面向微博用户标签推荐的关系约束主题模型

社会化标签系统允许用户使用个性化的词汇对网络中的资源进行标注而被用户广泛接受。在微博网络中，用户可以为自己加注标签以推广自己或者方便别人找到自己。深入分析了微博用户数据，总结了微博用户标签的特点，针对LDA（latent Dirichlet allocation）主题模型在处理短文本时存在的不足，提出了一种基于好友关系约束主题模型。在此基础上对微博用户标签进行主题分析，计算用户的主题分布，对标签词进行聚类，并最终为用户推荐标签。通过对比实验证明了该方法可以提高标签推荐的准确度。

Relationship Bind Topic Model Toward Tag Recommendation for Micro-Blog Users

Social tagging systems which allow users to use personalized words to annotate resources are widely accepted by users. In micro-blog network, user can attach personalized labels to himself/herself so as to promote his/her recognition or facilitate others to find him/her. With the deep analysis of the micro-blog dataset, this paper summarizes the characteristics of user＇ s tag. Aiming at the shortcomings of LDA （latent Dirichlet allocation） topic model in dealing with short texts, this paper proposes a new relationship bind topic model. After that, the user＇ s topic distribution is calculated, and topic words are clustered. At last, tag recommendation is performed. The comparative experiments show that the proposed method can improve the accuracy and diversity of tag recommendation task.