Variational Fuzzy Neural Network Algorithm for Music Intelligence Marketing Strategy Optimization
Variational Fuzzy Neural Network Algorithm for Music Intelligence Marketing Strategy Optimization
Sun, Juan
2022-01-06 00:00:00
Hindawi Computational Intelligence and Neuroscience Volume 2022, Article ID 9051058, 10 pages https://doi.org/10.1155/2022/9051058 Research Article Variational Fuzzy Neural Network Algorithm for Music Intelligence Marketing Strategy Optimization Juan Sun Department of Music, Handan University, Handan City, Hebei Province 056005, China Correspondence should be addressed to Juan Sun; sj03011015@hdc.edu.cn Received 10 September 2021; Revised 7 November 2021; Accepted 27 December 2021; Published 6 January 2022 Academic Editor: Suneet Kumar Gupta Copyright © 2022 Juan Sun. *is is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. In this paper, we use a variational fuzzy neural network algorithm to conduct an in-depth analysis and research on the optimization of music intelligent marketing strategy. *e music recommendation system proposed in this paper includes a user modelling module, audio feature extraction module, and recommendation algorithm module. *e basic idea of the rec- ommendation algorithm is as follows: firstly, the historical behavioural information of music users is collected, and the user preference model is constructed by using the method of matrix decomposition of the hidden semantic model; then, the audio resources in the system are preprocessed and the spectrum map that can represent the music features is extracted; the similarity between the user’s preferred features and the music potential features are calculated to generate recommendations for the target user. *e user-music dataset for model training and testing is constructed in-house, and the network model structure used for system experiments is designed based on a typical convolutional neural network model, while the model training tuning parameters are compared and selected. Finally, the model is trained and tested in this paper, and the system is evaluated in terms of both prediction rating accuracy and recommendation list generation accuracy using root mean square error, accuracy, recall, and F1 value as recommendation quality evaluation metrics. *e experimental results show that the recommendation algorithm in this paper has certain feasibility and effectiveness. Compared with other traditional music recommendation algorithms, this paper makes full use of the powerful advantage of deep neural networks to automatically extract features and obtain higher-level music feature representations from the audio content, while incorporating the historical behavioural information of user interactions with music, which can effectively alleviate the problems such as cold start in recommendation systems. problem of information overload, search engines and 1. Introduction recommendation systems have emerged as two comple- *e rapid development of the Internet has brought a large mentary tools. However, for users, search engines are amount of information to people, meeting their needs for passive. Users need to have some knowledge of the in- information in the information age and benefiting them formation they want and submit their search terms to the from it, but it has also brought the problem of information search engine before the system backend will return the overload, and both consumers and producers of infor- corresponding result information, while recommendation mation have encountered great challenges. Faced with the systems are active, users do not need to be overly familiar huge amount of information, as consumers of information, with the information, products, and so on and do not need people cannot get the information that is useful to them, to provide clear needs, and the system will analyze the the efficiency of using information is reduced instead, and a user’s historical behaviour and actively recommend to the large amount of invalid information plagues human daily user information that they may be interested in or need [2]. life; as producers of information, it becomes very difficult to In the music market, with the continuous development and make the information they produce stand out and be application of digital multimedia technology, increased noticed by most users [1]. In response to the growing music industries are turning to online music services, but as 2 Computational Intelligence and Neuroscience differences of different types of users’ digital music infor- music libraries become larger and larger and music re- sources become increasingly abundant [3]. mation acquiring behaviour, to improve the music. It is expected to be of great value and use in improving music Recommendation system technology has been improved by researchers in recent years, and researchers have intro- information theory and user information behaviour theory, duced many algorithms one after another in their research guiding users to effectively acquire digital music information work and practical applications to enable more accurate and the healthy development of digital music industry. optimization of recommendation systems. Research insti- tutions or companies are faced with problems such as lack of 2. Status of Research prediction accuracy of recommendation algorithms, the cold start of recommendation systems, and how to achieve *e development of Internet technology and its ripple effects maximum structural engineering of recommendation sys- have brought a lot of conveniences, intelligence, and en- tems. Solving these problems requires continued research tertainment to users’ lives. Users are no longer bound by and practice by a wide range of scholars [4]. Initializing the time, place, and environment and can read books, listen to weight with the smallest possible value will easily lead to too music, and watch videos remotely anytime and anywhere. At long network training time, too many iterations, and easy to the same time, user transactions are gradually replaced by fall into a local optimal solution. *e research significance of online transactions, which means