Social Recommendation Models Based On Fine-grained Trust Prediction

With the rapid growth of information quantity,information overload has already become an inevitable problem,it is becoming increasingly difficult to obtain effective data quickly and accurately.Recommender systems which arising in this environment,can effectively mine the behavior characteristics and interest of users,they can provide users with personalized information filtering and commodity recommendation services on the basis of modeling according to user preferences.In the field of information retrieval and e-commerce,recommender systems have become an effective way to deal with information overload and improve user satisfaction.Collaborative filtering has become one of the most widely used recommendation techniques because of its simple operation and high accuracy.However,its performance is greatly affected by the data sparsity and cold start problems.Development of social networks has extended a new research field in recommender systems,and users are no longer regarded as independent individuals because of the social relations between them.Trust is the core and key part of social relations,trust based methods can alleviate data sparsity,improve the accuracy and coverage of systems,and also can improve the system security by providing more reliable recommendations.Further enhancing the performance of recommender systems based on social trust is mainly discussed in this paper.We propose a fine-grained trust prediction model through a combination of multiple trust attributes,and then extend two classical collaborative filtering algorithm to construct new social recommendation models.Based on the attributes of transitivity,similarity and multi-aspects of trust,we build a fine-grained trust prediction model.First of all,to extend the coverage of the trust network,a single trust propagation method based on mining of key users and an optimization method based on similarity are proposed;then,trust weights are generated through the social interaction and frequency between users;finally,implicit factors which influence the fine-grained social trust are generated by employing multi-dimensional trust prediction,we use these factors to further fill missing fine-grained social relations.Experiments show that this method can accurately generate missing social relations and their weights.Based on fine-grained trust prediction model,we extend the nearest neighbor algorithm and latent factor model,then propose the fine-grained trust based neighbor model NNST and the social collective matrix factorization model CMFST.Multiple experiments verify our proposed social recommendation models can effectively solve problems existing in the traditional personalized recommender systems and can provide more accurate social recommendations for online users.