng B, Gerritsen R. What do we know about social media in tourism? A review. Tourism Manage Perspect. 2014;10:27–36. [18] Huang J-H, Peng K-H. Fuzzy rasch model in TOPSIS: A new approach for generating fuzzy num- bers to assess the competitiveness of the tourism industries in asian countries. Tourism Manage. 2012;33(2):456–465. 48 S. FOROUZANDEH ET AL. [19] White WG, Chandrasekar V. TOPSIS to optimize performance, reliability, and life cycle costs dur- ing analysis of alternatives. 2017 Annual Reliability and Maintainability Symposium (RAMS). 2017. IEEE. [20] Forouzandeh S, Berahmand K, Rostami M. Presentation a trust walker for rating prediction in recommender system with biased random walk: effects of H-index centrality, similarity in items and friends. Eng Appl Artif Intell. 2021;104:104325. [21] Forouzandeh S, Berahmand K, Rostami M. Presentation of a recommender system with ensemble learning and graph embedding: a case on MovieLens. Multimed Tools Appl. 2020;30:1–28. [22] Forouzandeh S, Sheikhahmadi A, Rezaei Aghdam A, et al. New centrality measure for nodes based on user social status and behavior on Facebook. Int J Web Inf Sys. 2018;14(2):158– [23] Rostami M, Berahmand K, Forouzandeh S. A novel community detection based genetic algorithm for feature selection. J Big Data. 2021;8(1):1–27. [24] Rostami M, Berahmand K, Forouzandeh S. A novel method of constrained feature selection by the measurement of pairwise constraints uncertainty. J Big Data. 2020;7(1):1–21. [25] Berahmand K, Nasiri E, Forouzandeh S, et al. A preference random walk algorithm for link pre- diction through mutual influence nodes in complex networks. J King Saud Univ Comput Inf Sci. 2021. DOI:10.1016/j.jksuci.2021.05.006 [26] Forouzandeh S, Aghdam AR, Barkhordari M, et al. Recommender system for users of internet of things (IOT). IJCSNS. 2017;17(8):46. [27] Forouzandeh S, Soltanpanah H, Sheikhahmadi A. Application of data mining in designing a recommender system on social networks. Int J Comput Appl. 2015;124(1):32–38. [28] Rafsanjani AHN, Salim N, Aghdam AR, et al. Recommendation systems: a review. Int J Comput Eng Res. 2013;3(5):47–52. [29] Munuswamy S, Saranya MS, Ganapathy S, et al. Sentiment analysis techniques for social media- based recommendation systems. Natl Acad Sci Lett. 2021;44:281–287. [30] Kulkarni PV, Rai S, Kale R. Recommender System in eLearning: A Survey. Proceeding of Interna- tional Conference on Computational Science and Applications. 2020. Springer. [31] Wang X, Li XR, Zhen F, et al. How smart is your tourist attraction? Measuring Tourist Pref- erences of Smart Tourism Attractions via a FCEM-AHP and IPA Approach. Tourism Manage. 2016;54:309–320. [32] Han D-I, Jung T, Gibson A. Dublin AR: Implementing Augmented Reality in Tourism, in Informa- tion and Communication Technologies in Tourism 2014. 2013, Springer. p. 511-523. [33] Garcia I, Sebastia L, Onaindia E. On the design of individual and group recommender systems for tourism. Expert Syst Appl. 2011;38(6):7683–7692. [34] Nilashi M, bin Ibrahim O, Ithnin N, et al. A multi-criteria collaborative filtering recommender system for the tourism domain using expectation maximization (EM) and PCA–ANFIS. Electron Commer Res Appl. 2015;14(6):542–562. [35] Bucur C. Using opinion mining techniques in tourism. Procedia Econ Finance. 2015;23:1666–1673. [36] Braunhofer M, Ricci F, Lamche B, et al. A context-aware model for proactive recommender systems in the tourism domain. Proceedings of the 17th International Conference on Human- Computer Interaction with Mobile Devices and Services Adjunct. 2015. [37] Jaiswal A, Mishra R. Cloud service selection using topsis and fuzzy topsis with ahp and anp. Proceedings of the 2017 International Conference on Machine Learning and Soft Computing. [38] Alrababah SAA, Gan KH, Tan T-P. Comparative analysis of MCDM methods for Product Aspect Ranking: TOPSIS and VIKOR. 2017 8th International Conference on Information and Communi- cation Systems (ICICS). 2017. IEEE. [39] Liu L, Xu J, Liao SS, et al. A real-time personalized route recommendation system for self-drive tourists based on vehicle to vehicle communication. Expert Syst Appl. 2014;41(7):3409–3417. [40] Benitez JM, Martín JC, Román C. Using fuzzy number for measuring quality of service in the hotel industry. Tourism Manage. 2007;28(2):544–555. [41] Chou T-Y, Hsu C-L, Chen M-C. A fuzzy multi-criteria decision model for international tourist hotels location selection. Int J Hosp Manag. 2008;27(2):293–301. FUZZY INFORMATION AND ENGINEERING 49 [42] Ju C, Xu C. A new collaborative recommendation approach based on users clustering using artificial bee colony algorithm. Sci World J. 2013;2013:1–9. [43] Hsu C-C, Chen H-C, Huang K-K, et al. A personalized auxiliary material recommendation system based on learning style on Facebook applying an artificial bee colony algorithm. Comput Math Appl. 2012;64(5):1506–1513. [44] Roy A, Tavana M, Banerjee S, et al. A secured context-aware tourism recommender system using artificial bee colony and simulated annealing. Int J Appl Manage Sci. 2016;8(2):93–113. [45] Hamdan S, Cheaitou A. Green supplier selection and order allocation using an integrated fuzzy TOPSIS, AHP and IP approach. 2015 International Conference on Industrial Engineering and Operations Management (IEOM). 2015. IEEE. [46] Solanki R, Gulati G, Tiwari A, et al. A correlation based Intuitionistic fuzzy TOPSIS method on sup- plier selection problem. 2016 IEEE International Conference on Fuzzy Systems (FUZZ-IEEE). 2016. IEEE. [47] Tzeng G-H, Huang J-J. Multiple attribute decision making: methods and applications. New York: CRC press; 2011. [48] Behzadian M, Khanmohammadi Otaghsara S, Yazdani M, et al. A state-of the-art survey of TOPSIS applications. Expert Syst Appl. 2012;39(17):13051–13069. [49] Lu X, Chang W, Zhou S. Quality evaluation of diesel engine using an improved TOPSIS method based on K-means algorithm. 2016 35th Chinese Control Conference (CCC). 2016. IEEE. [50] Wei L, Qin T, Yan Y, et al. Competitive ability evaluation for generation enterprises based on com- binational weight and improved TOPSIS. 2016 Chinese Control and Decision Conference (CCDC). 2016. IEEE. [51] Chen X, Tianfield H, Li K. Self-adaptive differential artificial bee colony algorithm for global optimization problems. Swarm Evol Comput. 2019;45:70–91. [52] Hannam K, Butler G, Paris CM. Developments and key issues in tourism mobilities. Ann Tourism Res. 2014;44:171–185. [53] Lerato M, Esan OA, Ebunoluwa AD, et al. A survey of recommender system feedback tech- niques, comparison and evaluation metrics. 2015 International Conference on Computing, Communication and Security (ICCCS). [54] Bouchibane F, Bensebti M. Parameter tuning of Artificial Bee Colony algorithm for energy effi- ciency optimization in massive MIMO systems. 2017 Seminar on Detection Systems Architectures and Technologies (DAT). 2017. IEEE. [55] Bansal JC, Gopal A, Nagar AK. Stability analysis of artificial bee colony optimization algorithm. Swarm Evol Comput. 2018;41:9–19. [56] Zhou J, Yao X, Chan FTS, et al. An individual dependent multi-colony artificial bee colony algorithm. Inf Sci (Ny). 2019;485:114–140. [57] Gao W-f, Huang L-l, Wang J, et al. Enhanced artificial bee colony algorithm through differential evolution. Appl Soft Comput. 2016;48:137–150. [58] Ricci F, Rokach L, Shapira B. Introduction to recommender systems handbook. In: Recommender systems handbook. Boston (MA): Springer; 2011.p.1–35. [59] Nagpal D, Kaur S, Gujral S, et al. Fr: A recommender for finding faculty based on CF technique. Procedia Comput Sci. 2015;70:499–507. [60] Karimi M, Jannach D, Jugovac M. News recommender systems – survey and roads ahead. Inf Process Manag. 2018;54(6):1203–1227. [61] Bentaleb A, El Bouzekri Y, Lahcen AA, et al. Context Aware Recommender Systems for Tourism: A Concise Review. 2018 IEEE 5th International Congress on Information Science and Technology (CiSt). 2018. IEEE. [62] Garcia LM, Aciar S, Mendoza R, et al. Smart tourism platform based on microservice architecture and recommender services. International Conference on Mobile Web and Intelligent Information Systems. 2018. Springer. [63] Marcheggiani D, Täckström O, Esuli A, et al. Hierarchical multi-label conditional random fields for aspect-oriented opinion mining. European conference on information retrieval. 2014. Springer. 50 S. FOROUZANDEH ET AL. [64] Wang H, Lu Y, Zhai C. Latent aspect rating analysis on review text data: a rating regression approach. Proceedings of the 16th ACM SIGKDD international conference on Knowledge dis- covery and data mining. 2010. [65] Nilashi M, Ibrahim O, Yadegaridehkordi E, et al. Travelers decision making using online review in social network sites: A case on TripAdvisor. J Comput Sci. 2018;28:168–179. [66] Borràs J, Moreno A, Valls A. Intelligent tourism recommender systems: A survey. Expert Syst Appl. 2014;41(16):7370–7389. [67] Nilashi M, Ibrahim O, Yadegaridehkordi E, et al. Preference learning for eco-friendly hotels recom- mendation: A multi-criteria collaborative filtering approach. J Cleaner Prod. 2019;215:767–783. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Fuzzy Information and Engineering Taylor & Francis

A Hybrid Method for Recommendation Systems based on Tourism with an Evolutionary Algorithm and Topsis Model

Download PDF
25 pages

Loading next page...
