一种基于自适应重心向量的主题检测方法  

Topic Detection Approach Based on Adaptive Center Vector

作　　者：潘渊[1] 李弼程[1] 张先飞[1] 

机构地区：[1]解放军信息工程大学信息工程学院,郑州450002

出　　处：《计算机工程》2009年第3期80-82,共3页Computer Engineering

基　　金：国家"863"计划基金资助项目(2007AA01Z439)

摘　　要：针对影响主题检测性能的2个重要因素——相似主题的判定和主题漂移问题,提出一种基于自适应重心向量的主题检测方法。该方法将命名实体信息应用到特征表示上,将命名实体向量和关键词向量相结合表示主题的重心向量,以有效区分相似主题。采用增量聚类检测主题,在增量聚类过程中不断修正主题重心,以解决主题漂移的问题。实验结果与性能比较表明,该方法能有效提高主题检测的性能。Similar topic detection and topic excursion are two important factors which affect the performance of topic detection. For these two problems, this paper proposes a topic detection approach based on adaptive center vector. By using information of name-entity in feature representation, it combines name-entity vector and keyword vector to construct topic center vector, which can detect similar topic efficiently. Based on the idea of single-pass clustering, the algorithm modifies topic center dynamically. Experimental results show that the algorithm can improve the performance of topic detection effectively.

关 键 词：主题检测 主题漂移 命名实体 主题重心向量 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]