Access the full text.
Sign up today, get DeepDyve free for 14 days.
Kurt Braunlich, Carol Seger, K. Jentink, I. Buard, B. Kluger, M. Thaut (2018)
Rhythmic auditory cues shape neural network recruitment in Parkinson's disease during repetitive motor behaviorEuropean Journal of Neuroscience, 49
Bingying Li, Yongxin Li, Haitao Wang, Yuming Ma, Hu Qiang, Fang Ge (2018)
Compensation of automatic weighing error of belt weigher based on BP neural networkMeasurement
Bo Yao, H. Hagras, Mohammed Alhaddad, Daniyal Al-Ghazzawi (2015)
A fuzzy logic-based system for the automation of human behavior recognition using machine vision in intelligent environmentsSoft Computing, 19
Shariq Mohammed (2016)
Bankruptcy Prediction by Using the Altman Z-score Model in Oman: A Case Study of Raysut Cement Company SAOG and its subsidiariesThe Australasian Accounting Business and Finance Journal, 10
Okba Tibermacine, Chouki Tibermacine, Cherif Foudil (2018)
Estimating the reputation of newcomer web services using a regression-Based methodJ. Syst. Softw., 145
Sadia Lodhi, M. Manzar, R. Zahoor (2017)
Fractional neural network models for nonlinear Riccati systemsNeural Computing and Applications, 31
Wentao Wei, Yongkang Wong, Yu Du, Yu Hu, M. Kankanhalli, Wei-dong Geng (2017)
A multi-stream convolutional neural network for sEMG-based gesture recognition in muscle-computer interfacePattern Recognit. Lett., 119
(2017)
Blockchain technology evolution and industrial application prospect
Pei-Ju Wu, Kun Lin (2018)
Unstructured big data analytics for retrieving e-commerce logistics knowledgeTelematics Informatics, 35
Saeed Rouhani, Amir Ashrafi, Ahad Ravasan, Samira Afshari (2016)
The impact model of business intelligence on decision support and organizational benefitsJ. Enterp. Inf. Manag., 29
A. Goswami, G. Parashari, Ruchir Gupta (2018)
Evolutionary Stability of Reputation-Based Incentive Mechanisms in P2P SystemsIEEE Communications Letters, 22
T. Oliveira, Matilde Alhinho, P. Rita, G. Dhillon (2017)
Modelling and testing consumer trust dimensions in e-commerceComput. Hum. Behav., 71
(2016)
Analysis on the construction of cross-border E-commerce credit evaluation system under the background of big data
Weiqiang Zhu, G. Beroza (2018)
PhaseNet: A Deep-Neural-Network-Based Seismic Arrival Time Picking MethodGeophysical Journal International
He Fang, Li Xu, Xinyi Huang (2017)
Self-adaptive trust management based on game theory in fuzzy large-scale networksSoft Computing, 21
Kamil Bortko, Piotr Bartków, Jarosław Jankowski, D. Kuras, Piotr Sulikowski (2019)
Multi-criteria Evaluation of Recommending Interfaces towards Habituation Reduction and Limited Negative Impact on User Experience
Shiwen Shen, Simon Han, D. Aberle, A. Bui, William Hsu (2018)
An Interpretable Deep Hierarchical Semantic Convolutional Neural Network for Lung Nodule Malignancy ClassificationExpert systems with applications, 128
(2019)
Using blockchain technology to promote the development of cross border e-commerce in China
S. Akuma, R. Iqbal, Chrisina Jayne, F. Doctor (2016)
Comparative analysis of relevance feedback methods based on two user studiesComput. Hum. Behav., 60
Y. Zhang (2018)
Building China EU cross border e-commerce ecosystem with blockchain technology,China’s Circulation Economy, 2
Liang Liu, Xiaoyu Jiang, Jian Wu (2019)
Research on the Construction of China-Europe Cross-border E-commerce Ecosphere Based on Blockchain TechnologyJournal of Finance Research
Chuang Wei, Zhaoji Yu, Xiaonan Chen (2017)
Research on social e-commerce reputation formation and state-introduced modelKybernetes, 46
Blockchain transportation : a case study of freight logistics and its market governance
Bhakti Ahirwadkar, S. Deshmukh (2019)
Deep Neural Networks for Recommender SystemsInternational Journal of Innovative Technology and Exploring Engineering
Quanwang Wu, Qingsheng Zhu, Peng Li (2015)
A neural network based reputation bootstrapping approach for service selectionEnterprise Information Systems, 9
Piotr Sulikowski, Tomasz Zdziebko, Dominik Turzyński, E. Kańtoch (2018)
Human-website interaction monitoring in recommender systems
Tao Zhou, Kim-Han Thung, Xiaofeng Zhu, D. Shen (2018)
Effective feature learning and fusion of multimodality data using stage‐wise deep neural network for dementia diagnosisHuman Brain Mapping, 40
A. Singraber, J. Behler, C. Dellago (2019)
Library-Based LAMMPS Implementation of High-Dimensional Neural Network Potentials.Journal of chemical theory and computation, 15 3
Research on the application of blockchain in cross border payment
Yixue Hao, Mohd Usama, Jun Yang, M. Hossain, Ahmed Ghoneim (2019)
Recurrent convolutional neural network based multimodal disease risk predictionFuture Gener. Comput. Syst., 92
C. Fan, Jiayuan Wang, Wenjie Gang, Shenghan Li (2019)
Assessment of deep recurrent neural network-based strategies for short-term building energy predictionsApplied Energy
Hindawi Computational Intelligence and Neuroscience Volume 2022, Article ID 3088915, 10 pages https://doi.org/10.1155/2022/3088915 Research Article E-Commerce Credit Risk Assessment Based on Fuzzy Neural Network 1 2 Lina Wang and Hui Song School of Finacial Technology, Hebei Finance University, Baoding, Hebei 071051, China School of Information and Communication Engineering, Beijing University of Posts and Telecommunications, Haidian, Beijing 100876, China Correspondence should be addressed to Lina Wang; wanglina@hbfu.edu.cn Received 8 November 2021; Accepted 14 December 2021; Published 7 January 2022 Academic Editor: Akshi Kumar Copyright © 2022 Lina Wang and Hui Song. ,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 propose a cooperative strategy-based self-organization mechanism to reconstruct the network. ,e mechanism includes a comprehensive evaluation algorithm and structure adjustment mechanism. ,e self-organization mechanism can be carried out simultaneously with the parameter optimization process. By calculating the similarity and independent contribution of normative neurons, the effectiveness of fuzzy rules can be jointly evaluated, and effective structural changes can be realized. Moreover, this mechanism should not set the threshold in advance in practical application. In order to optimize the parameters