 
/lp/taylor-francis/a-hybrid-method-for-recommendation-systems-based-on-tourism-with-an-IVcvbZJ5dK

References (69)

Publisher
Taylor & Francis
Copyright
© 2022 The Authors. Published by Taylor & Francis Group on behalf of the Fuzzy Information and Engineering Branch of the Operations Research Society, Guangdong Province Operations Research Society of China.
ISSN
1616-8666
eISSN
1616-8658
DOI
10.1080/16168658.2021.2019430
Publisher site
See Article on Publisher Site

Abstract

FUZZY INFORMATION AND ENGINEERING 2022, VOL. 14, NO. 1, 26–50 https://doi.org/10.1080/16168658.2021.2019430 A Hybrid Method for Recommendation Systems based on Tourism with an Evolutionary Algorithm and Topsis Model a b c Saman Forouzandeh , Mehrdad Rostami and Kamal Berahmand School of Mathematics and Statistics, Faculty of Science, University of New South Wales, Sydney, NSW, Australia; Centre of Machine Vision and Signal Processing, Faculty of Information Technology, University of Oulu, Oulu, Finland; School of Computer Sciences, Science and Engineering Faculty, Queensland University of Technology (QUT), Brisbane, Australia ABSTRACT ARTICLE HISTORY Received 16 September 2020 Recommender systems have been pervasively applied as a technique Revised 11 July 2021 of suggesting travel recommendations to tourists. Actually, recom- Accepted 14 December 2021 mendation systems significantly contribute to the decision-making process of tourists. A new approach of recommendation systems in KEYWORDS the tourism industry by a combination of the Artificial Bee Colony Recommender systems; ABC (ABC) algorithm and Fuzzy TOPSIS is proposed in the present paper. algorithm; topsis model; A multi-criteria decision-making method called the Techniques for tourism; Facebook social network Order of Preference by Similarity to Ideal Solution (TOPSIS) has been applied for the purpose of optimizing the system. Data were gath- ered through a 1015 online questionnaire on the Facebook social media site. In the first stage, the TOPSIS model defines a positive ideal solution in the form of a matrix with four columns, which indicates factors that get involved in this study. In the second stage, the ABC algorithm starts to search amongst destinations and recommends the best tourist spot to users. 1. Introduction One of the inevitable aspects of human life is decision making. Faced with many options, lack of proper expertise, and time constraints resulted in hectic human judgments. Recom- mendation systems are therefore provided to resolve these issues [1–3]. The knowledge discovery methods and statistical techniques are applied by recommendation systems for suggesting suitable items to clients [4–6]. These systems operate on the basis of the pref- erences and behavior of the user to recommend an option that is strongly adapted to the preferences of the client to make a proper decision [7–9]. In several different domains, rec- ommendation systems are utilized and recently implemented in the tourism industry to suggest a suitable destination for visitors [10,11]. The objective of travel recommendation systems is to match consumer requirements with the characteristics of tourism and leisure resources or attractions [2]. For recommending the best place, this system needs some information about tourist attractions and destinations, tourists visiting schedules, rating score by other users for each destination, priorities, and preferences of travelers, the length CONTACT Saman Forouzandeh s.forouzandeh@unsw.edu.au © 2022 The Authors. Published by Taylor & Francis Group on behalf of the Fuzzy Information and Engineering Branch of the Operations Research Society, Guangdong Province Operations Research Society of China. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons. org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. FUZZY INFORMATION AND ENGINEERING 27 of travel days, starting and the ending time of visiting tourist attractions etc. Through this type of information, the itinerary is offered to customers. Therefore, in suggesting the best travel to users, different criteria should be taken into account. The TOPSIS model can cover various criteria in providing recommendations to users, and it is used in this research to solve the problem. In order to cover users’ multi-criteria decisions, the authors use the TOPSIS model, and their preferred criteria are allocated impact factors depending on their preferences. Afterward, according to client pri- orities, the ABC algorithm is applied to search for qualified tourist places. The objective of applying the ABC algorithm is to discover different places in the selected range of the user, following their preferences with high ratings according to their criteria. Applying the ABC algorithm does the search for tourist places for users in three stages, and also the search for places is based on defined criteria in the TOPSIS model. In essence, the recommendation system offers personalized information for users regarding the trip plan [12,13]. Also, a couple of factors play an important role in the itinerary, including time managing and selecting the best option based on tourist prefer- ences [14,15]. There are plenty of tourist destinations, activities, and transportations in every tourist spot, and there are various types of travel plans as well. Since travelers have different preferences and flavors, the uniform travel plan would not be proper for all tourists. There- fore, a flexible and customized system is required to input some parameters by the user, then an excellent choice is offered to tourists to provide them a marvelous time during their vacation [16]. In this regard, the significant role of social media can be drawn con- cerning tourism development. For instance, social media users have a chance to check the travel experiences of their friends [17]. Moreover, mobile phones significantly contribute to the recommendation for tourists through discovering a current location of users, rec- ommending a particular near destination through the use of recommendation techniques [14]. Due to the growing number of options for users to find a tourist destination, Fuzzy logic systems were produced in this field. All possible choices were categorized by a multi- criteria decision-making system, providing a situation for users to select an appropriate solution [18,19]. The objective of the present research is to apply the TOPSIS model and ABC algorithm for optimizing tourist destinations according to specific principles and finally recommend these options to visitors. The article is structured as follows. Section 2 provides a literature review on recommen- dation systems and methods in the tourism, Section 3 addresses the multi-criteria decision method, and Section 4 outlines the ABC algorithm. Section 5 discusses the methodology of the research, Section 6 includes analysis and results, and the conclusion is presented in Section 7. 2. Literature Review One of the purposes of recommendation systems is detecting the user preferences and suggesting items or services for meeting users’ requirements and favors [20–22]. These systems are being developed as an efficient research area and possess applications in var- ious fields, including medical science, business, education, e-commerce, etc. [23–28]. The primary objective of recommendation systems is proposing a considerable suggestion of goodsorservicestousers [28–30]. There are several types of research in this area that are carried out in the tourism industry. For instance, Wang and Lee studied a methodological 28 S. FOROUZANDEH ET AL. technique for assessing smart tourist attraction (STA) tourist preference; afterward, a fuzzy comprehensive calculation strategy and analytical hierarchy procedure were introduced to the Zoo’s STA assessment, which is one of China’s tourist attractions. After evaluating the data and conducting the fuzzy assessment, they distributed 680 questionnaires to col- lect information from zoo visitors, and findings derived primary STA evaluation factors, including intelligent tourism management, smart information systems, smart forecasting, etc. [31]. Some researchers have provided new technologies, including Augmented Reality (AR), to satisfy visitors with smart recommendations. They created an AR tourist application for smartphones. In order to evaluate the results, 26 in-depth interviews with 26 international and local tourism and thematic study were used. The results revealed that the knowledge generated by social networks containing feedback, attractions, and destinations was com- mon with travelers and should be taken into consideration in AR applications to encourage visitors to use this modern technology and recommend tourist attractions and urban her- itage to them [32]. Some considerations could play an essential role in tourist choices. For example, one research suggested a recommendation framework focused on the users’ flavor, the destinations they have visited previously, and also their demographic catego- rization. The authors presented a recommendation framework that could be provided to a single user or a group of users. The questionnaires were distributed among 60 people, asking about overall preferences, demographic information, and the places they visited. The findings revealed that the combination of three basic recommendation approaches: demographic, content-based, and general like filtering, was conclusive in recommending the alternative for a single or group of users [33]. A recommender system was developed to recommend hotels to consumers in another study performed by Nilashi and Ibrahim. In order to fulfill the