基于内容分析的话题检测研究.pdfVIP

38 10 V ol38 N o10 2 0 0 6 10 JOURNAL OF HARBIN INSTITUTE OF TECHNOLOGY O ct. 2006 赵华, 赵铁军, 张姝, 王浩畅 ( , 1 001, Em ail: huazhao@ m tlab. h it. edu. cn) : 通过分析大量的英文报道的特点, 针对目前话题检测研究中存在的难以区分两次 同的火车事故 或爆炸事件的问题提出了基于内容分析的话题检测算法. 该算法以 S ingle- Pass聚类策略为基础, 通过内容 分析将话题表示成两 个中心向量: 标识中心向量及内容中心向量. 实验证明基于内容分析的话题检测算法 但简单易行, 而且对于解决上述的难以区分问题非常有效. : 话题检测; 内容分析; 错误检测开销; 标识词; 内容词 : TP391 : A : 0367- 6234( 2006) 10- 1740- 04 Top ic detection research based on con ten t analysis ZHAO Hua, ZHAO T iejun, ZHANG Shu, WANG H aochang ( School of Com puter Science and Technology, H arb in Institute of Technology, H arb in 1 0001, China, Em ail: huazhao@ m tlab. hit. edu. cn) Abs tract: Based on the analysis of lots ofEnglish stories, w e propose a ContentAnalysis- based top ic detec tion algorithm, w hich ami s to solve the problem existing in the topic detection research, wh ich is d ifficu lt to detect two d istinct train or exp losion accidents as d ifferent events. Based on SinglePass clustering techn ique, using Content Analysis, the algorithm expresses topics as two centroids: identifier centroid and content cen troid. Expermi ent resu lts prove that ContentAnalysisbased topic detection algorithm is not only easy, but also effective on solving the d ifficu lttod istingu ish problem. K ey w ord s: top ic detection; content analysis; detection error cost; identifierw ord; content w ord (Top ic Detection and T rack ing, TDT) . [ 1- 4] . , : (), . . . [ ] . 1 基本概念 , , .