of SC-IR2FNN, we developed a parameter optimization mechanism based on an interaction strategy. ,e parameter optimization mechanism based on a joint strategy, namely multilayer optimization engine, can split SC-IR2FNN parameters into nonlinear and linear parameters for joint optimization. ,e nonlinear parameters are optimized by an advanced two-level algorithm, and the linear parameters are updated with the minimum biological multiplication. Two parameter optimization algorithms optimize nonlinear and linear parameters, reduce the computational complexity of SC-IR2FNN, and improve the learning rate. Using the principal component factor analysis method, seven representative common factors are selected to replace the original variables, which include the profitability factor of the financing enterprise, the solvency factor of the financing enterprise, the profitability factor of the core enterprise, the operation guarantee factor, and the growth ability of the financing enterprise. Factors, supply chain online degree factors, financing enterprise quality, and cooperation factors, can well measure the credit risk of online supply chains. ,e logistic model shows that the profitability factor of the financing company, the debt repayment factor of the financing company, and the profitability of the core company are three factors that have a significant impact on the credit risk of online supply chain finance. Based on the improved credit calculation model, we developed an online clue risk calculation. ,is method is based on site conditions and can evaluate credit risk. From the test results, the improved credit scoring system is the result of facing speculative and circular credit fraud and implies that the traders of risk commentators are in a leading position in each electronic device. ,e results show that risk analysis is effective in any case. e-commerce credit risk assessment [2]. Fuzzy control is to 1. Introduction imitate people’s judgment and plot reasoning to deal with At present, artificial intelligence is developing rapidly, and problems that are difficult to solve by general methods, artificial neural network algorithm is the core of research [1]. especially nonlinear problems, problems that cannot be ,e combination of artificial neural networks and traditional modeled but require precision. ,e membership function industries is an effective way to solve traditional agricultural and rule design in fuzzy logic control are all artificially set, problems. ,e research purpose of this subject is to combine and in software design, the more the rules there are, the the fuzzy logic system and the artificial neural network into worse the real-time performance of the control operation the fuzzy neural network through research and use it in will be. ,e artificial neural network has good self-adaptive 2 Computational Intelligence and Neuroscience ,e scoring of “good reviews” and “bad reviews” changes learning ability. It mainly works by simulating the biological neural network, but the disadvantage is that its ability to dynamically with the progress of transactions. ,is method effectively suppresses periodic fraud by increasing the op- express rules is poor. ,erefore, the fuzzy logic system and neural network are combined to make full use of the ad- portunity cost of users with high credit values. Using the vantages of both, make up for the shortcomings of both, and range occupancy rate of commodity prices to score effec- form a fuzzy neural network algorithm [3, 4]. tively suppresses credit speculation and protects the interests With the development of the e-commerce industry, its of sellers who only sell low-priced commodities. ,e ex- risk issues have become increasingly prominent [5]. Credit perimental results show that the scoring method is effective risk has become an important reason hindering the devel- and feasible. opment of the industry. Various disputes and complaints caused by credit risk issues reduce consumer trust and repeat 2. Related Work purchase rates, increase customer acquisition costs and user transaction costs of e-commerce platforms, and hinder the Relevant scholars improved the genetic algorithm and then further development of e-commerce [6]. According to the used the improved algorithm to optimize the weights of the investigation of the Internet fraud observation website, in feedforward network so that the training accuracy of the areas with mature e-commerce markets such as the United algorithm was significantly improved [11]. Researchers States, the amount of money lost by traders due to seller combined the traditional BP algorithm and the differential credit risk is also increasing year by year; while China’s evolution algorithm to propose a new weight training market system is imperfect, legal and credit systems are method and used it in breast cancer prediction experiments imperfect. ,e traders are more likely to suffer losses because and achieved good results [12]. Related scholars combined of this. Relevant scholars believe that the issue of credit risk the differential evolution algorithm with EANN to construct has become an industry and social problem that needs to be the MPANN network model, which improved the function resolved urgently [7]. ,e influencing factors of e-commerce approximation ability [13]. credit risk issues include technical reasons not only caused Relevant scholars introduced mutation operators on the by the separation of time and space on the Internet but also basis of the PSO algorithm and then intelligently fused with related to whether the management of the virtual market is wavelet neural network, diagnosed transformer faults, and perfect and the soundness of the legal system, and the most obtained good experimental results [14]. Relevant scholars important thing is the credit choice of the transaction subject optimized the convolutional neural network and proposed a [8]. ,e information asymmetry caused by the virtual nature convolutional neural network acceleration algorithm based of the network has aggravated the inequality of information on parameter and feature redundancy compression, which between the parties to the transaction. From the perspective was used for image processing