needs of a consumer on data originating from a similar user, this system used recommendation techniques. Through dimensional reduction and estimation tech- niques, they suggested a novel hybrid approach. In order to estimate tourist preferences, the proposed model was focused on data mining techniques, EM algorithms, cluster- ing techniques, and ranking items in user profiles, recommending a suitable option for them [34]. Another research used opinion mining techniques to recommend a particular hotel based on information from clients. Opinion mining methods were applied to data by har- vesting and then classifying hotel reviews from the Internet. The collection of data consists of 3000 reviews posted on tripadvisor.com by users. The findings of this analysis showed that both visitors looking for accommodation and hotels are seeking to develop their ser- vices by utilizing these strategies [35]. In the research study conducted by Braunhofer and Lamche, a prototype of a recommender system for a mobile point of interest (POI) was presented. This system operated on the basis of text mining and user location and five fac- tors that might touch a user-perceived, including travel time accessible to the user, visit time, weather, and the POI visit history of the user. The findings of this analysis indicate that user’s preferences and various forms of heuristics were derived by the proposed model to estimate the total usefulness of the POI recommendation to the target user in the current contextual position [36]. Multi-criteria TOPSIS decision model and fuzzy logic were applied in recent years to evaluate the quality of travel websites from the consumer’s viewpoint [37]. The deci- sion matrix has been created in this method, and certain parameters are regarded in FUZZY INFORMATION AND ENGINEERING 29 terms of the purpose. In this technique, the ranking procedure is centered on the short- est distance to the best positive answer and the farthest distance from the negative answer [38]. A study indicates a TOPSIS-based combination of genetic and fuzzy for a real-time route recommendation system for tourists in tourist hot spots in this regard. This analysis carried out three phases: data collection, TOPSIS-based data phase regard- ing route optimization, and the generation of new routes on the genetic algorithm to choose the best route. The data was gathered through questionnaires provided to self- driven travelers. The results of this study showed that this personalized route recom- mendation system in real-time could enable the self-drive tourists to finish their trip more effectively, leading to reduce the congestion of traffic in the tourist hotspot. In addition, this proposed model saves average visiting time and satisfies visitors’ specific preferences [39]. Another analysis applied a fuzzy number and a TOPSIS model to quantify service quality and rank each hotel’s service performance, respectively. The authors distributed a survey to hotel visitors to gather data on the level of service at three Grand Canaria Island hotels. After implementing the fuzzy set theory on data, they concluded that fuzzy numbers act as an appropriate method to resolve the vagueness of views relevant to the subjective judgments of human beings. Also, detecting the customer perception and preferences about service quality in the hotel are two key points to boost marketing effort and improve the perfor- mance of the recommendation systems [40]. In hotel location selection for visitors, fuzzy logic is also applied. Fuzzy multi-criteria decision making (FMCDM) has been performed in research conducted by Clou to help travelers in selecting the appropriate hotel. Twenty- one criteria that play an important role in consumer satisfaction in hotel selection were derived by the authors. The results of this study indicate that FMCDM would enable tourists to choose a suitable hotel location [41]. The Artificial Bee Colony algorithm is another feature used in the recommendation system (ABC). The paper offered a modern recommendation system centered on a col- laborative filtering recommendation system on the basis of the ABC algorithm and the K-means clustering algorithm. It actually contributes to better clustering accuracy. The ABC algorithm is often used for the purpose of solving this clustering technique’s local optimal problem. The results revealed that this collaborative filtering system focused on the algorithm of tourist clustering showed better performance than several other approaches of recommendation [42]. A recommendation system based on the ABC algorithm on Facebook, which is the most prominent social network worldwide, was pro- posed by researchers in another study. The data gathered by questionnaires, the explo- ration of popular topics, and the difficulty of materials were valued. In the next level, a system optimized its performance using the ABC algorithm. The results showed that this approach enhanced the implementation time and made the recommendation sys- tem suggestions accurate in an effective way [43]. Furthermore, the ABC algorithm can boost the effectiveness of the recommendation system. Arup Roy, in a research study, developed a recommender system using the ABC algorithm and the Simulated Anneal- ing (SA) technique. Arup Roy applied the Irish TripAdvisor datasets, including 29,799 reviews of Irish hotels by 21,851 users. Results argued that it might be useful to use the ABC algorithm to identify an optimal or near-optimal choice for user recommendation advice [44]. 30 S. FOROUZANDEH ET AL. 3. Multi-Criteria Decision Method Multi-criteria decision analysis is a method in decision-making that estimates multi- incompatible criteria. There are two key aspects of this method; multi-criteria optimization and multi-criteria decision analysis. Although multi-criteria decision-making concentrates on multi-condition disciplines with few solutions and in an uncertain situation, multi- criteria optimization covers more than one goal [45]. It is necessary to compare all options comprehensively in the multi-criteria decision process and take into consideration all cri- teria. All responses and ideal solutions have been described in the numeric format in this approach [46]. The TOPSIS model was offered in 1981 [47], which is one of the best multi- criteria decision-making models. This model is on the basis of selecting the farthest negative ideal solution and the nearest positive ideal solution between the alternatives that is the best solution [41,48]. TOPSIS model includes the m∗n matrix in which ‘m’ implies the solu- tions and ‘n’ refers to criteria [41,48]. All qualitative measures should be converted into quantitative criteria for calculating mathematic techniques, as presented in Table 1. TOPSIS consists of several steps, which are expressed as follows: In the first step, a decision matrix is formed according to the desired indicators and cri- teria that the weight of each of these indicators may differ from each other. The decision matrix is m × n, in which m is equal to the desired indicators and criteria and is equal to m = 4 in the present research because four criteria are considered to evaluate tourist loca- tions. n is also equal to the number of tourist places in each city; the amount of which depends on the considered city. Matrix (1) is formed based on m×n matrix as follows (decision matrix): ⎡ ⎤ a a ... a 11 12 1n ⎢ ⎥ a a ... a 21 22 2n ⎢ ⎥ ⎢ ⎥ .. . . ⎢ ⎥ A = (1) ⎢ ⎥ ij ⎢ .. . . ⎥ ⎢ ⎥ ⎣ ⎦ .. . . a a ... a m1 m2 mn unscaled matrix is one of the most important steps in multi-criteria decision-making and is used to unscaling the decision matrix. In the present research, the Euclidean method is used to unscale the decision matrix. For normalization and creating the standard matrix, Equation (2) is defined as follows [49,50]: ij r =  (2) ij kj k=1 In Equation (3), positive and negative ideal values are specified so that the distance of other points from these positive and negative ideal points can be determined. Where ω is the Table 1. Converting qualitative criteria to quantitative criteria. Qualitative Criteria Too Low Low Medium Much Too Much Quantitative Criteria 1 3 5 7 10 FUZZY INFORMATION AND ENGINEERING 31 weight of the j-th criterion; w =1(3) i=1 The criteria and measures do not possess the same prioritized index, and the index crite- ria table solves this impediment. In this regard, a high indexed value has a high weight. The matrix, presented in the following, illustrates the value of the standard index for each criterion: ⎡ ⎤ w r w r ... w r 1 11 2 12 2 1n ⎢ ⎥ w r w r ... w r ⎢ 1 21 2 22 n 2n ⎥ ⎢ ⎥ .. . . ⎢ ⎥ v = ⎢ ⎥ (4) ij ⎢ .. . . ⎥ ⎢ ⎥ ⎣ .. . . ⎦ w r w r ... w r 1 m1 2 m2 n mn ∗ − ∗ Determining a value for S and S calculated based on (5) and (6) formula. S represents a i i i distance of solution from ideal positive criteria and S represents a distance from negative solution criteria. There are two methods for calculating the distance of other points from the positive and negative ideal points: The euclidean method and the block method. The Euclidean method is used in the present research, Equation (5) calculates the distance of option i from the positive ideal, and Equation (6) calculates the distance of option i from the negative ideal. ∗ ∗ S = (v − v ) (5) ij j=1 Figure 1. TOPSIS model implementation steps. 