基于LDA话题关联的话题演化_楚克明.pdf

44 11 上 海 交 通 大 学 学 报 Vol. 44 No. 11 2010 11 JOURNAL OF SH ANGH AI JIAOT ONG UNIVERSIT Y Nov. 2010 : 1006-2467( 2010) 11- 1496-05 LDA 楚克明, 李 芳 ( , 200240) : 话题演化可以帮助人们快速获取信息和了解 势. 提出了一种挖掘话题随时间变化的方 法, 通过话题抽取和话题关联实现话题的演化. 对不同时间段的文集进行话题的自动抽取, 话题数 目在不同时间段是可变的; 计算相邻时间段中任意2 个话题的分布距离和话题的特征向量相似度 实现话题的关联. 实验结果证明, 该方法不但可以描述同一 个话题随时间的强度变化, 还可以描述 新话题的产生, 旧话题的消失以及话题内容随时间的演化. : 话题探测; 话题关联; 话题演化; 潜在狄里特里分配 : T P 91 : A Topic Evolut ion Based on LDA and Topic Association CH U K e-M ing , L I Fang ( School of Electronic, Information and Electrical Engineering, Shanghai Jiaotong U niversity, Shanghai 200240, China) Abstract: Topic evolution w ill help people to learn information quickly. In this paper, a method w as pro- posed to discover topic s evolution over time by topic detection and relating topics in different time peri- ods. T he method applies LDA model on temporal documents to extract topics. T he number of topics in dif- ferent time periods is different. Relating topics in consecutive time periods is based on Jensen-Shannon di- vergence and features similarity. Experiments show that the method can detect new topics and describe topics evolution over time effectively. It not only show s that the topics evolve w ith time, but also that the content of topics change w ith time. Key words: topic detection; topic association; topic evolution; latent Dirichlet allocation ( LDA) (T opic Detection and , T racking, TDT) , . . ( Latent Dirichlet A lloca- [ 1] , tion, LDA) ,