and proved the optimization of the transaction characteristics of e-commerce, the performance of the algorithm [15]. Neural network tech- e-commerce platform has large user traffic and low sticki- nology has gradually developed and been used in fault ness, which leads to short-term interests driven by sellers detection, prediction, and classification. and triggers their actions [9, 10]. It realizes the processing and expression of the ambiguity ,e interval class II fuzzy neural network structure asks in data through membership functions and fuzzy neurons. the problems of solution difficulty and computational Relevant scholars combine evolutionary computing with complexity to design the interval class II structure. ,e neural networks to propose multiple groups of classification interval second kind of fuzzy Shinto network is constructed algorithms to effectively adopt the membership functions of in an organized structure, and the computational complexity the fuzzer and defuzzer to the data set and use actual eco- is low. B2C electronics led event supply finance credit risk nomic data to successfully test [16]. Related scholars have crunch creates a comment price model and defines a proposed a new online sequential learning evolution RL comment price index by analyzing the models attributed to neuro-fuzzy model design and developed a dynamic evolu- factor analysis and logistic back. When calculating the risk of tionary fuzzy neural network (DENFIS) function approaching a transaction, according to historical transaction records, the the RL system method [17]. Relevant scholars apply a fuzzy risk calculation is divided into three categories. When cal- neural network to environmental safety assessment, using culating the risk, the commodity price, seller credit value, fuzzy neural networks to deal with the characteristics of fuzzy seller credit rating, historical average price, and transaction phenomena, and have achieved good results [18, 19]. failure threshold are considered. ,e calculation of the risk Relevant scholars use weighted Euclidean distance to of cycle deception also takes into account the important improve traditional clustering, use rough sets to reduce factor of the deception cycle. At the same time, this article attributes, build a neural network based on the clustering also gives the algorithm idea of seeking the deception cycle. results, and establish a new prediction model and have ,rough experiments on periodic fraud and credit specu- achieved good results in wind speed prediction [20]. ,e lation transactions, we can see the impact of commodity researchers gave a theoretical overview of the fuzzy nervous prices, credit values, credit ratings, and transaction failure system, discussed related knowledge, introduced two net- rates on the value of risk. ,e graphic image drawn by work models, and proved the performance of the model MATLAB illustrates the historical transaction process. ,e through simulation experiments [21]. Relevant scholars credit scoring method proposed in this article closely links to proposed that the data can be input in batches during the credit and risk and divides commodity prices into ranges. reasoning process, and the data generated by the overall rules Computational Intelligence and Neuroscience 3 can be divided, thereby reducing the number of rules and identification and control of complex nonlinear systems completing the algorithm optimization of the fuzzy neural with uncertainties and time-varying properties. Different network, which makes it more advantageous in processing from a type 1 fuzzy neural network, IT2FNN adopts an high-dimensional data [22]. Relevant scholars analyzed bio- interval type 2 membership function to convert the exact logical principles and proposed the method of using gene value into an interval fuzzy set so that it can better deal with overlap to optimize the fuzzy neural network, generating T-S the uncertainty in the system. In addition, IT2FNN avoids fuzzy rules through genetic code shifting, combining genetic the use of primary and secondary membership functions in algorithm with fuzzy control, realizing genetic mutation, and the type 2 fuzzy neural network to cause an overly com- improving the performance of the algorithm [23]. plicated calculation process so that IT2FNN can be better Relevant scholars have incorporated average purchase applied in practical engineering. prices, transaction density, and historical transaction rec- ,e structure of IT2FNN is shown in Figure 1. Its ords into the evaluation index system to comprehensively structure specifically includes five layers of neurons, which reflect the credit status of sellers in the transaction process are the input layer, subordinate layer, rule layer, subsequent [24]. Researchers and others introduce commodity prices layer, and output layer. and the credit of the evaluation subject into the credit ,e output of IT2FNN is as follows: evaluation model to identify whether there are false eval- ′ ″ uations in the transaction process between buyers and y(t) � y (t)q(t + 1) − y (t)[1 − q(t + 1)], sellers, thereby reducing the uncertainty of credit evaluation M−1 results [25]. Relevant scholars have proposed that an ef- f (t)h (t + 1) j�0 j j y (t) � , fective way to distinguish honest sellers and prevent dis- M−1 f (t) j�0 j honest sellers from trading is to establish an effective credit evaluation mechanism [26]. M−1 (1) f (t)h (t + 1) On the basis of previous studies, relevant scholars have j�0 j j y (t) � , M−1 investigated the historical transaction volume of sellers and f (t + 1) j�0 j the credit level of evaluation subjects and objects into the credit evaluation model [27]. Relevant scholars pointed out n−1 that the current credit evaluation model indicators cannot h (t) � w (t + 1)x (t) + b (t). j ij i+1 j provide sufficient differentiation [28]. ,erefore, indicators i�0 such as the historical credit of buyers and sellers and the ′ ″ where y (t) and y (t) are the lower and upper output number and amount of transactions should be increased to bounds of the consequent layer at time t, q(t) is the scale more accurately reflect the user’s credit. factor, h (t) is the consequence of the j-th fuzzy rule, w (t) is