that the marketing model recommendation algorithms helps e-commerce companies of digital content is also transitioning from traditional understand user behaviour habits and expand their business marketing to online marketing [7]. From the user’s per- functions. In this paper, we study the recommendation spective, the recommendation system addresses the implicit algorithm that fits the business scenario, how the recom- needs faced by users such as information overload, lack of mendation engine works. *e optimization direction of the knowledge, and limited time [8]. With the massive music algorithm is given in conjunction with the actual problem library, users can search, listen, or share their favourite songs [5]. *e idea of solving the problem of cold start and user anytime and anywhere. However, the massive space of interest prediction in the music recommendation scenario is digital music exceeds users’ basic needs and ability to filter given, which is a guideline for the implementation in rec- information, resulting in severe information overload [9]. ommendation system engineering applications. *e rec- When users select their preferred songs, they must spend ommendation system can reach the demand of a time listening to them to know which songs they prefer, and personalized selection of users, help e-commerce enterprises this evaluation cost is unrecoverable, and users must listen to to better serve users, improve competitiveness and achieve songs that they do not prefer to give them an appropriate profitability. With the increasing demand for music infor- rating. *e recommendation system digs into the user’s mation and the access and utilization of music information music preferences based on their behavioural history and and the rapid development of the music industry, domestic then quickly finds the music of interest for them [10]. In and foreign scholars have actively explored the issues related addition, recommendation systems greatly improve user to music information [6]. *eir research mainly involves the satisfaction, surprise, and listening experience. For example, concept, characteristics,and functionsof music information, the popular real-time song streaming recommendations music information demand and behaviour, music infor- allow users to see what they want, giving them the initiative mation retrieval, library music information resource con- to get short-term “rewards” by simply swiping and im- struction and service, digital music intellectual property mersing them in the recommendation service. rights, the music industry, and other aspects. However, there From a data perspective, the recommendation system are not many studies devoted to the basic theory of digital mitigates the negative impact of the Matthew effect, while music information and music information acquisition be- enhancing the ability to tap into long-tail content. In the haviour, the research content is not comprehensive, the context of massive digital music, the traditional operation research depth is not enough, there is a lack of in-depth strategy makes the Matthew effect particularly obvious [11]. discussion on theoretical issues such as the motivation of *e long-lasting invariance of the popular song charts makes users’ digital music information acquisition behaviour, the it difficult to discover new songs on the shelves and exac- factors influencing the object of acquisition behaviour, the erbates the problem of the sparsity of user listening be- mechanism of choosing the way of acquisition behaviour, haviour [12]. *e recommendation system, through and so on, and there is a lack of comprehensive and sys- content-based filtering and model mining methods, allows tematic investigation and analysis of users’ digital music new songs to be recommended to users, and these new songs information acquisition behaviour and comparative analysis match users’ music preferences [13]. Overall, the recom- of different types of users. mendation system avoids the reoccurrence of the 2/8 phe- In this paper, based on the existing relevant research nomenon and improves the consistency of the user-song results and exploring the basic theories of music information interaction matrix. From the perspective of music artists, the and music information behaviour, we conduct a more in- recommendation system allows each artist’s music to be depth study on the motives of users’ digital music infor- recommended to groups of users who may like their work, mation acquiring behaviour, the objects of acquiring be- which greatly improves the distribution efficiency and haviour, and the mechanical mechanism of acquiring publicity ratio of music works. For unknown artists, it is behaviour mode, investigate and analyze the commonality of often expensive to get their music to be sampled or pur- users’ digital music information acquiring behaviour and the chased by users who might like it [14]. Today’s Computational Intelligence and Neuroscience 3 recommendation system helps them to quickly increase the robustness but also can achieve false alarm resistance and popularity of their works by recommending them directly to fault detection for abnormal modes such as hardware failure and strong environmental interference. those listeners who are more likely to be interested [15]. At the same time, this efficient content distribution mechanism Based on the prerequisites for the fusion of RBF neural also inspires and motivates artists to create. networks with TS fuzzy inference systems, this chapter Admittedly, these current algorithms have different proposes a self-organizing RBF fuzzy neural network advantages and disadvantages, and each algorithm is suitable (RDSO-TS-RBFNN) applied to a fused TS model for robust for different scenarios of application, and different algo- detection. Figure 1 shows the model structure of the model rithms will bring different performances [16]. *e focus on applied to the binary classification problem, which is like different technologies in the recommendation system will be that of the traditional TS-type fuzzy neural network, i.e., presented in different product forms. In the past, the