Trust Semirings Model Based On Similarity For Trust Propagation

Entering the big data era,there are more and more information resources on the network,so it might be hard for users to find things they want,especially when they are uncertain about what they need.While the personalized recommendation system,according to the user's history record,can perform data analysis,build the user's model,actively judge the user's needs,and then provide users with information which may interest them to meet their needs,which can improve users' experience and satisfaction.The most widely applied and most highly accepted algorithm in the personalized recommendation system is the collaborative filtering algorithm.In brief,the collaborative filtering algorithm refers to making use of similar items or interests-similar users(collaborative)to remove the uninterested items or users(filtering),so as to recommend items or users they might be interested in.However,there are some inherent problems,like data sparsity,cold start and so on,which could lead to the poor prediction accuracy or the unpredictable results for new users.It is discovered in many studies conducted by scientific researchers for the above situation that these problems can be alleviated in some degree by adding social network in the recommendation.Since the social network encompasses rich social information,such as user information,relationships between users,location,it can facilitate the building of the user characteristic model for further recommendation.Then,how can the user's social information be better utilized and how can the recommendation be improved? These are hotspots in current research as well as the research content and targets of this thesis.Firstly,we discuss the measurement of trust between users,and propose a mixed trust measure method based on common friends and favorite items,which can transfer the simple binary trust into the non-binary trust that can differentiate trust degree,and choose appropriate trust weights for the calculation of direct trust degree between users through experiments.Secondly,we put forward trust propagation model based on semiring.On the basis of trust-semiring,we define the trust opinion vector as the trust propagation unit,perfect the original model with similarity as the certainty model,and consider the predilection of the user in the trust propagation to acquire higher quality trust user.In the end,the results we acquired from the 5-fold cross validation on FilmTrust data set show that the improved algorithm actually improves the accuracy of recommendation and effectively reduces errors.