j ij Researchers propose an online credit evaluation method the consequent weight of the i-th input corresponding to the that measures the similarity between new transactions and j-th regular neuron, and b (t) is the j-th deviation. In ad- past transactions in the dimensions of product type, number dition, n represents the number of input neurons in the of sold products, and transaction amount, thereby estab- input layer; M represents the number of regular neurons in lishing a multidimensional credit evaluation index system the regular layer; and f (t) and f (t) are the lower and upper [29]. From the perspective of evaluation semantics, relevant bounds of the activation intensity of regular neurons, re- scholars have proposed an online credit evaluation system spectively. It can be expressed as follows: based on context, which aims to assist consumers in mea- suring the credibility of sellers and to screen whether n−1 consumers’ evaluation opinions are of reference [30]. ′ ′ f (t) � u (t), j ij ,e topology of the neural network has a greater impact i�0 (2) on its performance and calculation speed. Compared with n−1 the general neural network, the IT2FNN structure is more ″ ″ f (t) � u (t + 1). j ij complex and contains more parameters and faces a greater i�0 computational burden in practical applications. In general, more neurons can ensure that the neural network has better performance, but too many neurons will make the calcu- lation of the neural network too complicated, which is not 3.2. Design of Self-Organization Mechanism of Self-Con- conducive to practical applications. In addition, fewer structed Interval Type 2 Fuzzy Neural Network. In order to neurons will reduce the performance of the neural network. improve the performance of IT2FNN, the article proposes a self-constructed interval type 2 fuzzy neural network based ,erefore, how to determine the appropriate IT2FNN structure has always been the focus of research. on collaborative strategy, including a self-organization mechanism based on collaborative strategy and a param- eter optimization mechanism based on collaborative 3. Method strategy. ,e self-organization mechanism uses a com- 3.1. Interval Type 2 Fuzzy Neural Network. IT2FNN not only prehensive evaluation algorithm and a structure adjust- ment mechanism to make the structure adjustment of the has excellent uncertainty processing capabilities but also has adaptive learning capabilities, so it can well realize the self-constructed interval type 2 fuzzy neural network 4 Computational Intelligence and Neuroscience Interval Type 2 Rule layer Membership Function Post-layer h (t) Uncertainty handling capacity u (t) X (t) q (t) .. y (t) =y (t)q (t+1)–y (t)[tq (t+1)] X (t) 1-q (t) Adaptive learning ability y (t) q (t) X (t) Output layer Input layer 1-q (t) u (t) h (t) nM M n–1 h (t) = ∏ [w (t+1)x (t)]+b (t) j ij i+1 j i = 0 Membership layer p Figure 1: Interval type 2 fuzzy neural network structure. coordinate with the parameter optimization process. ,e making self-organizing judgments in the learning process. comprehensive evaluation algorithm uses the interneuron ,is feature is conducive to practical applications. In IT2FNN, the activation intensity of regular neurons and interlayer information to comprehensively evaluate the structure of SC-IT2FNN and then uses the structure ad- reflects the ability of fuzzy rules. Effective or redundant fuzzy justment mechanism to add and delete fuzzy rules to realize rules can be found so that SC-IT2FNN has a suitable net- the structural self-organization of SC-IT2FNN. In addition, work structure. In the comprehensive evaluation algorithm, SC-IT2FNN does not need to preset any threshold when the similarity is as follows: Z−1 ′ ″ F (z − t − 1) + F (z + t + 1)F (z + 2t − 1) − F (z − t + 2) z�0 j i S (t) � , ij 1/2 1/2 2 2 ′ ″ F (z − t − 1) − F (z + t + 1) · F (z + 2t − 1) + F (z − t + 2) j i ′ ″ F (t + z − 1) � 0.5 f (t − 2z − 1) − f (2t + z − 1) , j j j M−1 t + 2z F (t − 2z − 1) � F , M − 1 j�0 (3) S (t) ij C (t) � , d (t) 1/2 −1 d (t) � F (t + 1) − Y(t − 1)V(t) F (t + 1) − Y(t − 1) , j j j −1 F (t − 1) F (t) . . . F (t − 1 + Z) F (t) � , j j j −1 Y (t) � y(t − 1) y(t) . . . y(t − 1 + Z) . ,e comprehensive evaluation algorithm uses the sim- indicate the necessity of each fuzzy rule, and independent ilarity and independent contribution of rule neurons to contribution degree can indicate the effectiveness of each evaluate the effectiveness of fuzzy rules. Similarity can fuzzy rule. ,rough these two evaluation indicators, it can be ... ... ... ... ... ... ... ... ... ... ... ... ... ... Computational Intelligence and Neuroscience 5 judged whether each fuzzy rule in SC-IT2FNN is redundant process is optimal for construction and general parameter and effective. At the same time, when calculating the sim- modification, as well as nonlinear and general linear pa- ilarity and independent contribution of fuzzy rules, the state rameters. ,e improved quadratic algorithm and least value of the rule neuron is used to ensure the accuracy of the square algorithm are used in the structure transformation similarity and independent contribution. process. ,e SC-IR2FNN network structure is used to adjust In the design of this coordination mechanism, the and total parameters. We downloaded it based on sales connection of fuzzy weights and the independent trans- connect policy SC-IR2FNN and learned about the real mission of degrees are three structural transformation states, stacking steps of processing. By designing a self-adjusting which correspond to three stages of adjustment. Structure learning rate, the difficulty of choosing a learning rate is diagram of adjustment mechanism is shown in Figure 2. solved. For SC-IT2FNN, the layered optimization mecha- nism can effectively improve accuracy and reduce compu- tational complexity. Using multiple sets of sample data to 3.3. Design of Parameter Optimization Mechanism for Self- calculate the similarity and independent contribution of the Constructed Interval Type 2 Fuzzy Neural Network. ruled neurons not only avoids frequent adjustments to the Traditional optimization algorithms such as back- network structure that may cause the network to fail to propagation algorithm, gradient descent algorithm, Newton converge but also ensures the accuracy of the network method, and Levenberg–Marquardt algorithm are widely structure evaluation and reduces the computational burden. used in the parameter optimization of neural networks, but ,erefore, SC-IT2FNN has the characteristics of good these algorithms have the problems of slow convergence generalization performance and fast calculation speed. ,e speed and difficulty in obtaining optimal solutions, such as specific implementation process of the learning process of backpropagation algorithm and gradient descent algorithm. SC-IT2FNN is based on collaborative strategy as shown in ,e calculation process of Newton’s method and Lev- Figure 3. enberg–Marquardt algorithm is complicated and time- consuming. ,ere are many restrictions in the use process. 