both contain a network of antecedents and back pieces, and products of recommendation systems appeared in the form the number of neurons in the hidden layer is the same as the of relevant recommendations in the corners of the main number of fuzzy subsets contained in each input variable, i.e., each neuron corresponds to a specific fuzzy rule. *e products, and this form of recommendation products was only a supporting role, and if such a recommendation overall output of the network is a linear combination of the system failed, it did not affect much. Today, recommen- activation strength of the normalized fuzzy rules in the dation products have evolved into information flow, as the antecedent network and the output of the fuzzy rules in the main form of Internet products carried. From the initial consequent network [18]. recommendations of friends on social networking sites, graphic information on news apps to the current short video E � t + y . (1) r r recommendations. Information flow has become a form of r�1 recommendation system products and compared to the *e kernel approach has always been in recent years in form of relevant recommendations, information flow has scientific research as an optional and efficient way to solve become a powerful tool for user time and attention difficult problems when linear thinking cannot be applied to harvesting. the input data internally. *e second step is to imitate the internal correlation of the training set to obtain the pa- 3. Analysis of Music Intelligence Marketing rameter set of the predictive model when the expression is Strategies for Variational Fuzzy Neural obtained in the high-dimensional range. *e thinking of the Network Algorithms specific approachaimsat projectingthe problem tobe solved into a higher dimensional range, where the form of the 3.1. Variational Fuzzy Neural Network Algorithm. *is solution becomes linear in that range. If linear thinking chapter first addresses the problem that the model cannot cannot be applied to solve the puzzle within the input data, suppress the output of outlier sample points due to the the initial low-dimensional part within the initial input inclusion of a defuzzification layer in the antecedent part of sample is expressed in the high-dimensional range using a the network when the traditional TS-RBF neural network is specific mapping process; the second step obtains the set of applied to model the classification detection task. To reduce parameters of the prediction model by mimicking the in- the false alarms caused by abnormal samples under normal ternal correlation of the training set when the expression is conditions of industrial inspection and to extend the fault obtained in the high-dimensional range utilizing linear detection function, a new bias term is added to the com- thinking. putation of the activation strength of the normalized fuzzy rules in the precursor part of the TS-type fuzzy neural φ(x): x ⟶ β , network so that the model can suppress the output of outlier (2) sample points. Secondly, a self-organization mechanism of K(p, q) � φ(p) · φq . the neurons (fuzzy rules) of the TS-type fuzzy neural net- work is proposed to optimize the structure of the model by As can be seen from the above expression, the central three mechanisms: adding, merging, and deleting to im- thrust and outstanding advantage of the method is that in prove the robustness and generalization ability of the model. the process of machine learning, i.e., when the model can Finally, to solve the problem that the stochastic gradient apply the internal correlation of the training set to the descentalgorithmiseasyto fallintothelocal optimalpoint,a predicted output of the test set after mimicking it, only the set of adaptive learning mechanisms is proposed to enhance kernel function that can easily produce the result is assumed, the fitting ability of the model [17]. Normalize the data. and no specific form describing the mapping relationship is When adjusting the error, the adjustment of the weight is revealed. Wavelets are well known for their excellent pre- converted into the adjustment of the cloud model expec- dictions that are not limited to a small number of sample tation and entropy, which reduces the number of iterations datasets, their ability to make outstanding performances of the error to a certain extent, thereby improving the even when additionally predicting new sample sets, and their operating efficiency of the algorithm. *e experiments on unique advantage in dealing with signals related to temporal normal working conditions of three-band flame detection continuity, allowing for clearer, more concise, and more and data anomalies verify that the model not only has intuitive representations of signals. *is chapter, therefore, significant improvement in classification accuracy and adds wavelet functions to convolutional fuzzy polynomial 4 Computational Intelligence and Neuroscience Hidden Fuzzy x1 layer Input layer Anteceden Output x2 t network TS-RBF Hidden neural x3 layer network y1 classificati Hidden on layer y2 detection x4 Output False y3 layer alarms x5 y4 Normalizat Abnormal ion layer samples y5 x6 Hidden Fault layer detection y6 Figure 1: Structure of variational fuzzy neural network model. neurons to create convolutional fuzzy wavelet polynomial *ings are grouped “ gives rise to collaborative filtering neural networks. *ere is no doubt that wavelet selection is based on songs, i.e., first calculate the list of most similar crucial here, and it is important to consider the wavelet that songs for each song, and then recommend similar songs to most closely fits the model and can handle the actual users based on their favourite songs. People are divided by problem when building the model; it is especially important groups” gives rise to user-based collaborative filtering, i.e., to select the optimal one among the many wavelet functions. first finding user groups with similar interests to the target user, and then recommending songs that the user group has liked or purchased to the target user, as shown in Figure 2. (3) φ(t) � exp expjw t. According to the