Research On Trust-based Recommendation System

People's lives have been greatly improved by the Internet which offers almost all the knowledge,music,news and any other services.However,we are overwhelmed simultaneously by too much "choice".Although “information overload” can be solved effectively by exploiting the recommendation system problems,such as data sparsity,cold start,"care" attack and scalability remain enormous challenges for the traditional recommendation algorithm.In order to solve these problems,the researchers have tried various methods,the trust-based recommendation system(TARS),which introduces the trust mechanism on the basis of the traditional recommendation algorithm,emerges as the times require.By exploiting the trust networks of users and the trust between users,this kind of recommendation system connects the original isolated users exploiting the trust between users so that it can cope with those problems well.Therefore,the study of TARS has become one of the important research topics in the field of recommendation system and significant achievements have been made.However,the performance of existing schemes need to be improved,especially the accuracy of rating prediction.In this thesis,we focus on further promoting the performance of TARS.Firstly,the knowledge of the trust management and the common trust model are introduced.What's more,we study the most widely used algorithms in the field of recommendation system,and the advantages and disadvantages of these algorithms are analyzed.Then,a new collaborative filtering algorithm DTCF based on trust network is designed.This algorithm considers not only the trust between users but also the similarity between users.It consists of three stages: acquiring recommender list,forecasting user's rating,and updating trust network.In the phase of forecasting user's rating,the concepts of "user's reputation" and "quality of item" are introduced and incorporated into rating prediction to improve algorithm's accuracy.In the stage of updating trust network,based on the correlation between the trust degree and the consensus of “If entity A trusts the entity B,the probability that the entity A is similar to the entity B is relatively large”,the algorithm adopts a new method to update the trust network,which further promote algorithm's operating efficiency and accuracy.Finally,six experiments are designed on the real data set including Epinions and Flixster.We have verified the correlation between the trust degree and the similarity degree,and studied the influence of the parameters on the DTCF algorithm.Then the DTCF algorithm is compared with the classical collaborative filtering algorithms and the existing trust-based recommendation algorithms in terms of the coverage and accuracy of the algorithms.Experimental results show that the DTCF algorithm has better performance under the same conditions.