基于反馈报道的话题模型动态修正方法  

Dynamic topic model amending method based on feedback stories

作　　者：郑燕[1,2] 鲁燃[1,2] 赵爱华[1,2] 

机构地区：[1]山东师范大学信息科学与工程学院,济南250014 [2]山东省分布式计算机软件新技术重点实验室,济南250014

出　　处：《计算机应用》2012年第5期1343-1346,共4页journal of Computer Applications

基　　金：国家自然科学基金资助项目(60873247);山东省高新自主创新专项工程(2008ZZ28);山东省自然科学基金资助项目(ZR2009GZ007);济南市青年科技明星计划项目(20080201)

摘　　要：在话题追踪过程中,由于给定的初始话题相关报道少,而且话题具有动态演变的特点造成话题模型不准确。针对这一问题,提出了利用动态阈值收集反馈报道构造话题修正模型,实现了话题模型的动态修正;同时结合命名实体能够更加有效地区分不同话题的特性,提出了在修正话题模型时增大相关命名实体权重的方法,从而获得更准确的话题表示模型。实验结果表明,该方法能有效避免话题漂移现象,降低话题追踪过程中的漏报率和错报率。In topic tracking, the initial topic related stories are few and topic evolves dynamically, which leads to the topic model could not express topic accurately. Concerning this problem, it was proposed to build amended topic model by feedback stories collected by dynamic threshold, to amend topic model dynamically. And in combination of the feature that the named entity could differentiate different topics more effectively, it was suggested to increase the weight of named entity when amending topic model, to express a topic better. The experimental results indicate that, this method can solve the topic shifting problem effectively, and the miss tracking rate and fault tracking rate decrease a lot in topic tracking.

关 键 词：话题追踪 话题模型 动态阈值 命名 实体 反馈报道 

分 类 号：TP391.3[自动化与计算机技术—计算机应用技术]