different ways of implementation, col- laborative filtering algorithms can be divided into two major Since the application of this wavelet is considered to be categories: one is memory-based collaborative filtering and able to carry out the mapping of the input signal data from the other is model-based collaborative filtering. Considering the time domain to the frequency domain representation, that their core idea is to calculate the similarity between the intrinsic mapping logic at this point is expressed in the users or songs, this thesis will focus on memory-based following equation: collaborative filtering and model-based collaborative filter- ing with the user in mind. √�� � k w + w ⎛ ⎝ ⎞ ⎠ ψ(w) � kπ exp (4) . However, the memory-based collaborative filtering al- gorithm has certain limitations, the song cold start problem. For a new song on the shelf, since there are only a few user Collaborative filtering is a very classical recommenda- actions, which is equivalent to the fact that the columns tion algorithm; the core idea is to use the group behaviour to corresponding to the song in the user behaviour matrix are find similarity, that is, the similarity between users or the zero, it is impossible to calculate the similar songs of the similarity between songs, by calculating the similarity to song, and at the same time the song will not appear in the make decisions and recommendations for users. *is idea similar list of other songs, and it is impossible to recommend originated from a very simple natural philosophical thought: the new song out. Streaming platforms generally have a “things gather by class, people are divided by group.” library of millions of music tracks, and the user-song in- � � � � � �2 � �2 T teraction matrix is usually extremely sparse compared to the � � � � max r + p q − λ�p � − �q � . uv u u v v (5) number of songs users have listened to. *us, using this (u,v)∈A sparse matrix to predict the recommendation results is Computational Intelligence and Neuroscience 5 CM-FNN uses the Initializes the weights by Often too complicated in more mature BP selecting random determining parameters algorithm Physical test results Climate change Self-comparative concern Process of network Small as possible values Initialization training involves Particle swarm algorithm Exercise adherence Exercise internal External regulation motivation Parameter Traditiona Annealing adjustment l method algorithm Intraphoto adjustment Leads to long Studying the specific BP algorithm Integrated regulation Motivation network training content dimensions Interview method Identity adjustment Internal motivation goal-oriented behavior model Many iterations Local optimal Respond Prepare Recover Exercise persistence Exercise intention solutions Figure 2: Collaborative filtering based on songs. imprecise. Collaborative filtering uses a group wisdom 3.2. Experiments in Optimizing Music Intelligence Marketing Strategies. Current music recommendation methods can recommendation mechanism, which makes the model more inclined to recommend popular music, resulting in popular only mine the general relationship between users and songs songs gaining increased exposure, while generally long-tail and cannot distinguish the differential preferences of dif- songs have very little, few, or no user actions, thus making it ferent users for the same song. To this end, this chapter more difficult for this algorithm to distribute long-tail songs proposes a music recommendation algorithm based on to more users. Music preferences are more personalized. multilayer attentional representation, which uses user at- User preferences are relatively independent in the music tribute information and song content information to learn domain, i.e., the assumption that users with similar be- song representations from multiple dimensions and mine haviours may have the same musical tastes has not been the preference relationship between users and songs. To distinguish the differential preferences of users on multi- effectively tested in music platforms. *e object of users’ digital music information acqui- domain features of songs, a user feature-dependent attention network is designed; to distinguish the differential prefer- sition behaviour is mainly digital music art information. *ere are various types of digital music art information, ences of different historical behaviours on users and to mine and the internal and external quality varies greatly, so what the temporal dependencies of user behaviours, a song-de- type and quality of digital music art information users pendent attention network is designed. Finally, the SoftMax acquire will be influenced by various factors, and different function is used to calculate the distribution of users’ factors will affect different aspects of digital music art preferences for candidate songs and generate recommen- information acquisition objects. *eoretically, exploring dations. Experimental results on the 30 Music and MIGU the types of digital music information acquisition behav- datasets show that the model proposed in this chapter iour objects and their influencing factors is one of the key achieves significant improvements in both Recall and MRR compared to existing recommendation algorithms. issues in the study of users’ digital music information acquisition behaviour. In this chapter, based on describing In digital music streaming platforms, users’ listening the composition of user digital music information acqui- behaviour is often played in an automated manner. Tradi- sition behaviour objects, the factors influencing user digital tional recommendation algorithms mine the user’s music music information acquisition behaviour objects are firstly preferences from historical behavioural information to build determined by expert consultation method, and then the a model that predicts the next song the user is likely to like hierarchical relationship of the influencing factors of each [19]. *is approach does not consider the temporal de- digital music information acquisition behaviour object is pendencies in the listening sessions and ignores the huge calculated