The Study Of Trustworthy Recommendation Approach Based On Multidimensional Trust Model

With the rapid development of Internet, there has been a great increase in the amountof information. So how to make use of online resources effectively becomes a challengingtask. Recommender systems, as a kind of information filtering technology, have becomean effective way to solve the information overload problem. Specially, among thenumerous recommendation methods, collaborative filtering is one of the most widely usedand successful recommendation technologies. However, due to the openness ofrecommender systems, some malicious users might try to insert a lot of fake profiles intosystems in order to bias the recommendation results, which results in a decline in users’satisfaction with the recommender systems. Therefore, with the emergence of shillingattacks, how to provide trustworthy recommendation for target user has become a keyissue to be solved. In this thesis, we have made some deep research of trustworthyrecommendation approaches from the following four aspects: the credibility of data source,the strategy of choosing neighbors, the robustness of recommender systems and thecredibility of top-N recommendation.Firstly, aiming at the problem that the existing computational approaches of trust cannot measure the trust relationship between users accurately, a multidimensional trustmodel incorporating users’ degree of suspicion is proposed. According to users’ ratings onitems, the users’ degree of suspicion are calculated through introducing the theory ofentropy and the idea of density-based local outlier factor. On the basis of that, in order toimprove the calculation accuracy for the degree of trust between users, the trust attributesare analyzed and measured from different angles using the source credibility theory.Secondly, aiming at the problem that the existing heuristic recommendationalgorithms suffer from lower recommendation quality, a collaborative filteringrecommendation algorithm based on double neighbor choosing strategy is proposed. Onthe basis of the computational result of similarity, the preference similar users of targetuser are chosen dynamically. We can measure the degree of trust between target user andpreference similar user using the proposed multidimensional trust model incorporating users’ degree of suspicion. The trustworthy neighbor set of target user is selected inaccordance with the degree of trust between users. The recommendation for target user isgenerated by incorporating the double neighbor choosing strategy with the conventionalcollaborative filtering recommendation algorithm.Thirdly, aiming at the problem that the existing model-based recommendationalgorithms have the disadvantage of weaker robustness, a robust collaborativerecommendation algorithm incorporating trustworthy neighborhood model is proposed. Atrustworthy neighborhood model is constructed by incorporating the proposedmultidimensional trust model with baseline estimate approach. On the basis of that, wecan provide recommendation for target user using the M-estimator based matrixfactorization approach.Fourthly, aiming at the problems that the existing top-N recommendation approachessuffer from lower recall and weaker robustness, a top-N recommendation approach basedon reliable users is proposed. We can measure the degree of trust between users using theproposed multidimensional trust model. And the computational result is used to choosereliable users for target user. Moreover, based on the rating information of reliable users,we can choose candidate recommendation items using some approaches and providetop-N recommendation for target user using the strategy of average aggregation.Lastly, we have compared the performance between the proposed approaches andothers to demonstrate the effectiveness of the proposed approaches. Also, we propose thefurther research work.