异质社交网络中多通道特征融合的好友推荐模型

合理有效的好友推荐算法对于社交网络的发展和扩张有重大的意义。然而随着社交网络的复杂化和异质化,传统推荐系统中协同过滤推荐方法不能满足需求。针对异质社交网络中存在着大量的内容相关信息这一特点,根据好友推荐的需求,提出了多通道特征融合的好友推荐模型。该模型对用户相关的多维特征进行挖掘与利用,包括显性特征(如用户profile,用户tag,社交关系等)和隐性特征(如用户重要度,挖掘用户标注发现其领域兴趣等),并进一步将这些内容相关的多特征融合到协同排序算法中进行学习训练。实验结果表明,随着多个内容特征的逐步融合,算法的MAP值稳步提高,最终相对未融合的协同排序方法提高了12%,并在一定程度上的解决了冷启动问题,提高了好友推荐的多样性。

Friend Recommendation Model with Multidimensional Features in Heterogeneous Social Network

Reasonable and effective friend recommendation algorithm is very significant for the development and expan_sion of social network. However, with the development of social network, more and more applications in the social network make it a heterogeneous network, which disables the traditional collaborative filtering recommendation methods. Consider the large amounts of contents relevant information in heterogeneous social network, as well as the demand of friend recom_mendation, multidimensional features analysis based friend recommendation model is proposed. In this model, the user rele_vant multidimensional features are mined, which include explicit features such as user profiles, user tags and social relation_ship and latent features such as users’ authority, users’ research interests and so on. Those features are taken into considera_tion and fused with former collaborative ranking model. The results of the experiment show that, through fusion of content features gradually, the MAP values increased steadily, and the MAP raises as much as 12% compared with the ones with no multidimensional features. What’s more, this method solves the problem of cold start to a certain extent.