by explaining the structural model method, and behavioural noise data brought by the user’s listening finally, the hierarchical relationship of the influencing process, which brings great disturbance and challenge to factors of user digital music information acquisition be- music recommendation and thus affects the accuracy of the model in mining users’ music preferences. In addition, users’ haviour objects is explained and the role mechanism of each factor is analyzed. preferences for a song are often more specific and fine- 6 Computational Intelligence and Neuroscience information of interest more accurately to the user and grained, and it is crucial to improve the accuracy of rec- ommendation results if the local information features of the better give the reason for the recommendation to the user, as shown in Figure 4. song are prioritized and input into the model for training. To this end, this chapter proposes a temporal recommendation Even in the normal working condition due to the un- algorithm based on the attention mechanism. In general, certainty of the hardware, data processing process of un- long-tail songs have very few operating behaviours, with few reasonable and other factors will also generate part of the user operations or even no user operations. *erefore, this abnormal sample points, but in the normal working con- algorithm is more difficult to distribute long-tail songs to dition, the continuity and frequency of the abnormal sample more users. Music preferences are more personalized. In the points are much lower than the continuity and frequency of the abnormal sample points under the fault condition. Based music field, user preferences are relatively independent. It is a two-layer neural network, the first layer, used to extract on the above fact, a predetermined value can be set in the practical application [21]. If the frequency of the current high-dimensional semantic representation from the multi- dimensional information of songs as song representations; outlier sample is lower than this value, the recognition result of the outlier sample will be output as “reject recognition” to the secondlayer, the listofsongs inthe userlistening session. *e initial play sequence is used to learn the temporal de- avoid false alarm, while if the frequency is higher than the pendencies of listening behaviour using a threshold-con- predetermined value, it can be judged that “fault exists in the trolledrecurrent neural network,and auser-drivenattention system” to achieve fault detection. mechanism is constructed to learn the importance of each Now it is only necessary to suppress the output of the historical listening behaviour, to achieve the effect of re- outlier samples to a definite value to achieve the false alarm ducing the noise data, as shown in Figure 3. resistance of the model and the fault detection of the system, while the fuzzy activation strength of all the fuzzy rules in the Explicit user feedback compared to implicit user feedback requires users to rate and comment, which can be TS-RBF fuzzy neural network will be very small during the calculation of the anomalous samples, i.e., there is no a good indication of their true preferences for items. Im- plicit user feedback differs from it in that it requires suitable fuzzy rule in the current network to fit this situation. Unfortunately, some unsuitable fuzzy rules under the action analysis and processing of some data, and if the processed data is not guaranteed to be accurate or the analyzed be- of normalization (defuzzification) layer dominate the whole haviour has a large noise interference, then implicit user inference processofthe modelandplay a majorcontribution feedback will have a less obvious effect. Implicit user to the output, which leads to the difficulty of the model to feedback requires the selected behavioural features to be suppress the output of outlier sample points to a particular clear, and on Internet e-commerce sites, the user’s purchase value. behaviour is the implicit feedback that can show the user’s preferences and can be selected as a behavioural feature. It 4. Results and Analysis is important to note that the selected behavioural features may vary in different scenario applications [20]. *e rec- 4.1. Variational Fuzzy Neural Network Performance Analysis. ommendation engine will adopt different recommendation It can be seen from Figure 5 that compared to the SO-TS- mechanisms. One recommendation mechanism is to an- RBFNN and TS-RBFNN models, the adaptive learning al- alyze the data in the data source to get certain rules and gorithm proposed in this chapter can effectively overcome recommend items to the user based on these rules. Another the local optimum problem in the optimization process of recommendation mechanism is to predict and calculate the the traditional stochastic gradient descent method. Figure 5 target user’s preference for the item and recommend the shows the comparison of simulation results between dif- item to the user based on the ranking of the preference. It ferent models. Although the differences in recognition rates should be noted that in different scenarios, the selected are small, the four models differ significantly in RMSE error, behaviour characteristics may be different. *e recom- an evaluation scale more suitable for evaluating model ac- mendation engine will adopt different recommendation curacy, generalization ability, and robustness. *e RDSO- mechanisms. A recommendation mechanism is to analyze TS-RBFNN model proposed in this chapter is the best the data in the data source to obtain certain rules and among the four models in terms of both accuracy and model recommend items to users according to these rules. It is generalization ability, while the PSO algorithm can signif- very simple to design a recommendation engine using only icantly optimize the model fitting accuracy on the training one recommendation strategy in a recommendation sys- set, but the method has the worst generalization ability on tem, and we often use a recommendation strategy that the test set. *e main reason for the limited generalization incorporates