Abstract: As micro-blog grows more popular and widely used, micro-blogging site such as Sina has become a huge source of information, although the traditional method of texts, topic mining has been extensively applied research. For this special kind of text of micro-blogging, traditional text mining algorithm can not be well studied. In order to compensate the deficiencies of current topic mining for micro-blogging platform and considering the sparsity and multidimensional characteristics of micro-blogging, this paper proposes targeted pretreatment method and combines the users’ micro-blogging data with AT model, then mining the micro-blog topics by gibbs sampling, getting users’ interest through extracting the topics of authors. Through the experiments on a real data sets, as well as comparison with LDA models prove that the model can get micro-blog topics effectively.

Key words: micro-blog, topic mining, author-topic model, Gibbs sampling

摘要： 随着微博的日趋流行与广泛使用，新浪等微博网站已经成为海量信息的来源，虽然传统的文本主题挖掘方法已经得到广泛的应用研究，但对于微博这种特殊结构的文本，传统的挖掘算法不能很好地对其进行研究。为了弥补目前微博平台主题挖掘方法的不足，以及考虑到微博信息的稀疏性，多维性等特点，提出有针对性的预处理方法，将用户微博数据与AT模型结合，通过吉布斯采样进行微博主题挖掘，对作者主题进一步提取得到用户兴趣。通过在真实数据集上的实验，以及与LDA模型对比，证明该模型能有效得到微博主题。

关键词: 微博, 主题挖掘, AT模型, 吉布斯采样