4. Result Analysis Due to the numerous parameters of the interval type 2 fuzzy neural network, more calculation time is required in the 4.1. Logistic Model and Data Selection. In terms of supply process of optimizing the parameters. chain financial risk management, foreign countries have In order to optimize the parameters of SC-IT2FNN, the developed a variety of methods to measure supply financial article proposes a parameter optimization mechanism based risk. Traditional methods include expert scoring method, on a collaborative strategy, that is, a hierarchical optimi- fuzzy comprehensive evaluation method, credit rating zation mechanism. In this hierarchical optimization method, BP neural network, and logistic model. ,e logistic mechanism, the parameters are divided into nonlinear and model has the following advantages: linear parameters for collaborative optimization. Nonlinear First, the logistic model has relatively simple require- parameters are updated using an improved second-order ments for data collection and processing and has strong algorithm, and linear parameters are updated using a least- operability; second, the model predicts the probability that squares algorithm. ,is layered optimization mechanism the result is between 0 and 1, which can intuitively see the can quickly update the SC-IT2FNN parameters, effectively credit risk of the financing enterprise; ,ird, the model’s improve learning accuracy, and reduce computational preconditions are relatively loose and can be applied to complexity. ,e hierarchical optimization mechanism in- continuous or categorical independent variables. ,erefore, cludes two parts: parameter analysis and parameter opti- this paper finally chooses the logistic model to evaluate the mization. In the parameter analysis, the SC-IT2FNN credit risk of B2C e-commerce online supply chain finance. parameters including uncertain mean, standard deviation, Suppose the conditional probability of credit risk in scale factor, subsequent weights, and deviations will be financing enterprises is p. When the value of p is closer to 1, divided into two types: nonlinear parameters and linear it indicates that the credit status of the enterprise is better; parameters. ,e following parameters are nonlinear otherwise, the credit is not good. ,e logistic model does not parameters: theoretically have a critical value, and 0.5 is used as the critical value during model analysis. ,erefore, this paper σ (t − 1) q(t) m (t + 1) Φ (t) � . (4) ij ij also uses 0.5 as the critical value when studying corporate credit risk. If the calculated value is greater than or equal to where Φ (t) is a nonlinear parameter set containing 0.5, the company’s credit is considered good; otherwise, the three nonlinear parameters. ,erefore, SC-IT2FNN can be company’s credit is bad. expressed as follows: ,is article takes an e-commerce company’s participa- Y(t) � Φ (t + 1)E(t) + Ψx(t),Φ (t), L N tion in online supply chain finance as an example, applies the (5) evaluation index system established above, comprehensively w (t − 1) b (t) Φ (t) � . ij j evaluates the company’s credit status based on the online ,ere are two parts: updating nonlinear parameters supply chain financing model, and compares and analyzes using a modified second algorithm and optimizing linear the results. ,e example quantitatively analyzes the effec- parameters using at least two algorithms. But the usual tiveness of the logistic model on the financial credit risk parameter normalizer is the best. In addition, SC-IR2FNN control of the B2C e-commerce online supply chain. 6 Computational Intelligence and Neuroscience Determine whether each fuzzy ere is no need to preset any rule in SC-IT2FNN is redundant threshold when adjusting the structure of SC-IT2FNN and effective Delete the corresponding Add a new Indicate the necessity of redundant fuzzy rules fuzzy rule each fuzzy rule Show the validity of each fuzzy rule Exclude Growth stage Evaluation Structural Adjustment conditions index adjustment condition Independent contribution Growth Other conditions Cut-off stage Stable phase conditions Similarity e number of fuzzy rules will not change Structural adjustment conditions ree structural adjustment conditions ree adjustment stages Structural adjustment mechanism Figure 2: Structural adjustment mechanism. Start Determine the number of input and output variables Uncertain mean End Construct the initial structure of Standard SC-IT2FNN deviation Stop calculation e parameters are Scale factor initialized Yes Subsequent weight Use SC-IT2FNN to Does it meet the No calculate network output set accuracy or the Deviation based on input data number of iteration steps? Self-adjusting learning rate Update nonlinear parameters using improved second-order Number of fuzzy algorithm rules Update linear parameters according to update rules Total number of samples Calculate the vector and Is the number of Linear matrix in the least squares No input samples an optimization algorithm integer multiple of process sample Z? size Generate new fuzzy rules Sample size Yes during structural adjustment Calculate the similarity and Initialize new fuzzy rule independent contribution of all parameters regular neurons Figure 3: ,e specific implementation process of the learning process of SC-IT2FNN based on collaborative strategy. Computational Intelligence and Neuroscience 7 E-commerce is one of the previous generation B2C 0.65 online shopping platforms. ,ere are many kinds of 0.64 household appliances, such as traditional household appli- ances, 3C household appliances, and daily necessities. 