multiple recommendation mechanisms in line ability of the PSO-TS-RBFNN model is that it overly pursues with business scenarios to make the recommendation high accuracy in the training phase and neglects the gen- system achieve better recommendation results. For ex- eralization ability of the model and the inference ability of ample, in the Youku recommendation system, the rec- the fuzzy system. In the training phase, the semantic in- ommendation mechanisms considered for fusion include formation in the fuzzy set will be largely destroyed in the recommending similar videos based on videos viewed or optimization process of the PSO algorithm, and some un- commented on by users’ historical behaviour, recom- reasonable fuzzy rules will be generated in the system, which mending popular videos based on the popularity of videos, may have low influence or even not activated in the training and so on. *e purpose is to recommend the video set, but in the case of data uncertainty in the test set, it is very Computational Intelligence and Neuroscience 7 Played in an automated Temporal Music streaming platforms Users' listening behavior manner dependencies Traditional recommendation Music preferences from Behavioral information algorithm Song 1 Song 2 Song 3 Song 4 Song 5 Song 6 Song 7 Song 8 Song 9 listening Input 4 Input 3 Input 2 Input 1 sessions Input 6 Input 7 Input 8 Input 9 Input 11 Output Delay distribution Delay distribution Delay distribution Figure 3: How the recommendation engine works. 3.0 60 93.2 93.2 94.0 85.0 85.8 85.8 93.2 85.8 88.8 86.5 86.5 86.5 82.8 82.8 55 92.5 88.0 87.3 87.3 89.5 91.7 87.3 88.0 84.3 88.8 91.0 85.0 88.0 2.5 89.5 83.5 90.2 90.2 90.2 83.5 84.3 91.0 82.8 89.5 87.3 91.0 91.7 89.5 88.8 92.5 93.2 90.2 88.0 91.7 82.8 93.2 81.3 94.7 2.0 84.3 85.0 86.5 94.0 86.5 92.5 86.5 87.3 91.088.8 86.5 88.8 89.5 82.0 91.7 84.3 82.0 91.0 90.2 83.5 91.7 92.5 88.8 85.0 89.5 88.0 85.8 89.5 88.8 88.0 88.8 1.5 88.0 85.8 87.3 87.3 92.5 85.0 91.7 91.0 88.0 85.8 92.5 25 93.2 85.0 85.8 94.0 91.7 90.2 94.7 85.8 95.5 82.0 96.2 92.5 84.3 85.0 20 97.0 93.2 1.0 0 50 100 150 200 2 4 6 8 10 12 14 16 Epoches Function RDSO-TS-RBFNN TS-RBFNN Figure 4: Summary of network observations and statistical results SO-TS-RBFNN PSO-TS-RBFNN of additional quality elements of digital music information platforms. Figure 5: Number of fuzzy rules in the training process. likely to lead to the disorder of fuzzy rule priority and thus affect the model’s recognition effect. database containing audio content and user’s historical In addition, most of the above-mentioned publicly behaviour by itself, and based on this data preprocessing, we available datasets are mainly English songs, while data on finally obtain a dataset that meets the needs of the system Chinese songs are rare. However, Chinese songs have their experiment. In the music recommendation system, the user characteristics, both in terms of melody and lyrics, especially and the music are two main parts, while the user and the from the audio perspective. Chinese pop songs have some music can be linked together by the matrix decomposition special ethnicinstrumentsandsingingstyles,whicharequite method. *e music dataset obtained in the previous section different from foreign language songs. Given the many contains the number of times each user played each piece of problems of the current music-related publicly available music, and the user’splayingbehaviour canbe seenas a form datasets mentioned above, this paper collects a music of implicit feedback. *is is because although the dataset Degree Number of rules 8 Computational Intelligence and Neuroscience records the number of times a user has listened to each song, the user does not explicitly rate each piece of music. However, it may be useful to assume that if a user loves a particular song, then that user will be more inclined to go listen to that song multiple times, viewing the user’s playing behaviour for the music as a potential rating. *e signifi- cance of this is that it avoids having the user display the ratingonthe onehand, andon theother hand,it allowsusers to change the rating data after listening to music, rather than only determining the rating by the age and singer of the 25 song. If a user has never listened to a song, there could be multiple reasons, such as the user may not like it or the user is not yet aware of it, as shown in Figure 6. 5 10 15 20 25 30 Applying the convolution layer, it is necessary to extract Serial number the features of the neighbouring data sets and feature them Figure 6: Data set output results. in each step, so a representative set of signals is obtained using a convolution kernel computed jointly in the hori- zontal and vertical directions. *e convolution is first per- they are unable to decide for themselves what digital music formed in two-step units according to the neighbourhood access behaviours to choose or when they do not know what and then in three-step units according to the neighbourhood digital music access behaviours are available to them. so that the features are obtained continuously. *e conti- Sometimes users seek recommendations from multiple nuity of CSI data is maintained as much as possible in gait people for different digital music information access be- haviours and choose one of them to use, as shown in recognition, which provides a guarantee for the accuracy of gait recognition. Taking a walk as an example, 30 packets of Figure 7. *e digital music information content and form of the data from three sets of antennas for one signal acquisition are taken and 300 signal acquisitions occur for each walk. platform has 25.83% of excitement type quality elements, A fusion then segmentation operation is performed. *e 25.59% of expectation type quality elements, 10.66% of fusion is a packet of three groups of antenna values for the essential type quality elements, 32.94% of undifferentiated average value calculation to get the average value. After the quality elements, 1.42% of reverse type quality elements, and 30 values as a row, walking time a total of 300 rows of data, 3.55% of suspicious results, so the quality