Research On Collaborative Filtering Recommendation In Social Networks

In the internet,mass information resources have allowed people to enjoy convenience. In recent years, however, users are increasingly suffering from the information overload. It has long been a problem for scholars to delve into the methods that could filter out the information necessary to meet users' needs.Personalized recommended system is a solution to solve the information overload issue by mining users' preferences from historical data that users have left in the system and recommending items based on their preferences. With a rising number of system users, however, many problems arise including rating data sparse and recommended results that lack incredibility.To solve these problems, based on the study at home and abroad, this paper proposes a collaborative filtering recommendation algorithm in social networks.Collaborative filtering algorithm is one of the most widely used recommended algorithms. But few data on historical users existing in the system is inadequate to support system analysis in exploring their interests. Social network provides a new way to understand users' interests. It uses trust information among users in social networks to present the user similarity calculation model based on the social trust. The social trust can be divided into global trust, the local trust, trust in a single path,multipath trust, trust in one direction, two-way trust and other dimensions. In terms of the lack of recommendation diversity and long-tail problem, this paper proposes a collaborative filtering method based on time context and item confidence. First, it constructs a trusted user-item rating matrix using the result of the previous similarity calculation method. Then, we calculate the similarity of the target users by integrating item popularities, user interest ratings and time decay factor. Third, it selects the nearest neighbors of target users and get predicted ratings.At the end, this paper compare the proposed algorithm with other similar ones based on users and social trust using Extend Epinions data set. Experimental results show that the algorithm has an effect on accuracy and coverage.