The Recommendation Service Based On Multi-dimensional Trust And Data Detection

With the rapid development of Internet technology,the rapid growth of Internet information has been far beyond our ability to receive information,bringing the problem of information overload.Therefore,the method of finding interesting information resources deserves,in order to solve this problem,recommender system emerges as the times require.One of the most popular and successful recommendation algorithms used in recommender system is collaborative filtering.As one of collaborative filtering algorithms,the Slope One algorithm is easy to implement and has good scalability.In addition,due to the wide application of the recommender system and its huge economic benefits,the security problem of the recommendation algorithm becomes more obvious.Some malicious users began to make many fraud ratings,which not only lead to the uncertainty of recommendation results.Therefore,in terms of this kind of user profile injection attack,how to provide users with more accurate and satisfactory recommendation results is an urgent problem to solve.In this thesis,the Slope One algorithm of multi-dimensional trust value is studied from the aspects of the authenticity of the data set,the trust relationship between users and the credibility of the score.Firstly,this thesis analyzes the shortcomings of the traditional Slope One algorithm and weighted Slope One algorithm.In order to solve the problem of low accuracy associated with these two algorithms,we build a credible recommendation model that combines the direct trusted relationship between users and the trust ratio for rating values.Based on this model,we present our improved Slope One algorithm.Secondly,in consideration of the existence of a large number of fraud ratings among data set,this thesis introduces the security problem of the recommendation algorithm.By employing Bayesian model,we can identify those fraud ratings from original data set based on ratings and corresponding timestamps,then we can conduct data detection on data set.After that,we conclude our improved algorithm that combining user trust and rating trust.At last,with the MAE as measure for prediction accuracy,we compare and analyze these two algorithms,which confirms that our improved algorithm has better prediction accuracy and can improve the quality of recommendation service.Finally,this thesis introduces the system architecture of the prototype recommender system and the design and implementation of key modules,and then we simply analyze the performance advantage of the recommender system.