elements are this is the fusion of the walking matrix. *en step-by-step undifferentiated type quality elements. segmentation is carried out to convolve the whole matrix Extracting features from digital music audio signals can according to two steps for the acquisition frequency, then not only effectively avoid the complex computation brought by directly processing the original audio data three steps for the acquisition frequency, and so on, after alternate convolution and pooling layers, a highly abstracted content but also obtain more distinct audio features for feature signal set F is obtained. subsequent processing and application. *ere are various audio feature analysis methods, among which the more common ones are acoustic spectrogram, Meier spectrum, 4.2. Experimental Results of Music Intelligence Marketing and Meier cestrum coefficient (MFCC). In recent years, as Strategy Optimization. *e influence of other people’s convolutional neural networks CNNs have shown very digital music information acquisition behaviours refers to powerful capabilities in image processing, the Mel Spec- users’ following or imitating others’ information acquisition trogram features of audio signals are increasingly widely behaviours in the process of digital music information ac- used in deep neural network models, even more than Mel quisition, which is the herding behaviour in digital music Cestrum Coefficients MFCC. When training a supervised deep learning model, the information acquisition. Herding behaviour is a behaviour in which individuals are influenced by the group and change original data set is usually divided into three parts: the their ideas to be in line with the majority, and it arises from training set, the validation set, and the test set. As the name both objective aspects and users’ factors. Crowd-following implies, the training set is the data set used to train the behaviour of digital music information access behaviour is model, which is used to determine the parameters of the manifested as access route crowding, access method network model, while the test set is the data set used to test crowding, and access tool for crowding. Most people around the model after it is trained, which is used to check the a user often use certain platforms to get digital music in- generalization ability of the network model. In the process of formation, so the user will also think that these platforms are model training, the validation set can be used to observe the fit of the model and stop the training in time when the model good to use and follow them; a user’s family or friends often use certain functions of digital music information platforms, appears to be overfitted. Also, the validation set can be used to determine some hyperparameters to assist in model so the user will also choose these functions for synchronizing with them or for easy contact. Following the advice of others optimization, such as the learning rate required for training in the choice of digital music information access behaviour. based on the convergence on the validation set. *e reason Often, users accept the recommendations of others when the aboveoperation isnotdone onthetrainingdatasetisthat Values Computational Intelligence and Neuroscience 9 13.0 10.4 7.8 5.2 2.6 Book1_C12 Book1_C13 Book1_C14 Book1_C15 Book1_C16 like very much As it should be It doesn't matter Can bear Do not like Name Figure 7: Model ratio matrix of digital music information content and form. as the network model is continuously trained, it is likely to 13.57% 7.56% cause the model to overfit the data on the training set, which 4.22% would be meaningless if the data on the training set is still 11.92% used to check and test the accuracy of the model in the end, as shown in Figure 8. 14.5% In summary, we find that both have their advantages and disadvantages in terms of recommendation performance in different recommendation tasks. Music preferences, the 7.94% higher the recommendation quality when N is larger, but in practice, the number of songs that users are most likely to 2.54% 4.89% browse is less than 40. 10.15% 8.5% By doing a cross-sectional comparison and systematic 1.92% validation of the algorithms proposed in this thesis, the 6.31% 5.98% recommendation accuracy and the explanation of reasons Physical Activity for the respective recommendation algorithms in different economic Inu fl ence recommendation tasks are derived. Recommendation Nature working mechanisms that consider integration include recom- hobby Music mending similar videos based on user’s historical behaviour Preference Regional browsing or comments and recommending popular videos Quality other based on video popularity. *e purpose is to recommend the Motivation video information of interest more accurately to the user and Figure 8: Optimization results. to better give the user the reason for the recommendation. *e system validated results show that the HARM algorithm is suitable for the next song recommendation task, which mainly reveals the session-specific preferences encoded in time required for a user to consume a song leads to a playlist the latest user interaction, focuses on the short-term music that usually contains not only one song but several songs in preferences of users, and distinguishes the difference in the listening session, focusing on users’ long-term music, preferences of different users for the same song from the filtering out behavioural noise data from users’ historical currently listened song sequence to improve the accuracy of listening lists, and then learning users’ long-term music recommending the next song; the ASR algorithm is suitable preferences to finally improve the quality of song recom- for the automatic playlist continuation task. Since the short mendations in automatic playlists. Type Values 10 Computational Intelligence and Neuroscience [5] J. Tang, G. Liu, and Q. Pan, “A review on representative 5. Conclusion swarm intelligence algorithms for solving optimization problems: applications and trends,” IEEE/CAA Journal of In digital music marketing, how to distinguish the differ- Automatica Sinica, vol. 8, no. 10, pp. 1627–1643, 2021. ences in music preferences among users is the key to im- [6] X. Zhou, X. Yang, J. Ma, and K. I.