32 S. FOROUZANDEH ET AL. − − S = (v − v ) (6) ij i j j=1 Where, i and j belong to [1..m] and [1..n], respectively. The relative closeness of i-th alterna- tive S with respect to S∗ described as follows: i i ∗ i C = (7) − ∗ S − S i i It is evident that the rating of options was calculated by C ∗. Equation (7) is used to calculate the relative closeness of options to the ideal solution. Finally, the options are ranked based on their closeness to the ideal solution. Accordingly, the TOPSIS model consists of six steps, as shown in Figure 1: Algorithm 1 TOPSIS Method in the 6 steps as below: Algorithm 1: TOPSIS method Step 1: Calculate the normalized decision matrix (nij) as: ij N = , i = 1, ... , m; j = 1, ... , n. ij i−1 ij Step 2: Calculate the weighted normalized fuzzy decision matrix (r )as: ij x = w × n , i = 1, ... , m ; j = 1, ... , n. ij j ij Step 3: Obtain the positive (PIS) and negative (NIS) ideal solutions as: max(r |i = 1, ... , m), IFj ∈ B ij + + + PIS ={r , ... , r , ... , r }, where v = PIS = 1 j n j min(r |i = 1, ... , m), IFj ∈ C ij max(r |i = 1, ... , m), IFj ∈ B ij − − − − {r , ... , r , ... , r }, where v = 1 j n j min(r |i = 1, ... , m), IFj ∈ C ij Step 4: Calculate  the distances of each alternative i in relation to the ideal solutions as: + n + S = (r − r ) , i = 1, ... , m ij i j − − S = (r − r ) i = 1, ... , m ij i j j Step 5: Calculate the closeness coefficient of the alternatives (CCi) as: CC = i + − S +S i i Step 6: Sort the alternatives in descending order. The highest CC the value indicates the best performance in relation to the evaluation criteria. 4. ABC Algorithm The artificial Bee Colony (ABC) algorithm is an optimization method to simulate the foraging behavior of honey bees. Karaboga introduced this algorithm in 2005 [51–53]. Honey bees can spread far from their hive, searching for food. The colony of artificial bees is classified into three groups [54]: employed bees, onlooker bees, and scout bees. The search mecha- nism for food begins with the bees, which are employed. In their memory, the employed bees investigate food sources inside the neighborhood; in the meantime, they share the information from these sources of food to the onlooker bees. Onlookers choose a source of food from the food sources selected by them within the neighborhood. An employed bee from which the source has been discarded becomes a scout and begins spontaneously FUZZY INFORMATION AND ENGINEERING 33 to look for a new food source. The fulfillment of initializing a value in the ABC algorithm is done as the following [54–56]. 1. Employed bees: in this level, artificial bee X forms a new candidate solution V in the i i neighborhood of its current location. Selecting a location of new food source is carried out as follows: V = x + Ø ( x – x )j = 1, 2, ... ., D ; k = 1, 2, ... , SN (8) ij ij ij ij kj Here v = [v ,v ... ., v ] represents current bee’s location and x = [x ,x ... ., x ] indi- i i1 i2, in i i1 i2, in th cates a i location of bee’s. k(k#j) implies a random number index chosen from the set {1,2, ... ,n} and SN presents a number of artificial bees, and D represents the solutions. Ø ij refers to a random number within [−1, 1]. Choosing a random x applied as the following formula: x = L + rand(0, 1) × (U − L ) (9) ij j j j Here U and L are the upper bond and the lower bond of the X , respectively, and rand () j j i denotes a random number within [0,1]. Subsequent to selecting the location of a new food source, the optimization value should be measured by the following formula: f ≥ 0 1 + f fit = (10) 1 + abs(f ) f < 0 i i In this study, on the basis of increasing repetition of employed bees, their search range is increased, and each time they search for 3 points for a similar purpose improves the likelihood that they can reach the optimal points. Therefore, at each output, the most appropriate point is optimized compared to the other two points and closer to the aim of the user (Max Points). Figure 2. ABC algorithm implementation step. 34 S. FOROUZANDEH ET AL. 2. Onlooker bees: employer bees share the food source knowledge with onlooker bees via waggle dance after the search phase is completed by them. At this level, the informa- tion collected from the bees employed is estimated by each onlooker bee, selecting a food source with a likelihood correlated with food resources. The following formulas present this probability: fit p = (11) SN fit j=1 D = [v p , v p , ... , v p (12) i i1 i i2 i in i th Where fit implies the fitness value of the i solution in the swarm. It is clear that the th higher probability of i is obtained by the better solution i. After picking all food sources by onlooker bees, the location of the neighbors’ food source is measured by Equation (9) and this cycle will be continued until getting the required goal. In Equation (11), the enhance- ment of the function of the onlooker bees is relevant to the location of their movement, and their search for food commences from the location where the employer bees were in that position (the newest) last time (V = [v ,v , ... ., v ]). The onlooker bees are then around i i1 i2 in the locations where the employer bees seek and the neighborhood between reviewed points and this local search for improvement, and in this way, further points can be found. 3. Scout bees: In the ABC algorithm, some bees give up their solution method, becom- ing scout bees to search randomly in the domain if the target cannot be reached after repetitive default processes. Applying this type of method will increase search efficiency. S = S – (V and Di) (13) ij b b Equation (12) represents that the Sb discover the points that the employed bees and onlooker bees (Vij and Di) have not visited previously. Accordingly, the ABC algorithm performs the search in three steps according to Figure 2: The optimized ABC algorithm is as the Algorithm 2 [57]: 5. Methodology The following study tries to present a recommender system for recommending an attrac- tive tourist zone for travelers. For this purpose, the multi-criteria decision analysis TOPSIS model has been applied that works based on fuzzy logic as well as a tourist destination ABC algorithm has been utilized for searching. The method of this research consists of two main parts. In the first part, the TOPSIS model has been used to define and model the problem, and in the second part, the ABC algorithm has been applied to find tourist spots to suggest to the target user. The procedure is as follows: studying the number of five cities that have many tourist places, using the experi- ence of users in traveling to the desired tourist destinations through a questionnaire and four criteria are set to assess the quality of each of them and users are asked to rate each of the criteria in the form of a questionnaire, and qualitative criteria have been used to make the questionnaire simple for users. Subsequently, the proposed method is presented FUZZY INFORMATION AND ENGINEERING 35 Algorithm 2: Artificial bee colony algorithm 01. Randomly generate SN points in the search space to form an initial population. 02. Evaluate the objective function values of the population 03. cycle = 1, trial = 0, i = 1, 2, ... , SN 04. repeat {The employed bee phase} 05. for i = 1to SN do 06. Generate a candidate solution V by Equation 7 07. Evaluate f(V ) 08. If f(V ) < f(X ), set X = V , trial = trial + 1, otherwise trial = 0 i i i i i i i 09. end for 10. Calculate the probability values p by Equations 10, 11 {The onlookers phase} 11. t = 0, i = 1 12. while t < = SN do 13. if rand(0, 1) < p then 14. Generate a candidate solution V by Equation 7 15. Evaluate f(V ) 16. If f(V ) < f(X ),set X = V , trial = trial + 1, otherwise trial = 0 i i i i i i i 17. t = t + 1 18. end if 19. Set i = i + 1, if i = SN,set i = 1 {The scouts phase} 20. If max(trial ) > limit and X –(V and D )by then i i ij i 21. Replace X with a new randomly produced solution by Equations 8, 12 22. end if 23. Memorize the best solution found so far 24. cycle = cycle+1 25. until (cycle = Maximum Cycle Number) using the TOPSIS model. Since there are several tourist places to suggest to the user, sev- eral ideal points (places with the highest score) should be defined as the best options to offer to the target user, and finally, a search should be done (using the ABC algorithm) for places that have the shortest distance to these ideal locations. For this purpose, the TOPSIS model is used; the procedure is first to convert qualitative values  into quan- titative values and then create a decision matrix numerically and based on the desired criteria. The TOPSIS method is one of the Multiple Attribute Decision Making (MADM) meth- ods that rank options. In this method, two concepts of ‘ideal solution’ and ‘similarity to ideal solution’ are used. The ideal solution is the solution that is the best choice in every way that we trying to approach it. In order to measure the similarity of a design (or option) to an ideal (places with the highest score) and anti-ideal (places with the lowest score) solution, the distance of that points (places) from the ideal and anti-ideal solu- tion is measured. The points are then evaluated and ranked based on the distance them from the ideal solution and anti-ideal solution. Accordingly, in this research, tourist places ranked and in the ideal case, each feature has 10 scores and in the worst case, it has 1 score. Ideally, each place has a maximum score of 40 and we are looking for places that have the least distance with the number 40. The method is based on the concept that the found place should have the shortest distance from the positive-ideal solution and the longest distance from the negative ideal solution. TOPSIS defines an indicator called similarity with the ideal positive solution and distance from the ideal negative solu- tion. The ABC Algorithm then searches for places with maximum similarity to the ideal 