Design And Implementation Of Trust - Based Recommendation Method In Social Network

At present, the recommender technology has been widely applied in various fields to solve the problem of "information overload", while the traditional recommendation technology has the problem that the accuracy not high, the coverage not wide. With the development and the rise of social network, people will more and more concentrate on it. Combining the information in social network with traditional recommendation technology can largely slove the above shortcomings. This paper fully explored the relationship between the social networks users, combined with the traditional algorithm to predict users’ ratings of the target item. Main contents of this thesis are as follows:1. The collaborative filtering algorithm based on expert trust. In the real life, people tend to trust the thought of expert. In our paper, we look for expert users in a social network, using experts’ perceptions of the target project to predict the target user’s rating. The user’s expert-degree based on global trust and local trust. The global trust is used to measure the users whether authority; Local trust is used to measure the quality and quantity of the user. Finally find the experts with the most similar to the target user. According to experts’ opinions to predict ratings using the collaborative filtering algorithm. Finally we verified by experiments that our algorithm can enhance the accuracy of recommendation result, and also can recommend items for cold start that they would be like.2. The recommender algorithm is based on rating propagate between trust users. In the real life, if you want to know the evaluation of an item, you can ask your friends, and then he will ask his friends too, finally, the result will feedback to you along the trust path. Based on this thought, our paper analysis how the rating propagate between the trust users, and then puts forward a simple model of rating propagate, then analyze the influencing factors of the model and improved, and then puts forward a complete model and state the algorithm detailed. Finally we verified the performance with experiment to prove our method is efficient.