0.63 Online finance and e-commerce is one of the main direc- 0.62 tions. ,e supply of financial consumption platoon, in- 0.61 0.005 vestment asset management, tips, ointment, prepaid cards, 0.004 0.003 0.002 690 preleasing, private equity financing, and other businesses 0.001 0 650 have been launched. Its attempts in the internet field have Sig. for Bartlett sphericity test Approximate chi−square further expanded the influence of an e-commerce company Figure 4: Bartlett and KMO tests. and realized collaboration and cooperation with upstream and downstream enterprises for a win-win situation. ,e supply chain finance business of an e-commerce 1.75 company is developed on the basis of its one-stop supply 1.5 chain service platform. Customers can realize the transac- 1.25 tion and financing process on the platform. However, as the business layout expands, an e-commerce company has also encountered applications. In order to help an e-commerce 0.75 company better cope with the credit risk of small- and medium-sized enterprises in online supply chain finance, 0.5 this article selected 40 e-commerce companies on an 0.25 e-commerce platform and upstream and downstream 1 2 3 4 5 67 companies in the supply chain from the Wonder database as 7 principal components samples. We take the lower value of the interest-bearing debt Sig. B ratio of each industry in 2020 as the limit. Among the df Wald existing 40 sample companies, 6 companies have breached S.E the contract, and 34 companies have not breached the contract. Figure 5: Logistic regression results. and their Sig. values are all less than 0.5, indicating that these 4.2. Factor Analysis. When using the logistic model to model the data, it is required that there should be no collinearity seven variables are very convincing for predicting the credit between the independent variables. ,erefore, we first use risk of financing enterprises. factor analysis to standardize the variables and select rep- resentative independent variables to replace the original 4.4. Results and Verification of Model Analysis. ,e seven indicator variables. ,e newly acquired common factors are principal component factors calculated by using SPSS linear combinations of the original variables. software are used as independent variables of the model, and First, we need to check tapinle’s data with Bartlett to see two types of enterprises (the value of enterprises with credit if the indicators selected by individuals are factor analysis. If risk is 1, and the value of enterprises without credit risk is 2) KMO is 0.9 or greater, factor analysis is considered to be are selected as dependent variables. ,e sample results are used. If it is 0.8–0.9, it is open. 0.6–0.8 spacing is normal. At shown in Table 1. one level, we then do factor analysis. ,e results of the As shown in Table 1, we can see that in these 40 samples, Bartlett and KMO tests are shown in Figure 4. among the 29 risk-free companies that have been observed, ,e KMO test coefficient in this article is greater than 29 risk-free companies have been predicted using the model, 0.5, and the partial correlation between variables is strong. with an accuracy rate of 100%. Among the 6 observed high- ,e Sig. of the Bartlett sphere test is less than 1%, indicating risk companies, 5 were predicted by the model to be risky that we can do factor analysis. companies, with an accuracy rate of 87%. ,erefore, the final comprehensive accuracy rate of this model reached 89%. 4.3. Logistic Regression Model Analysis. We have to substi- ,is shows that the prediction accuracy of the model is high. tute the logistic regression model for the seven principal Table 2 shows the results of the significance test using SPSS. components. ,e sample companies are used as dependent As shown in Table 2, the Sig. value of the model test is variables, and the seven principal component factors are 0.021, which is significantly lower than 0.05, indicating that used as independent variables. ,e factors F1∼F7 are the logistic regression equation obtained is significant at the substituted into the logistic model for regression analysis 95% level. ,is illustrates that the seven factors, namely, the using the entry method, that is, all variables are substituted profitability factor of the financing enterprise, the solvency factor of the financing enterprise, the profitability factor of into the equation at one time. ,e results of using SPSS software are shown in Figure 5. the core enterprise, the operation guarantee factor, the It can be seen from the regression results that the final growth ability factor of the financing enterprise, the online explanatory variables of the seven models are all retained, degree factor of the supply chain, and the quality of the Logistic regression value KMO test coefficient 8 Computational Intelligence and Neuroscience Table 1: Classification of inspection samples Predicted Observed E-commerce credit risk 1 2 1 6 0 87.01 E-commerce credit risk 2 5 29 93.40 Overall percentage 89.25 Table 2: Comprehensive test of model coefficients. Bangla Sig. df 14.2 0.021 6 13.9 0.021 7 15.1 0.021 5 15.2 0.021 7 100 14.7 0.021 6 13.8 0.021 5 15.4 0.021 7 0 50 100 150 200 250 300 350 400 450 500 Number of transactions financing enterprise and cooperation factors, have a sig- RBF neural network nificant relationship with the credit risk of the enterprise and Fuzzy neural network further illustrates the practical application value of the Figure 6: Comparison of the credit value growth process of cyclic model. deception. 