36 S. FOROUZANDEH ET AL. Figure 3. Research methodology. positive solution. To measure the similarity of a place with an ideal and non-ideal state, we consider the distance of that point from the ideal and non-ideal solution. In this research, we also use the ranking matrix and the distance of each place from the ideal state. Given that multiple tourist destinations may be offered to the user, multi-criteria decision-making is used, which is made possible by TOPSIS. Hence, in each of the stud- ied cities, tourist locations are ranked based on the TOPSIS model in the form of decision matrices and by converting quality criteria into quantitative values (based on users’ com- ments), and then places with the highest points are considered as ideal points. Accordingly, the ranking of each tourist place is determined according to Table 1. Afterward and in the second step, the points that have the shortest distance with ideal points in the decision matrix are searched using the ABC algorithm and according to the selection of the desired city by the user, and finally, in each searching process with ABC, three tourist places with higher points are found according to the desired criteria and then suggested to the user. The case study in the present study is about some tourist cities in Iran, and the research methodology steps are depicted in Figure 3. Our research methodology is categorized into eight levels as follows: FUZZY INFORMATION AND ENGINEERING 37 1. Selecting some tourist cities: The methodology procedures begin with selecting some tourist cities that each city has a specific tourist attraction. The purpose of this level is that when travelers go to one of these cities, our system will recommend the best tourist destination in this city based on their preferences and experiences. The authors nominate five cities at this level that have a great number of tourist attractions. These cities are Tehran, Isfahan, Shiraz, Mashhad, and Orumieh. 2. Selecting tourist destinations for each city: Each city has several tourist spots. At this level, ideal destinations highly matched with tourists’ preferences are considered to recommend to the user in the subsequent stages. 3. Applying the TOPSIS model: After selecting touristic cities and destinations for each city, the TOPSIS model has been carried out to form a matrix for each city, consisting of four columns as follows: • Distance • Road Condition • Traffic Congestion on The Road • Tourist Attractions of the City Essentially, all responses and optimal solutions have been clarified in numeric format; thus, it is necessary to convert qualitative parameters to quantitative measures in the pre- vious section and based on Table 1. According to the TOPSIS model, the qualitative variables were converted into a quantitative element rating based on Table 1 from 1 (the lowest point) to 10 (the highest point). The ideal matrix is a matrix that produces 10 for all parts, so the nearest rate to the 10-band score implies an ideal solution, which is one of the good options to recommend to the user. 4. Distributing questionnaires on Facebook: At this level, each tourist attraction is ranked by the user through an online questionnaire on the Facebook website. The users give the mark for each destination based on four criteria mentioned in the previous level. The questionnaire is divided into four sections: distance, road condition, traffic conges- tion, and tourist attraction. The authors arranged an online questionnaire uploading it on the Facebook social media website through several accounts and asked respon- dents to rate each item based on their travel experience to those cities. Finally, 1015 questionnaires were gathered from respondents. 5. Creating a matrix for each city on the basis of TOPSIS outcome: At this level, matrices are created for every single city according to questionnaire outcomes taken from users. Matrices are formed based on the following formula: 5 4 M = C && L (14) n i=[A→1,3,5,7,10B→1,3,5,7,10C→1,3,5,7,10D→1,3,5,7,10] n=1 i=4 Where implies each city, and i = 1 means that a minimum of one city must be picked n=1 and the maximum number of cities is equal to five. Indicates four filed for the quality i=4 of tourist attractions. The upper bond of 4 means that all 4 fields must be customized. An AND operator exists between C cities and L tourist destinations since our function is only n i true when both values are true. Furthermore, L has four fields; A, B, C, D, presenting four i 38 S. FOROUZANDEH ET AL. criteria for destination quality, and every single field of the matrix can be customized by 1,3,5,7,10 values; the best quality is obtained by higher numbers. As mentioned, in relation to the weights intended for tourist places, we used the quality criteria according to Table 1, then we convert them to have quantitative criteria. Weighing and scoring for each tourist place is based on the dataset and Tripadvisor site. Also, quality criteria have been selected in such a way that users can easily evaluate the quality of the tourist places. Then, a score is considered for each quality criterion, and through this, the weighing of each tourist place is done. 6. Selecting a specific city by the user: The user chooses one of the five cities in this level, and then the recommender system starts to suggest a tourist destination to the user. 7. Applying the ABC algorithm for a searching destination with minimum distance and an ideal matrix: After choosing a city by the user, the recommender system starts to search and analyze tourist destinations in the selected city. The ABC algorithm is used to fulfill the searching process for the purpose of discovering an optimized point. This searching process is carried out in two general levels. First, the user selects preference criteria, including traffic congestion, then the system finds a solution, which includes the lowest traffic jam via employed bees. Second, onlooker bees find the nearest ideal location to the user’s preferences. 8. Recommendation to tourists: Finally, the recommender system finds three places based on the results obtained from level 7 and recommend these tourist destinations to the user. 6. Analysis and Findings In this paper, five tourist cities like Tehran, Isfahan, Shiraz, Mashhad, and Orumieh were selected for conducting a pilot study. Hence, these cities with some tourist attractions were recommended for new arrivals. Each of these cities converted to a matrix with specific val- ues on the basis of the rating system (Table 1) via the TOPSIS model. Table 2 demonstrates a matrix for each city: It is clear that every city has its specific matrix and each row of the matrix represents one tourist’s spot in this city, and all fields are rated according to Table 1. The user selects the area he/she needs to travel; then, the system suggests the best place to visit according to the importance of each factor. Secondly, the ABC algorithm is carried out, looking by employed bees for the best ideal matrix, then onlooker bees searching for other nearest ideal two matrices, proposing all to the user. The outcome of this algorithm for cities is described in Table 3 as follows: As shown in Table 3, the ABC algorithm was implemented for five cities with the outcome of different values for each city’s factors. Every city has three matrices that show three tourist destinations. At the top of the matrix, the cost of each matrix is specified, and the best matrix is defined in terms of the cost outlined in the figure. For example, with a cost of 37, the best matrix for Tehran is the second one, and it is the closest ideal matrix to the best positive ideal matrix. At this point, with 1000 repeats, the employed bee starts looking in the loop to find a first matrix based on the given value. Then, onlooker bees start searching and finding the other two matrices. Consequently, three matrices are found and recommend to the user. FUZZY INFORMATION AND ENGINEERING 39 The evaluation of a proposed model contains three sections. In the first section, the authors assumed and compared the accuracy of recommending items to a tourist based on the questionnaires. In the second section, the ABC algorithm’s accuracy about finding an optimized matrix is assessed by a specific function. In the third section, the proposed method in the present research has been implemented on the Tripadvisor dataset and compared with other algorithms. 