Trust Network Random Walk Model Based On The User Preferences

With the development of network communication technology and the expansion of Internet information resources,the phenomenon of information overload is becoming more and more serious.The emergence of traditional information services eases the problem of information overload,however could not meet the user's personalized needs because of its general characteristics of the public demand.Information filtering represented by personalized recommended systems could predict users' interests,provide valuable or potentially valuable information for the users by analyzing the needs of users,which solve the problem of personalized needs.Recommendation system used in information service widely has become an indispensable information service form for the major information service providers.With the development of social network research,many researchers combine social networks with personalized recommendation to solve problems such as the difficulty of data sparsity,cold start,expandability and robustness of model,which improves the scalability and accuracy of recommendation systems effectively.However,most of the algorithms simply take advantages of a trust relationship to find the nearest neighbor set,failed to take into account that the suggestions of trust users may not be suitable for the target user.In view of the insufficiency of the TrustWalker random walk model,the main work of this thesis as follows:1.Using the matrix decomposition to reduce the dimensionality of the user scoring matrix,and retain the main information features.With the high dimensional sparse scoring matrix,the similarity can be calculated efficiently and accurately,which could alleviate the problem that the similarity of the high dimensional sparse score matrix can not be calculated easily.2.Taking advantages of clustering idea and user preference to reduce the search space of the project and explore the user's relationship of the project category and the weight of interest.In the process of selecting similar items in the random walk,the user preference is used to predict the user's score,which can accelerate the convergence of the model and ensure its effect of the recommended and interpretability.3.Giving full consideration to the user trust relationship information,making full use of authority and partial trust to refine the credibility and explore the relationship between trust and interest.Finally a trust network random walk model PTTrustWalker based on user's preference is proposed.