关键词: 隐含语义分析, 向量空间模型, 话题发现, 微博, 两阶段聚类

Abstract: As the large popularity of micro-blog and awareness continues to improve, hot topic of micro-blog detecting has become the current research focuses. For short texts, there exist high-dimension, sparse, synonymy and polysemy problems for Vector Space Model（VSM） text presentation, making it difficult to measure the similarity of the texts accurately. This paper presents a two-stage cluster based on Latent Semantic Analysis（LSA） topic detection approach. Firstly, the concept of hot topic is introduced to select micro-blogs with certain attention, using LSA to model the dataset. Then CURE algorithm of hierarchical clustering is employed to determine the initial centers. Finally, the hot topic clustering results are obtained through K-means clustering. Experimental results on real micro-blog dataset verify the validity of the method.

Key words: Latent Semantic Analysis（LSA）, Vector Space Model（VSM）, topic detection, micro-blog, two-stage clustering