6.1. Accuracy of the Recommender Systems For evaluating the TOPSIS model’s performance, one city is selected as an example, using an online questionnaire to assess the accuracy of this method. The outcome recommendation items are compared with users’ suggestions. If the similarity of these factors is high, it is concluded that the accuracy of our recommender system is acceptable. In order to calculate a difference between these two criteria, the Euclidean distance formula [58] was utilized as follows: d(x, y) = (X − Y ) (15) k k k=1 Table 2. Presenting tourism destinations in the form of a matrix using the TOSIS model. Tehran Esfahan Shiraz Mashhad Orumieh 110 3 710 135310 77510 53710 3 7 710 3 73710 10 3510 7 77310 5 10 10 7 7710 10 10 10 5 10 510 77710 553710 1 7510 37510 7 510 3710 5553155 10 10 75110 7 57575337510 10 1 3 10 10 510 7733510 10 5 17710 5 510 1 710 10 10 10 7773313710 57710 10 1 710 10 3335810 10 3337710 1 355 15710 75710 355710 10 10 10 10 5 57 710 10 711510 10 1355713310 10 3 55710 510 5 3577510 533110 7 10 77510 537110 55773310 3 5710 10 10 3710 17710 10 775710 3710 10 1 55710 357710 11357710 3 315 10 5317510 3 710 3315710 3 10 10 7 710 5 7573310 10 13735710 10 10 7 3 103 5 107 10 3 1 7 10107 5 10105 3 7 5 77710 73355137755710 3 510 73710 3510 1 7710 5 310 10 5310 7 5 5710 373510 3 10 355710 31715 7710 10 10 71310 310 1 17577710 7 ... ... ... ... ... Table 3. Performing the ABC algorithm and TOPSIS technique. Orumieh Mashhad Shiraz Esfahan Tehran 1 Cost 27 Best Ans Cost 25 Best Ans Cost 26 Best Ans Cost 22 Best Ans Cost 34 Best Ans 5 5 710 10 5 371 10 105 5 5 5710 710 7 2 Cost 34 Best Ans Cost 30 Best Ans Cost 32 Best Ans Cost 37 Best Ans Cost 31 Best Ans 7 7 10 10 10 7 3 10 5 7 10 10 10 10 7 10 10 7 7 7 3 Cost 37 Best Ans Cost 32 Best Ans Cost 30 Best Ans Cost 29 Best Ans Cost 29 Best Ans 10 10 10 7 10 7 5 10 3 10 7 10 10 7 7 5 7 10 7 5 40 S. FOROUZANDEH ET AL. Figure 4. Comparison of user choices and recommendation system results. Where n represents the number of characteristics, X and Y denote Euclidean vectors start- k k ing from the origin of the space, and their tips show two points. For example, ‘Esfahan’ is selected to evaluate in terms of the accuracy of the TOPSIS model. Esfahan consists of 20 tourist destinations that each destination formed into a matrix through the TOPSIS model. In this regard, a questionnaire was distributed on Facebook to ask respondents to select tourist spots based on four criteria: distance, road condition, traffic congestion, and tourist spot attractions. The numbers of 526 respondents were getting involved in this survey. They selected specific destinations according to their previous experiences of traveling to these tourist destinations. Afterward, the outcomes of online questionnaires were compared with the results of our recommender system. Results are presented in Figure 4 as follows: According to Figure 2, the blue color line indicates that user’s responses, including the places they selected, and the red color line, presents recommendation system out- comes. As can be seen, in most cases, the user’s answers and system recommendations are matched together, and it is concluded that the accuracy of our recommender system is high regarding offering places for tourists. The next comparison is related to quality. As mentioned in the previous section, a matrix with a maximum value of 10 points was formed for each column. Hence, in the best situ- ation, the sum of all columns in the ideal matrix would be 40 scores. Thus, an ideal option would be an option with the nearest distance to 40 scores in the ideal matrix. For exploring a similarity between recommender system results and users opinions, the following formula canbeused[59–62]: R = sim(C, C ) = argmax sim(C, C ) (16) c,c c,c ∈R By applying this method, the similarity between the user’s judgment and recommendation system suggestions, which is represented by C and C were examined by items such as sim (c,c ). Here, argmax picks a higher number between C and C , which means there is a com- parison between the user’s opinion and recommender system outcome for determining FUZZY INFORMATION AND ENGINEERING 41 Figure 5. Comparison of user choices and recommender system results. the nearest distance to the ideal purpose. Consequently, the outcome, which is near to 40, is selected as an ideal option. Figure 5 illustrates the results from this level. According to Figure 3, the blue color line presents users to review, and the red color line illustrates recommendation systems results. Besides, in some parts of the figure, there is a gap between the user’s suggestion and recommender system suggestions; the authors analyzed these differences based on their score and found that all suggestions by the rec- ommender system are outscored the user’s recommendation. Hence, in terms of scoring criteria, this is acceptable that the places were recommended by recommendation systems that have a near distance to the ideal matrix. 6.2. Accuracy of the ABC Algorithm In this section, the ABC algorithm’s accuracy to find the best places to recommend to users is evaluated by the function, which produces random numbers in a specific domain. This func- tion generates random numbers available in a matrix-domain, then compares these points with destinations offered by the recommender system. The random number generation function is defined as follows [42]: L = Rand(1, 10), i ∈{1, 3, 5, 7, 10} (17) i=1 This formula includes four digits ranging from 1 to 10 (TOPSIS model domain). In each cycle of performing this function, one random number has been generated, then all generated numbers sum together and calculate the cost function. The results of comparing the ABC algorithm and Random number generation function is presented in Figure 6 as follows: Both algorithms were applied five times for five cities, and the best results outcome was selected in each cycle of performing algorithms. Hence, 25 destinations were explored. The precision of the ABC algorithm was higher than the Random number generation algorithm because the outcomes of ABC have a minor distance with an ideal solution. The results of 42 S. FOROUZANDEH ET AL. Figure 6. Evaluating the accuracy of the ABC algorithm and Random number generation function. Table 4. Converting qualitative criteria to quantitative criteria on the Tripadvisor site Qualitative criteria Terrible Poor Average Very good Excellent Quantitative criteria 1 2 3 4 5 the ABC algorithm recommend appropriate options to users due to conclusive and highly valued outcomes. 6.3. Experimental Evaluation The method presented in this analysis was applied on the Tripadvisor dataset that contains the recommended hotel to the user and involves a review of 235,793 hotels collected by Crawler in one month [63–65]. The ratings provided by users are between 1 and 5, and according to the type of dataset studied by TripAdvisor, the criteria and rating associated with TOPSIS have changed in the research, and the selection of criteria for scoring is based on qualitative criteria on the Tripadvisor site and is considered as Table 4 as follows: Also, in this dataset to select each hotel has been intended for some criteria, and the users’ rating for hotels is according to Table 4 and these criteria. Hence, each hotel possesses an identification ID, and the criteria for hotels are as follows: 1. Business_Service 2. Service 3. Check in/front desk 4. Cleanliness 5. Location 6. Rooms 7. Value FUZZY INFORMATION AND ENGINEERING 43 On the basis of these seven criteria, the algorithm is then applied and tries to identify hotels that are based on each of these criteria and have an appropriate ranking in each of them to recommend to users. The accuracy of the algorithm provided in this study is also compared with other recommended system algorithms, such as Content-Based (CB) Systems, Collaborative Filtering (CF) Standard, and Collaborative (CL) Systems Algorithms. [66], that contains several algorithms (called Tourism Recommender Systems related algo- rithms). According to the following criteria, the accuracy of the recommender system in the present article was investigated by CF standard, CL, and CB algorithms on the Tripadvisor dataset. Ultimately, the authors evaluate two approaches and used precision, recall, and F1 mea- surements as recommended by Herlocker et al. to illustrate the efficacy of the proposed approach in recommending with the ABC & TOPSIS method and CF Standard and CL and CB for recommendation hotels (2004). The authors have therefore separated the ratings of each user on ABC & TOPSIS hotels into two main sets, the training set and the test set as used by Billsus and Pazzani (1998). On the training set, the model was trained, and then the test set was used to predict the Top-N hotels. In addition, the hotels were divided into two groups, irrelevant and relevant, to measure the accuracy and recall. The precision met- ric provides the probability that the hotel selected is relevant (Equation 18), and the recall metric provides the probability that the hotel selected would be selected (Equation 19). It is important to evaluate these two metrics together. The F1 measure that combines these two metrics can also be used as a single metric to evaluate the quality of recommendations, as suggested by Sarwar et al. [67]. Hotel relevant−selected Precision = (18) Hotel + Hotel relevant−selected relevant−selected Hotel relevant−selected Recall = (19) Total relevant(selected−notselected) 2 × Precision × Recall F1 = (20) Precision + Recall In order to assess the accuracy of the recommender system for the provision of hotels to users on the basis of the above criteria, a cost matrix has been established, and the mini- mum number of points associated with the created user matrix has been specified. For this purpose, it should be regarded that if all fields get an outstanding ranking, the cost of the matrix is equivalent to (7 ∗ 5 = 35), and this is the highest rating that can be earned by each matrix. The next high-quality criterion in the second of the maximum score is Very Good, with a score of 4, and the matrix cost is equal to (7 ∗ 4 = 28) if all the metrics have a Very Good rating. The recommendations presented in this research are therefore based on two appropriate parameters for users, and on this basis, there are appropriate matrices that achieve the lowest score (28), and the score achieved in the matrix must be at least equal to 4 for both of the criteria. Correspondingly, the following rules in Table 5 give an acceptable matrix: 6.4. Rules for Relevant Hotel As shown in Table 5, for providing recommendations hotels to users, there are accept- able matrices possessing all the features of that hotel or rated (‘Very good’ or ‘Excellent’) 44 S. FOROUZANDEH ET AL. Table 5. Rules for created acceptable