Recommendation Algorithm Regularized With User Trust

Recommender systems have been widely used to solve the problem of information overload and provide high quality personalized service.Many trust-aware recommender systems can effectively reduce the problem of data sparse and cold start and improve the performance of recommender systems by using explicit trust relationship,which is specified by users with binary value.However,explicit trust is lack of trust,user participation.To help resolve the issues mentioned above,some implicit trust-aware algorithms were proposed.These methods generally implicit trust as the result of single aspects,we considering impersonal aspects,for example,similarity and transitivity.In reality,trust is a multiple facet,which has not been well utilized in the existing recommender systems.For the issue mentioned above,we focus on the realization of trust predict model and the improvement of trust-aware recommender algorithm.In order to better predict trust relationship,we take the impersonal aspects and interaction aspects into consideration in a recommender system algorithm,and propose to a classification approach to address trust prediction problem.In addition,based on the assumption that trust users have similarity interests,we take the implicit and explicit influence of trust into consideration in a recommender system algorithm to improve the performance of the recommender systems.The main research contents and research achievements are summarized as follows:(1)To explore trust feature and build trust predict model.An analyze of the impersonal aspect(such as the number of terms and similar users)and interactive aspect(such as similarity,ability,predict value),and given the relevant compute formula.We are computationally feature based on users' historic ratings,the logistic regression model is used to train the above aspects and construct trust prediction model.(2)To propose a novel approach(TrustReg)that incorporates explicit and implicit influence of trusted users.The implicit influence of a user's trustors is used to build her trust-feature vector,that is,consider the contribution of trusted users in the rating predict task.Explicit influence is the regular item that add user and his trusted user's feature-specific vector in the loss function.(3)To conduct many experiments to evaluate the effectiveness of trust predict model and recommender algorithm.By using the trust predict model,we refined the explicit trust relationship and applied in the existing trust-aware recommender systems Experimental results on the real data sets show that the trust predict model can effectively realize the trust prediction and TrustReg achieves better accuracy than other recommendation algorithm.