4.5. Anticycle Deception Analysis. Here, we select a set of experimental data with this feature to compare and analyze the difference between the credit scoring method proposed in this article and Taobao’s credit scoring method. ,e experimental results are shown in Figure 6. It can be seen from Figure 6 that the scoring method using RBF neural network credit accumulation is completely powerless against periodic deception. ,e credit scoring method of a fuzzy 250 200 8 150 6 neural network describes the process of cyclical deception in 50 2 a very specific way and inhibits the growth of credit value, Number of transactions Number of traders which is conducive to the risk judgment of buyers and can (a) also play a supervisory role for sellers. Within the transaction failure threshold, the credit value will not be degraded, and when the threshold is exceeded, the credit value will be degraded. ,erefore, the setting of the threshold of trans- action failure rate is critical. If the number of negative re- views specified by the transaction failure rate threshold is not reached, the credit value begins to degrade, indicating that the threshold is set too large and the actual transaction 5 50 2 failure rate is small. Conversely, if the number of times Number of traders Number of transactions specified by the threshold is reached and the credit value has (b) not been degraded, it means that the threshold setting is too small, and the actual failure rate is relatively large. Figure 7: ,e credit value of the cycle deception after the credit hype: (a) results of RBF neural network method and (b) the results of the fuzzy neural network method. 4.6. Analysis of Anticredit Hype. We select realistic data that meet this situation for analysis, and the experimental results are shown in Figure 7. that the credit scoring method of the RBF neural network Under Taobao’s credit scoring method, although the cannot curb fraudulent behavior after credit speculation. seller did not cheat after the credit speculation, the credit at And through the fuzzy neural network, it can be clearly seen this time has been greatly reduced, and the buyer is easily that the deceptive credit value after the credit hype has been confused by the credit value. It can be seen from Figure 7 reduced. Credit value Credit value Credit value Computational Intelligence and Neuroscience 9 experiment shows that the self-constructed interval type 2 5. Conclusion fuzzy neural network is effective in punishing the periodic From the empirical results, it can be seen that indicators deception. such as net sales interest rate, return on net assets, and net profit rate of total assets are positively correlated with the Data Availability profitability factor of financing enterprises. At the same time, the profitability factor of financing enterprises is ,e data used to support the findings of this study are positively correlated with the compliance probability of available from the corresponding author upon request. online supply chain finance SMEs. ,erefore, the higher the net profit margin of the financing company’s sales and the Conflicts of Interest net profit margin of the financing company’s total assets, the lower the credit risk of SMEs participating in online supply ,e authors declare that there are no conflicts of interest. chain financing. ,e solvency of financing companies is negatively related to the probability of online credit risk. Acknowledgments ,at is, the stronger the solvency of SMEs, the smaller the credit risk in their online supply chain financing business. ,is work was supported by Hebei Finance University. From the empirical results, we can see that the current ratio, quick ratio, and interest coverage are positively correlated References with solvency, and the debt-to-asset ratio is negatively [1] C. Wei, Z. J. Yu, and X. N. Chen, “Research on social correlated with solvency. ,erefore, the higher the current e-commerce reputation formation and state-introduced ratio, quick ratio, and interest protection multiple, the model,” Kybernetes, vol. 46, no. 6, pp. 1021–1038, 2017. stronger the solvency of SMEs, and the lower the credit risk [2] P. J. Wu and K. C. Lin, “Unstructured big data analytics for of their participation in online supply chain financing; the retrieving e-commerce logistics knowledge,” Telematics and higher the debt-to-asset ratio, the lower their solvency. ,e Informatics, vol. 35, no. 1, pp. 237–244, 2018. online degree factor is negatively related to the probability of [3] Y. Zhang, “Building China EU cross border e-commerce the occurrence of financial credit risk in the online supply ecosystem with blockchain technology,” China’s Circulation chain. ,at is to say, the stronger the online degree of this Economy, vol. 2, no. 2, pp. 66–72, 2018. business, the smaller the credit risk of this online supply [4] Y. Hao, M. Usama, J. Yang, M. S. Hossain, Q. Liu, and chain financing business. It can be seen from the com- A. Ghoneim, “Recurrent convolutional neural network based multimodal disease risk prediction,” Future Generation ponent score coefficient matrix that the indicators Computer Systems, vol. 92, no. 1, pp. 76–83, 2019. reflecting the online degree of the supply chain, such as the [5] P. Sulikowski, T. Zdziebko, D. Turzynski, ´ and E. Kantoch, ´ electronic order processing capacity and the degree of “Human-website interaction monitoring in recommender information sharing, are positively correlated with the systems,” Procedia Computer Science, vol. 126, pp. 1587–1596, online degree factor of supply chain finance, and the online degree factor is related to the compliance of small- and [6] N. Dong and X. Zhu, “Blockchain technology evolution and medium-sized enterprises. ,e probability is positively industrial application prospect,” Information Security Re- correlated. ,erefore, the higher the electronic order search, no. 3, pp. 200–210, 2017. processing capability and the degree of information [7] S. Mohammed, “Bankruptcy prediction by using the altman sharing, the lower the credit risk of SMEs participating in Z-score model in Oman: a case study of raysut cement online financing business. When dealing with the problem company SAOG and its subsidiaries,” Australasian Ac- counting Business and Finance Journal, vol. 10, no. 4, of using low-priced commodities for credit speculation, the pp. 70–80, 2016. improved credit scoring method considers prices in dif- [8] C. Fan, J. Wang, W. Gang, and S. Li, “Assessment of deep ferent segments, and the increase in credit value is related recurrent neural network-based strategies for short-term to the ratio of price segments in all price segments. ,e building energy predictions,” Applied