matrixes in evaluating the accuracy of the recommender system in Tripadvisor 01 IF [ ‘Value’ = (‘ Very good’ or ‘Excellent’) OR ‘Value’ > = 4 AND Rating > = 28 ] 02 AND [‘Room’ = (‘ Very good’ or ‘Excellent’) OR ‘Room’ > = 4 AND Rating > = 28] 03 AND [‘Location’ = (‘ Very good’ or ‘Excellent’) OR ‘Location’ > = 4 AND Rating > = 28] 04 AND [‘Cleanliness’ = (‘ Very good’ or ‘Excellent’) OR ‘Cleanliness’ > = 4 AND Rating > = 28] 05 AND [‘Check in/front desk’ = (‘ Very good’ OR ‘Excellent’) OR ‘Check in/front desk’ > = 4 ] 06 AND [‘Service’ = (‘ Very good’ or ‘Excellent’) OR ‘Service’ > = 4 AND Rating > = 28] 07 AND [‘Business_Service’ = (‘ Very good’ OR ‘Excellent’) OR ‘Business_Service’ > = 4 AND Rating > = 28] 08 Then Generate Matrix: = Relevant Hotel; Figure 7. Accuracy of matrices produced in Precision and on the basis of user priorities for hotel-user recommendations on TripAdvisor. or at least possess 4 points (every user chooses a feature as a search priority among hotels). Therefore, seven times the algorithm is executed according to the seven prior- ities in selecting each hotel, and the archived matrices are determined by applying the desired approaches, and then the accuracy of each of them is achieved in different runs. Accordingly, the findings are presented in Figures 7–9. It is obvious that the diagram created in the vertical column includes Generate Matrix, and in the horizontal column involves Precision, Recall, and F1 and each of which was according to user priorities and the created rule (Every Priority > = Very Good or points 4). The number of matrix production has been made for every single diagram in ranges of 100, 200, 400, 600, 800, and 1000, which means that in the first stage, 100 matrices and the second stage 200 and finally 1000 matrices produced and specified the number of gener- ated matrices (based on percentage) with a bet (Every Priority > = Very Good or 4) and the precision of the matrices produced to provide the hotel to the users is measured and illustrated according to the users’ priorities. The important point regarding the increase of the production of matrices was that with the increase in their product ranges, the accu- racy of the matrices also increased, which shows the usefulness of the algorithm in offering recommendations to users in a large number. FUZZY INFORMATION AND ENGINEERING 45 Figure 8. Accuracy of matrices produced in Recall and on the basis of user priorities for hotel-user recommendations on TripAdvisor. Figure 9. Accuracy of matrices produced in F1 and on the basis of user priorities for hotel-user recom- mendations on TripAdvisor. 7. Conclusion Choosing a tourist destination to travel to is one of the hectic activities due to the numer- ous options travelers usually come across. There are various sorts of factors that should be considered to travel, ranging from activities to transportations, etc. Each traveler has special favors and preferences and usually faces many options. Thus, the uniform trip plan is not suitable for all kinds of tourists. A productive and customized system is essential to get some parameters via the user and offer an excellent choice for tourists to provide them a memo- rable trip. The present study provides a recommendation system to suggest an appropriate tourist destination for each type of visitors based on their preferences. This system works based on TOPSIS fuzzy model and the ABC algorithm. First of all, the TOPSIS model defines 46 S. FOROUZANDEH ET AL. a positive ideal solution in the form of a matrix with four columns, which indicates a factor that includes in this study. By distributing online questionnaires on the Facebook website, the respondent was asked to put a value of each factor out of 10 scores. Therefore, the desired indicators and criteria are evaluated using the TOPSIS model and decision matrix, and the ideal points are detected and by converting qualitative values into quantitative to solve the problem, then the distance of other points or the same tourist places in each city to the ideal points is calculated. After creating a matrix for every tourist’s destination, the ABC algorithm started to search amongst destinations and recommend the best tourist spot to users. Conclusive results of the present investigation indicate a high performance of these algorithms. The proposed recommendation system would help travelers select the best tourist attraction in the cities based on their preferences. 8. Limitation and Future Work In the present article, due to extensive tourist zones in Iran, the authors only selected the five most touristic cities; the number of cities can be increased in future research work since there are plenty of tourist cities and urban heritages in Iran that can be taken into account in future work. Another limitation of this research was about distributing the questionnaires. This task was done through online questionnaires within the Facebook website via several accounts for gathering data from users. Future works can be expanded through the sur- vey method by an in-depth interview with travelers and make face-to-face questionnaires. Besides, future studies can focus on the relationship between touristic cities and recom- mend other tourist attractions between these cities; when travelers want to travel between two cities, the recommender system can suggest some tourist destinations around these cities based on user location and routing system. Actually, through making location graphs and neighborhood matrix, and the related algorithms, more tourist hot spots can be recom- mended to travelers, particularly in far distances. Hence, future works can boost our novel recommender system to enhance the results. Disclosure statement No potential conflict of interest was reported by the author(s). Notes on contributors Saman Forouzandeh is currently a PhD student with the Faculty of Science, University of New South Wales (UNSW), Sydney, Australia. He received the M. Sc. degree in Software Engineering from Islamic Azad University, Science and Research Branch, Tehran, Iran in 2013 and B. Sc. degree of Software Engineering from University of Orumiyeh, Iran in 2009. His main research interests include Machine Learning, Graph Neural network, and Recommender Systems. s.forouzandeh@unsw.edu.au Mehrdad Rostami received the B. Sc. degree with honors in Computer Engineering from Razi Uni- versity, Kermanshah, Iran, in 2012, and the M. Sc. degree with honors in Artificial Intelligence from University of Kurdistan, Sanandaj, Iran, in 2014. He is currently a Ph. D. student and researcher at Uni- versity of Oulu, Oulu, Finland. His research interests include pattern recognition, medical diagnosis, machine learning, big data mining, and bioinformatics. FUZZY INFORMATION AND ENGINEERING 47 Kamal Berahmand received the M.S. degree in information technology engineering from Azarbai- jan Shahid Madani University, Tabriz, Iran, in 2017. He is currently a Ph. D. student with the Faculty of Computer Engineering and Information Technology, Queensland University of Technology, Brisbane, Australia. His current research interests include Machine learning, network science, graph neural network, and knowledge graph. ORCID Saman Forouzandeh http://orcid.org/0000-0002-5952-156X References [1] Tugaonkar PS, Chitre V. Survey on recent methodologies used for recommender system. 2016 International Conference on Electrical, Electronics, and Optimization Techniques (ICEEOT). 2016. IEEE. [2] Lu J, Wu D, Mao M, et al. Recommender system application developments: a survey. Decis Support Syst. 2015;74:12–32. [3] Esmaeili L, Mardani S, Golpayegani SAH, et al. A novel tourism recommender system in the context of social commerce. Expert Syst Appl. 2020;149:113301. [4] Pujahari A, Sisodia DS. Model-Based Collaborative Filtering for Recommender Systems: An Empirical Survey. 2020 First International Conference on Power, Control and Computing Tech- nologies (ICPC2T). 2020. IEEE. [5] Alhijawi B, Kilani Y. The recommender system: A survey. Int J Adv Intell Paradigms. 2020;15(3):229–251. [6] Ashley-Dejo E, Ngwira S, Zuva T. A survey of context-aware recommender system and services. 2015 International Conference on Computing, Communication and Security (ICCCS). 2015. IEEE. [7] Osadchiy T, Poliakov I, Olivier P, et al. Recommender system based on pairwise association rules. Expert Syst Appl. 2019;115:535–542. [8] Forouzandeh S, Aghdam AR. Health recommender system in social networks: A case of Facebook. Webology. 2019;16(1):40–54. [9] Forouzandeh S, Aghdam AR, Forouzandeh S, et al. Addressing the cold-start problem using data mining techniques and improving recommender systems by cuckoo algorithm: A case study of Facebook. Comput Sci Eng. 2020;22(4):62–73. [10] Braunhofer M, Ricci F. Contextual information elicitation in travel recommender systems. In: Inversini A, Schegg R, editors. Information and Communication Technologies in Tourism 2016. Cham: Springer; 2016. p. 579–592. DOI:10.1007/978-3-319-28231-2_42 [11] Renjith S, Sreekumar A, Jathavedan M. An extensive study on the evolution of context-aware personalized travel recommender systems. Inf Process Manag. 2020;57(1):102078. [12] Kenteris M, Gavalas D, Mpitziopoulos A. A mobile tourism recommender system. The IEEE symposium on Computers and communications. 2010. IEEE. [13] Logesh R, Subramaniyaswamy V. Exploring hybrid recommender systems for personalized travel applications. Cognitive informatics and soft computing. 2019, Springer. p. 535-544. [14] Gavalas D, Kasapakis V, Konstantopoulos C, et al. A survey on mobile tourism recommender systems. 