Design And Implementation Of Online Recommendation System With Filtering Fake Comments

With the rapid development of network technology,information exchange has become more and more frequent,which has brought difficulties in information selection.Hence,the recommendation system came into being.Based on the needs and interests of users,the recommendation system can recommend the suitable products to the user.In particularly,personalized recommendation systems are now widely used in many fields,especially on the e-commerce.However,in practical applications,the recommendation system is vulnerable to fake comments,and the recommendation effect and credibility of recommendation system are drastically decreased.Therefore,for the recommendation system,it is important to achieve effective fake comment filtering and accurate recommendations.At present,there are a large number of works on fake comments filtering and recommendation systems,but the following problems still exist in the research process: 1)The identification and filtering of fake comments relies on a large number of labeled true and fake comment data sets as the basis,while most of the fake comment data that has been generated is artificially generated and the amount of generated fake comment data is small.That is,how to obtain the comment data which mixes the fake comments and real comments is the basic to further develop fake comments filtering methods;2)Well-designed fake comments lead to a sharp drop in recommendation system performance,and the existing fake comments detection methods which use the statistical information of the text as features are hard to filtering the fake comments.So,how to effectively filter fake comments in the recommendation data is a crucial issue for the recommendation system;3)The existing recommendation algorithms have a poor perform when handling sparse data,and the data sparsity will directly affect the recommendation effect of the recommendation system.In view of the above problems,this paper proposes a set of offensive and defensive strategies for fake comments,and designs an online clustering recommendation system.In the offensive and defensive strategies for fake comments,this paper proposes a generative text method based on conditional text generation adversarial network(CTGAN),and designs a double-loop graph fake comment detection method that uses data recycling to improve the effect of filtering fake comments.Considering the problem of data sparsity,an online clustering recommendation system based on network embedding,namely N2 VSCDNNR,is designed.The specific research content includes the following parts:(1)In order to obtain the comment data which mixes the fake comments and real comments,this paper proposes a generative text method based on conditional text generation adversarial network(CTGAN).CTGAN generates variable length texts with specified emotion labels,and employs an automated word level replacement strategy to improve the quality and diversity of the generated text.The experiments of CTGAN have confirmed that it is difficult for a detector with only text information to filter fake comments generated by a machine learning method.(2)For filtering fake comments,this paper proposes a fake comment detection method based on double-loop graph.This paper proposes data recycling method to calculate the reliable user confidence and item confidence,and optimizes the initial user confidence and item confidence for the second recycling graph.Then,this paper designs a weighted graph filter that considers the user's personal influence on the item,and performs a second loop iteration based on the weight graph filter and initial optimized confidence to update the confidences.(3)To inprove the performance of recommendation system,this paper proposes a novel clustering recommender system based on node2 vec technology and rich information network,namely N2 VSCDNNR.It combines the new network representation learning method(node2vec)with the rich information network to effectively represent the potential features of each node(user or item)in network in the form of vectors.Then,to efficiently cluster users and items,this paper proposes a spectral clustering algorithm based on dynamic nearest-neighbors(SCDNN),which can automatically determine the clusters number and has high clustering effect.After clustering,we propose the staged personalized recommendation to realize the personalized recommendation of items for each user.