Energy, vol. 236, closer the price range of the goods sold, the faster the credit pp. 700–710, 2019. growth of sellers. Once the price range of the goods sold by [9] Q. Wu, Q. Zhu, and P. Li, “A neural network based reputation the seller fluctuates sharply, the growth of the credit value bootstrapping approach for service selection,” Enterprise slows down. In this way, credit hype can be curbed while Information Systems, vol. 9, no. 7, pp. 768–784, 2015. protecting the interests of sellers who only sell low-priced [10] K. Braunlich, C. A. Seger, K. G. Jentink, I. Buard, B. M. Kluger, goods. Comparative experiments show that the self-con- and M. H. ,aut, “Rhythmic auditory cues shape neural network recruitment in parkinson’s disease during repetitive structed interval type 2 fuzzy neural network is effective in motor behavior,” European Journal of Neuroscience, vol. 49, suppressing credit speculation. When dealing with the no. 6, pp. 849–858, 2019. problem of periodic fraud, the improved credit scoring [11] H. Li, “Using blockchain technology to promote the devel- method sets different credit levels according to the level of opment of cross border e-commerce in China,” Monthly credit value. In different credit levels, different deduction Finance and Accounting, no. 3, pp. 66–146, 2019. coefficients are set. A high credit rating sets a high de- [12] Z. F. Liu, “Analysis on the construction of cross-border duction factor to increase the opportunity cost of periodic E-commerce credit evaluation system under the background deception. ,rough this kind of punishment, the cycle of of big data,” China Business and Market, vol. 30, no. 6, deception is effectively curbed. ,e comparative pp. 58–64, 2016. 10 Computational Intelligence and Neuroscience [13] S. Shen, S. X. Han, D. R. Aberle, A. A. Bui, and W. Hsu, “An [30] B. Li, Y. Li, H. Wang, Y. Ma, Q. Hu, and F. Ge, “Compensation interpretable deep hierarchical semantic convolutional neural of automatic weighing error of belt weigher based on BP network for lung nodule malignancy classification,” Expert neural network,” Measurement, vol. 129, pp. 625–632, 2018. Systems with Applications, vol. 128, pp. 84–95, 2019. [14] O. Tibermacine, C. Tibermacine, and F. Cherif, “Estimating the reputation of newcomer web services using a regression- based method,” Journal of Systems and Software, vol. 145, pp. 112–124, 2018. [15] K. Bortko, P. Bartkow, ´ J. Jankowski, D. Kuras, and P. Sulikowski, “Multi-criteria evaluation of Recommending Interfaces towards habituation reduction and limited negative impact on user experience,” Procedia Computer Science, vol. 159, pp. 2240–2248, 2019. [16] Z. Lu and X. Ge, “Research on the application of blockchain in cross border payment,” Southwest Finance, no. 2, pp. 23–28, [17] T. Zhou, K. H. ,ung, X. Zhu, and D. Shen, “Effective feature learning and fusion of multimodality data using stage-wise deep neural network for dementia diagnosis,” Human Brain Mapping, vol. 40, no. 3, pp. 1001–1016, 2019. [18] B. Yao, H. Hagras, M. J. Alhaddad, and D. Alghazzawi, “A fuzzy logic-based system for the automation of human be- havior recognition using machine vision in intelligent envi- ronments,” Soft Computing, vol. 19, no. 2, pp. 499–506, 2015. [19] B. Ahirwadkar and S. N. Deshmukh, “Deep neural networks for recommender systems,” International Journal of Inno- vative Technology and Exploring Engineering, vol. 8, no. 12, pp. 4838–4842, 2019. [20] Y. Tian, G. Zhao, and L. Shen, “Blockchain transportation: a case study of freight logistics and its market governance,” China’s Circulation Economy, no. 2, pp. 50–56, 2018. [21] W. Zhu and G. C. Beroza, “PhaseNet: a deep-neural-network- based seismic arrival-time picking method,” Geophysical Journal International, vol. 216, no. 1, pp. 261–273, 2019. [22] A. Goswami, G. S. Parashari, and R. Gupta, “Evolutionary stability of reputation-based incentive mechanisms in P2P systems,” IEEE Communication Letter, vol. 22, no. 2, pp. 268–271, 2017. [23] S. Akuma, R. Iqbal, C. Jayne, and F. Doctor, “Comparative analysis of relevance feedback methods based on two user studies,” Computers in Human Behavior, vol. 60, pp. 138–146, [24] W. Wei, Y. Wong, Y. Du, Y. Hu, M. Kankanhalli, and W. Geng, “A multi-stream convolutional neural network for sEMG-based gesture recognition in muscle-computer inter- face,” Pattern Recognition Letters, vol. 119, pp. 131–138, 2019. [25] T. Oliveira, M. Alhinho, P. Rita, and G. Dhillon, “Modelling and testing consumer trust dimensions in e-commerce,” Computers in Human Behavior, vol. 71, pp. 153–164, 2017. [26] S. Rouhani, A. Ashrafi, A. Zare Ravasan, and S. Afshari, “,e impact model of business intelligence on decision support and organizational benefits,” Journal of Enterprise Information Management, vol. 29, no. 1, pp. 19–50, 2016. [27] S. Lodhi, M. A. Manzar, and M. A. Z. Raja, “Fractional neural network models for nonlinear riccati systems,” Neural Computing and Applications, vol. 31, no. 1, pp. 359–378, 2019. [28] H. Fang, L. Xu, and X. Huang, “Self-adaptive trust man- agement based on game theory in fuzzy large-scale networks,” Soft Computing, vol. 21, no. 4, pp. 907–921, 2017. [29] A. Singraber, J. Behler, and C. Dellago, “Library-based LAMMPS Implementation of high-dimensional neural net- work potentials,” Journal of Chemical ?eory and Compu- tation, vol. 15, no. 3, pp. 1827–1840, 2019.
Computational Intelligence and Neuroscience – Hindawi Publishing Corporation
Published: Jan 7, 2022
You can share this free article with as many people as you like with the url below! We hope you enjoy this feature!
Read and print from thousands of top scholarly journals.
Already have an account? Log in
Bookmark this article. You can see your Bookmarks on your DeepDyve Library.
To save an article, log in first, or sign up for a DeepDyve account if you don’t already have one.
Copy and paste the desired citation format or use the link below to download a file formatted for EndNote
Access the full text.
Sign up today, get DeepDyve free for 14 days.
All DeepDyve websites use cookies to improve your online experience. They were placed on your computer when you launched this website. You can change your cookie settings through your browser.