Abstract:  As the exponential growth of the number of users in social networks, how to recommend friends with similar interests from massive users has become one of the research focuses in social recommendation. This paper proposes a hybrid friend recommendation algorithm based on user-item-tag graph and users' personal interests. Firstly, the similarities between users are calculated by mass diffusion method in tripartite graph. Secondly, the relationship between users and tag graph of users is introduced, and the communities in the tag graphs of users are detected for the topics of users' interests. Then the similarities between users are measured by Kullback-Leibler divergence according to their topics distributions. Finally, two kinds of similarities are integrated for user recommendation by the harmonic mean method. The experimental results on Delicious and Last.fm datasets demonstrate that the proposed algorithm can effectively improve the accuracy of Top-N recommendation in terms of precision and recall. At the same time, the experimental results on the academic social network site—SCHOLAT for the scholar recommendation prove that the proposed algorithm improves the recommendation of core network members.

Key words: friend recommendation, tripartite graph, tag, social network, collaborative filtering

摘要： 随着社交网络的用户数量呈爆炸式增长，如何为用户推荐具有相同兴趣爱好的好友已成为当前研究的焦点。为此，提出了一种基于“用户-项目-用户兴趣标签图”的协同好友推荐算法。该算法首先利用基于“用户-项目-标签”的三部图物质扩散推荐算法来计算用户之间的相似度，并引入“用户-用户兴趣标签图”二元关系，通过用户的兴趣标签图来发掘用户的兴趣主题；然后根据用户主题分布，利用KL距离来计算用户之间的相似度；最后将两组结果采用调和平均数方式融合得到用户间的综合相似度，并进行好友的推荐。通过在Delicious和Last.fm数据集上的实验证明，该算法能有效提高Top-N推荐的准确率和召回率，同时通过在学术社交网站——学者网数据集上进行的学者推荐实验表明，该算法能有效提高核心用户的推荐度。

关键词: 好友推荐, 三部图, 标签, 社交网络, 协同推荐