2013 Third International Conference on Communications and Information Technology (ICCIT). 2013. IEEE. [15] Ricci F. Recommender systems in tourism. Handbook of e-Tourism, 2020: p. 1-18. [16] Gavalas D, Konstantopoulos C, Mastakas K, et al. Mobile recommender systems in tourism. J Network Comput Appl. 2014;39:319–333. [17] Zeng B, Gerritsen R. What do we know about social media in tourism? A review. Tourism Manage Perspect. 2014;10:27–36. [18] Huang J-H, Peng K-H. Fuzzy rasch model in TOPSIS: A new approach for generating fuzzy num- bers to assess the competitiveness of the tourism industries in asian countries. Tourism Manage. 2012;33(2):456–465. 48 S. FOROUZANDEH ET AL. [19] White WG, Chandrasekar V. TOPSIS to optimize performance, reliability, and life cycle costs dur- ing analysis of alternatives. 2017 Annual Reliability and Maintainability Symposium (RAMS). 2017. IEEE. [20] Forouzandeh S, Berahmand K, Rostami M. Presentation a trust walker for rating prediction in recommender system with biased random walk: effects of H-index centrality, similarity in items and friends. Eng Appl Artif Intell. 2021;104:104325. [21] Forouzandeh S, Berahmand K, Rostami M. Presentation of a recommender system with ensemble learning and graph embedding: a case on MovieLens. Multimed Tools Appl. 2020;30:1–28. [22] Forouzandeh S, Sheikhahmadi A, Rezaei Aghdam A, et al. New centrality measure for nodes based on user social status and behavior on Facebook. Int J Web Inf Sys. 2018;14(2):158– [23] Rostami M, Berahmand K, Forouzandeh S. A novel community detection based genetic algorithm for feature selection. J Big Data. 2021;8(1):1–27. [24] Rostami M, Berahmand K, Forouzandeh S. A novel method of constrained feature selection by the measurement of pairwise constraints uncertainty. J Big Data. 2020;7(1):1–21. [25] Berahmand K, Nasiri E, Forouzandeh S, et al. A preference random walk algorithm for link pre- diction through mutual influence nodes in complex networks. J King Saud Univ Comput Inf Sci. 2021. DOI:10.1016/j.jksuci.2021.05.006 [26] Forouzandeh S, Aghdam AR, Barkhordari M, et al. Recommender system for users of internet of things (IOT). IJCSNS. 2017;17(8):46. [27] Forouzandeh S, Soltanpanah H, Sheikhahmadi A. Application of data mining in designing a recommender system on social networks. Int J Comput Appl. 2015;124(1):32–38. [28] Rafsanjani AHN, Salim N, Aghdam AR, et al. Recommendation systems: a review. Int J Comput Eng Res. 2013;3(5):47–52. [29] Munuswamy S, Saranya MS, Ganapathy S, et al. Sentiment analysis techniques for social media- based recommendation systems. Natl Acad Sci Lett. 2021;44:281–287. [30] Kulkarni PV, Rai S, Kale R. Recommender System in eLearning: A Survey. Proceeding of Interna- tional Conference on Computational Science and Applications. 2020. Springer. [31] Wang X, Li XR, Zhen F, et al. How smart is your tourist attraction? Measuring Tourist Pref- erences of Smart Tourism Attractions via a FCEM-AHP and IPA Approach. Tourism Manage. 2016;54:309–320. [32] Han D-I, Jung T, Gibson A. Dublin AR: Implementing Augmented Reality in Tourism, in Informa- tion and Communication Technologies in Tourism 2014. 2013, Springer. p. 511-523. [33] Garcia I, Sebastia L, Onaindia E. On the design of individual and group recommender systems for tourism. Expert Syst Appl. 2011;38(6):7683–7692. [34] Nilashi M, bin Ibrahim O, Ithnin N, et al. A multi-criteria collaborative filtering recommender system for the tourism domain using expectation maximization (EM) and PCA–ANFIS. Electron Commer Res Appl. 2015;14(6):542–562. [35] Bucur C. Using opinion mining techniques in tourism. Procedia Econ Finance. 2015;23:1666–1673. [36] Braunhofer M, Ricci F, Lamche B, et al. A context-aware model for proactive recommender systems in the tourism domain. Proceedings of the 17th International Conference on Human- Computer Interaction with Mobile Devices and Services Adjunct. 2015. [37] Jaiswal A, Mishra R. Cloud service selection using topsis and fuzzy topsis with ahp and anp. Proceedings of the 2017 International Conference on Machine Learning and Soft Computing. [38] Alrababah SAA, Gan KH, Tan T-P. Comparative analysis of MCDM methods for Product Aspect Ranking: TOPSIS and VIKOR. 2017 8th International Conference on Information and Communi- cation Systems (ICICS). 2017. IEEE. [39] Liu L, Xu J, Liao SS, et al. A real-time personalized route recommendation system for self-drive tourists based on vehicle to vehicle communication. Expert Syst Appl. 2014;41(7):3409–3417. [40] Benitez JM, Martín JC, Román C. Using fuzzy number for measuring quality of service in the hotel industry. Tourism Manage. 2007;28(2):544–555. [41] Chou T-Y, Hsu C-L, Chen M-C. A fuzzy multi-criteria decision model for international tourist hotels location selection. Int J Hosp Manag. 2008;27(2):293–301. FUZZY INFORMATION AND ENGINEERING 49 [42] Ju C, Xu C. A new collaborative recommendation approach based on users clustering using artificial bee colony algorithm. Sci World J. 2013;2013:1–9. [43] Hsu C-C, Chen H-C, Huang K-K, et al. A personalized auxiliary material recommendation system based on learning style on Facebook applying an artificial bee colony algorithm. Comput Math Appl. 2012;64(5):1506–1513. [44] Roy A, Tavana M, Banerjee S, et al. A secured context-aware tourism recommender system using artificial bee colony and simulated annealing. Int J Appl Manage Sci. 2016;8(2):93–113. [45] Hamdan S, Cheaitou A. Green supplier selection and order allocation using an integrated fuzzy TOPSIS, AHP and IP approach. 2015 International Conference on Industrial Engineering and Operations Management (IEOM). 2015. IEEE. [46] Solanki R, Gulati G, Tiwari A, et al. A correlation based Intuitionistic fuzzy TOPSIS method on sup- plier selection problem. 2016 IEEE International Conference on Fuzzy Systems (FUZZ-IEEE). 2016. IEEE. [47] Tzeng G-H, Huang J-J. Multiple attribute decision making: methods and applications. New York: CRC press; 2011. [48] Behzadian M, Khanmohammadi Otaghsara S, Yazdani M, et al. A state-of the-art survey of TOPSIS applications. Expert Syst Appl. 2012;39(17):13051–13069. [49] Lu X, Chang W, Zhou S. Quality evaluation of diesel engine using an improved TOPSIS method based on K-means algorithm. 2016 35th Chinese Control Conference (CCC). 2016. IEEE. [50] Wei L, Qin T, Yan Y, et al. Competitive ability evaluation for generation enterprises based on com- binational weight and improved TOPSIS. 2016 Chinese Control and Decision Conference (CCDC). 2016. IEEE. [51] Chen X, Tianfield H, Li K. Self-adaptive differential artificial bee colony algorithm for global optimization problems. Swarm Evol Comput. 2019;45:70–91. [52] Hannam K, Butler G, Paris CM. Developments and key issues in tourism mobilities. Ann Tourism Res. 2014;44:171–185. [53] Lerato M, Esan OA, Ebunoluwa AD, et al. A survey of recommender system feedback tech- niques, comparison and evaluation metrics. 2015 International Conference on Computing, Communication and Security (ICCCS). [54] Bouchibane F, Bensebti M. Parameter tuning of Artificial Bee Colony algorithm for energy effi- ciency optimization in massive MIMO systems. 2017 Seminar on Detection Systems Architectures and Technologies (DAT). 2017. IEEE. [55] Bansal JC, Gopal A, Nagar AK. Stability analysis of artificial bee colony optimization algorithm. Swarm Evol Comput. 2018;41:9–19. [56] Zhou J, Yao X, Chan FTS, et al. An individual dependent multi-colony artificial bee colony algorithm. Inf Sci (Ny). 2019;485:114–140. [57] Gao W-f, Huang L-l, Wang J, et al. Enhanced artificial bee colony algorithm through differential evolution. Appl Soft Comput. 2016;48:137–150. [58] Ricci F, Rokach L, Shapira B. Introduction to recommender systems handbook. In: Recommender systems handbook. Boston (MA): Springer; 2011.p.1–35. [59] Nagpal D, Kaur S, Gujral S, et al. Fr: A recommender for finding faculty based on CF technique. Procedia Comput Sci. 2015;70:499–507. [60] Karimi M, Jannach D, Jugovac M. News recommender systems – survey and roads ahead. Inf Process Manag. 2018;54(6):1203–1227. [61] Bentaleb A, El Bouzekri Y, Lahcen AA, et al. Context Aware Recommender Systems for Tourism: A Concise Review. 2018 IEEE 5th International Congress on Information Science and Technology (CiSt). 2018. IEEE. [62] Garcia LM, Aciar S, Mendoza R, et al. Smart tourism platform based on microservice architecture and recommender services. International Conference on Mobile Web and Intelligent Information Systems. 2018. Springer. [63] Marcheggiani D, Täckström O, Esuli A, et al. Hierarchical multi-label conditional random fields for aspect-oriented opinion mining. European conference on information retrieval. 2014. Springer. 50 S. FOROUZANDEH ET AL. [64] Wang H, Lu Y, Zhai C. Latent aspect rating analysis on review text data: a rating regression approach. Proceedings of the 16th ACM SIGKDD international conference on Knowledge dis- covery and data mining. 2010. [65] Nilashi M, Ibrahim O, Yadegaridehkordi E, et al. Travelers decision making using online review in social network sites: A case on TripAdvisor. J Comput Sci. 2018;28:168–179. [66] Borràs J, Moreno A, Valls A. Intelligent tourism recommender systems: A survey. Expert Syst Appl. 2014;41(16):7370–7389. [67] Nilashi M, Ibrahim O, Yadegaridehkordi E, et al. Preference learning for eco-friendly hotels recom- mendation: A multi-criteria collaborative filtering approach. J Cleaner Prod. 2019;215:767–783.

Journal

Fuzzy Information and Engineering – Taylor & Francis

Published: Jan 2, 2022

Keywords: Recommender systems; ABC algorithm; topsis model; tourism; Facebook social network

There are no references for this article.