Research On Cross-Recommendation Algorithm Based On Social Trust

In the era of big data,it is a difficult task for both the provider to make the providing information attractive and the receiver to look for interesting information from the receiving information.The recommendation system is an effective way to solve this problem.Collaborative filtering is a typical method by using collective intelligence.However,some problems including cold start,data sparsity and poor scalability are affecting the performance of collaborative filtering algorithm.In order to alleviate the above problems,we proposed an improved collaborative filtering algorithm by using social network data.The main works are as follows:(1)The related technologies and applications of recommendation system including the state-of-the-art methods by using social data were introduced in the paper.Aiming at alleviating the existing problems of collaborative filtering recommendation algorithm in single domain,the existing collaborative filtering recommendation methods were discussed in detail.Furthermore,an overview of the relevant research works considering data from multi-source domains was presented.(2)To alleviate the cold start problem of the memory-based collaborative filtering recommendation algorithm,a memory-based collaborative filtering algorithm considering social trust in social network is proposed.Before computing user similarity and predicting ratings,the method achieves more accurate user similarity and predictive ratings by fulfilling the original user-item rating matrix using social trust information and alleviate the sparsity of user-item rating matrix effectively.Experimental results show that the proposed algorithm can alleviate the cold start problem of recommendation algorithm in single domain and improve the accuracy of recommendation system.(3)For the model-based collaborative filtering recommendation algorithm,an algorithm based on matrix decomposition that combines social trust is proposed.By using the user's social trust as a regular item in the matrix decomposition,such as the confidence between the user and his/her trusted friends is as weight,the user's preference becomes close to his/her trusted friends in the social network.The proposed method not only improves the accuracy of the recommendation system but also improves the interpretability of the recommending results.The process of matrix decomposition can be completed offline,which makes the method has better scalability and practicality.Experimental results show that the algorithm can improve the accuracy of recommendation system effectively.