Research And Implementation Of The Friends Filtering System Based On Trust Model And Topic Model

With the rapid development of web technology, social network services(Social Network Service, SNS) play a very important role in people's daily life. In a social network, people can build and extend their own social networks by adding unfamiliar user or following other people, what's more, people can share our latest imfprmation and get the imformation we are interested in. However, with the rapid expansion of the SNS, how to find a friend with the same interests and the high credibility become more and more difficult. So in order to solve this problem, friend recommendation based on social network have been proposed.Generally speaking, traditional content-based recommendation algorithm can recommend users with the same interests to other users by their historical information. But the acceptance of this algorithm is not high because of lack of the credibility of the recommended users. Meanwhile, the words generated by traditional content-based recommendation algorithm is insufficient in "polysemy". So in order to reflect user's interests preferences, we introduce the topic model to mining user's hidden interests. Therefore, we proposed a hybrid recommender model with combination of content-based recommendation algorithm and trust model to solve the difficult of lack of the credibility.Our model contains two models: Topic Model and Trust Model. Topic model uses assembled-based LDA model and Gibbs sampling method for estimating the parameters to get the user's interests preference, and calculate the interest's similarity by the user-topic probability. In trust model, we firstly define a trust propagation model and then combine rust propagation model and trust calculation method to calculate the credibility between the users. Finally, we calculate the ultimate recommend degree based on a combination of credibility and interest's similarity, based on which recommends Top-N friends to a target user.