基于LDA话题关联话题演化.PDF

44 11 上 海 交 通 大 学 学 报 V ol. 44 N o . 11 20 10 11 JO U RN A L OF SH A N GH AI JIA O T O NG U N IV ERSIT Y N ov . 20 10 :( 2010 L DA 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 391 : A Topic Evolut ion Based on LDA and Topic Association CH U K eM ing , L I Fang ( Scho ol o f Elect ronic, Inf ormat ion and Electr ical Engineering , Shanghai Jiaot ong U niv er sit y, Shang hai 200240, China Abstract: T opic ev olution w ill help people to learn inf orm at ion quickly. In t his paper, a m et hod w as pro po sed t o disco ver t opic s evolut io n ov er t ime by t opic det ectio n and relating t opics in dif ferent t ime peri ods. T he m et hod applies L DA mo del on temporal do cument s to ex t ract to pics. T he num ber of t opics in dif f erent t im e periods is dif f erent. Relat ing t opics in consecut ive t ime periods is based on JensenShannon di vergence and f eat ur es sim ilarit y. Ex periment s show t hat t he met hod can det ect new t opics and describe t opic s evo lut ion over t ime ef fect ively. It not only show s that t he to pics ev olve w it h t im e, but also t hat t he cont ent of t opics chang e w it h t im e. Key words: t opic det ect ion ; to pic associat io n; t opic evo lut ion ; lat ent Dirichlet allocat ion ( LDA ( T opic Det ect ion and , T racking , T DT