Abstract: In order to face the challenges of feature sparsely of short text messages for microblog hot topic detection, this paper proposes a hot topic detection method based on the combination of term mutual information and probabilistic topic model. Symmetric Nonnegative Matrix Factorization（sNMF） is performed on word co-occurrence with word mutual information and the matrix of term-topic matrix is thereafter inferred. Probabilistic Latent Semantic Analysis（pLSA） model is then adopted to model the topic-microblog. The hotness of topic is analyzed and sorted. Experiments show that this method can effectively cluster and detect the hot topics.

Key words: term co-occurrence matrix, symmetrical nonnegative matrix factorization, probabilistic latent semantic analysis, micro-blog hot topic detection

摘要： 为了解决短文本信息流的特征稀疏性对热点话题发现带来的挑战，提出了结合词语互信息和概率主题模型的微博热点话题发现方法。通过建立词共现矩阵并应用对称非负矩阵分解算法获取词项-主题矩阵，再利用概率潜在语义分析模型进行主题发现，最终通过定义微博热度分析和排序，有效地支持微博热点话题发现。实验表明，此方法能有效地进行话题聚类并检测出热点话题。

关键词: 词共现矩阵, 对称非负矩阵分解, 概率潜在语义分析, 微博热点话题发现