User’s Potential Interest Network Based Social Recommendation Method And Its Application

With the rapid development of social networks, social recommendation hasbecome an important branch of recommendation technology research. In the currentrecommendation technology, there are some disadvantages. On one hand, thetraditional collaborative filtering recommendation technology can provide somemore accurate personalized recommendations for users, but it cannot make use ofusers’ social information; on the other hand, the social recommendation can providerecommendations through users’ tags, trust and friends relationships, but it cannotmake full use of other information in the web. Therefore, we can consider the use ofinformation in social webs to improve the traditional collaborative filteringalgorithms.In this paper, we will take LEHU BT and LEHU forum as an experimentalsubjects, research on how to take full advantage of user’s activities in the forums, toexcavate potential interest relationship between users, use this information toimprove the traditional collaborative filtering algorithms, to improve the accuracy ofthe algorithm and solve the problem of data sparsity.This paper’s mainly work and innovation points are as follows:1. The relationship between user’s activity behaviors and their interest in forumis studied and the collaborative filtering algorithm named UPINCF (User’s PotentialInterest Network based Collaborative Filtering algorithm) based on user’s potentialinterest network is proposed. Firstly, we had studied the relationship between user’spotential interest and their activities in the forum. And then, the methods forcalculating the user’s potential interest and user’s potential interests similarity areproposed. Secondly, the user potential interest network which is built to express theinterests relationship between users is employed to improve the user-basedcollaborative filtering algorithm. And the recommendation results show that theproposed UPINCF has the high recommendation accuracy.2. The characteristics of user’s interest changes with time is studied and thecollaborative filtering algorithm based on users’ potential interest network namedIMUPINCF (Interest Migration improved User’s Potential Interest Network basedCollaborative Filtering algorithm) improved by the interest migration is proposed.The user potential interest contribution degree of one user activity is calculated usingthe user active time according to the characteristics of user’s interest changes withtime. And then importance degree of activity is used to describe the user’s potentialinterest changes with time. So the aim of UPINCF algorithm optimized by userinterest migration is realized. The IMUPINCF algorithm improves the accuracy ofrecommendation results and the ability of mining long tail projects.3. Studied the transferability of user’s interest, we had proposed thecollaborative filtering algorithm based on user’s potential interest network called ITUPINCF (Interest-Transmitting improved User’s Potential Interest Network basedCollaborative Filtering algorithm) improved by interest-transmitting. After studiedthe transferring characteristic of user potential interest similarity, we defined thetransferring rules and calculation formulas of user’s potential interest similarity. Andthen, interests transmission is employed to mine the relationship of indirectlypotential interest similarity among users in order to improve the recommendationaccuracy of UPINCF. This method solves the problem of data sparse in a certaindegree.4. The new methods proposed in this paper are applied to the personalizedrecommendation of LEHU BT, and the online test results show that the performanceof the new methods in precision and solving the user cold start is better thantraditional collaborative filtering algorithm for different types of users.