-K. Wang, “Energy efficient proving marketing efficiency, for which a recommendation smart routing based on link correlation mining for wireless algorithm based on multilayer attention representation is edge computing in IoT,” IEEE Internet of 8ings Journal, p.1, proposed. *e algorithm uses information such as user attributes and song content to learn the embedded repre- [7] K. Zatwarnickiand A. Zatwarnicka, “Anarchitecture of a two- sentations of songs from a multidimensional perspective and layer cloud-based web system using a fuzzy-neural request to mine the preference relationships between users and distribution,” Vietnam Journal of Computer Science, vol. 7, songs. It mainly solves the following problem, to mine the no. 3, pp. 251–262, 2020. differential preferences of different users for multidimen- [8] M.-H. Huang and R. T. Rust, “A strategic framework for sional features of the same song, an embedded represen- artificial intelligence in marketing,” Journal of the Academy of Marketing Science, vol. 49, no. 1, pp. 30–50, 2021. tation based on attention mechanism is proposed to learn [9] Q. Zhang, J. Lu, and Y. Jin, “Artificial intelligence in rec- song representations through user-based attention net- ommender systems,” Complex & Intelligent Systems, vol. 7, works, and then build song-based attention networks to no. 1, pp. 439–457, 2021. learn user preference representations based on the learned [10] Z. Cai and X. Zheng, “A private and efficient mechanism for song representations. To distinguish the degree of contri- data uploading in smart cyber-physical systems,” IEEE bution of different historical behaviours to users’ decisions, a Transactions on Network Science and Engineering, vol. 7, no. 2, temporal relationship recommendation algorithm based on pp. 766–775, 2020. the attention network is proposed to learn temporal de- [11] R. Mishra, N. Sharma, and H. Sharma, “Teaching learning- pendencies from listening behaviours and improve the ac- based optimisation algorithm: a survey,” International Journal curacy of song recommendations. Finally, the research of Swarm Intelligence, vol. 5, no. 2, pp. 153–181, 2020. results focus on mining users’ short-term music preferences, [12] W. K. Mashwani, A. Hamdi, M. Asif Jan, A. Gokta¸,s and F. Khan, “Large-scale global optimization based on hybrid which can be applied to recommendation tasks such as next swarm intelligence algorithm,” Journal of Intelligent & Fuzzy song similar songs. Systems, vol. 39, no. 1, pp. 1257–1275, 2020. [13] Z. Yan and Z. Lv, “*e influence of immersive virtual reality Data Availability systems on online social application,” Applied Sciences, vol.10, no. 15, p. 5058, 2020. *e data used to support the findings of this study are [14] M. Shehab, L. Abualigah, H. Al Hamad, H. Alabool, available upon request to the author. M. Alshinwan, and A. M. Khasawneh, “Moth–flame optimi- zation algorithm: variants and applications,” Neural Computing and Applications, vol. 32, no. 14, pp. 9859–9884, 2020. Conflicts of Interest [15] H. Hasanzadeh and J. Majidpour, “Application of fuzzy logic in order to plan and create software for converting sound to *e author declares that there are no conflicts of interest. text,” Bulletin of Electrical Engineering and Informatics, vol. 9, no. 3, pp. 1127–1136, 2020. [16] M. Jahanbakht, W. Xiang, L. Hanzo, and M. Rahimi Azghadi, Acknowledgments “Internet of underwater things and big marine data analy- tics–a comprehensive survey,” IEEE Communications Surveys *is work in this article was supported by Handan & Tutorials, vol. 23, no. 2, pp. 904–956, 2021. University. [17] M. D. Pawar and R. D. Kokate, “Convolution neural network based automatic speech emotion recognition using mel-fre- References quency cepstrum coefficients,” Multimedia Tools and Appli- cations, vol. 80, no. 10, pp. 15563–15587, 2021. [1] A. G. Hussien, M. Amin, and M. Abd El Aziz, “A compre- [18] R. Gao and H. Jiang, “Fuzzy control mathematical modelling hensive review of moth-flame optimisation: variants, hybrids, method based on dynamic particle swarm optimisation and applications,” Journal of Experimental & 8eoretical training,” International Journal of Innovative Computing and Artificial Intelligence, vol. 32, no. 4, pp. 705–725, 2020. Applications, vol. 12, no. 1, pp. 48–55, 2021. [2] T. Liu, Z. Yuan, L. Wu, and B. Badami, “An optimal brain [19] B. Yang, X. Li, Y. Hou et al., “Non-invasive (non-contact) tumor detection by convolutional neural network and en- measurements of human thermal physiology signals and hanced sparrow search algorithm,” Proceedings of the Insti- thermal comfort/discomfort poses-a review,” Energy and tution of Mechanical Engineers, Part H: Journal of Engineering Buildings, vol. 224, Article ID 110261, 2020. in Medicine, vol. 235, no. 4, pp. 459–469, 2021. [20] D. Molina, J. Poyatos, J. D. Ser, S. Garc´ıa, A. Hussain, and [3] N. Nedjah, L. D. M. Mourelle, and R. G. Morais, “Inspiration- F. Herrera, “Comprehensive taxonomies of nature- and bio- wise swarm intelligence meta-heuristics for continuous op- inspired optimization: inspiration versus algorithmic be- timisation: a survey-part I,” International Journal of Bio-In- havior, critical analysis recommendations,” Cognitive Com- spired Computation, vol. 15, no. 4, pp. 207–223, 2020. putation, vol. 12, no. 5, pp. 897–939, 2020. [4] V. A. Vuyyuru, G. A. Rao, and Y. V. S. Murthy, “A novel [21] S. Gupta, M. Kaur, S. Lakra, and Y. Dixit, “A comparative weather prediction model using a hybrid mechanism based on theoretical and empirical analysis of machine learning algo- MLP and VAE with fire-fly optimization algorithm,” Evolu- rithms,” Webology, vol. 17, no. 1, pp. 377–397, 2020. tionary Intelligence, vol. 14, no. 2, pp. 1173–1185, 2021.
http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png
Computational Intelligence and Neuroscience
Hindawi Publishing Corporation
http://www.deepdyve.com/lp/hindawi-publishing-corporation/variational-fuzzy-neural-network-algorithm-for-music-intelligence-94E5urGBwT