一种基于LDA模型的主题句抽取方法  

Approach for topical sentence extraction based on model LDA

作　　者：王力[1,2] 李培峰[1,2] 朱巧明[1,2] 

机构地区：[1]苏州大学计算机科学和技术学院,江苏苏州215006 [2]江苏省计算机信息处理技术重点实验室,江苏苏州215006

出　　处：《计算机工程与应用》2013年第2期160-164,257,共6页Computer Engineering and Applications

基　　金：国家自然科学基金(No.90920004;No.60970056;No.61070123;No.61003153);江苏省高校自然科学重大基础研究项目(No.08KJA520002)

摘　　要：在基于Web的主题关键词查询扩展,获取候选主题句的基础上,提出一种基于LDA模型的主题句抽取方法,以抽取粒度较细的主题信息,并增加主题信息的置信度。该方法通过多个侧面对目标主题的衬托,采用LDA模型对主题信息进行建模,利用各个主题概率分布的平滑度进行候选句的可信度计算来抽取主题句。在面向Web的主题句抽取的具体应用中,取得了较好的效果。This paper proposes a novel topic sentence extraction approach based on model LDA on basis of acquiring candidate topic sentences through the topic-related and query-based keyword expansion. It can extract fine granularity on the subject and increase the reliability of the certain topical information, On several sub-topics against a target topic, it extracts those topic sen- tences by means of the reliability calculation according to the smoothness of the topic-sentence probability distribution. The method achieves good result in the special application of sentence extraction on web topic.

关 键 词：隐含狄利克雷分配(LDA) 主题模型 主题句抽取 信息融合 

分 类 号：TP391.12[自动化与计算机技术—计算机应用技术]