ISSN 1000-1239 CN 11-1777/TP

计算机研究与发展 ›› 2015, Vol. 52 ›› Issue (11): 2441-2450.doi: 10.7544/issn1000-1239.2015.20140583

• 人工智能 • 上一篇    下一篇



  1. (语言语音教育部-微软重点实验室(哈尔滨工业大学) 哈尔滨 150001) (
  • 出版日期: 2015-11-01
  • 基金资助: 

Dynamic Incremental Analysis of Sub-Topic Evolution

Li Fenghuan, Zheng Dequan, Zhao Tiejun   

  1. (MoE-MS Key Laboratory of Natural Language Processing and Speech (Harbin Institute of Technology), Harbin 150001)
  • Online: 2015-11-01

摘要: 事件发展的持续性和相互影响性使人们对事件的后续进展越来越感兴趣,而传统的事件分析大多是针对基于句子的事件.针对专题事件,结合single-pass聚类方法、兼类思想以及动态增量思想,提出了一种增量式子主题动态演化分析模型.该模型基于专题事件的时序特征提出,包括动态阈值的设定、相似度平滑、子主题动态增量策略等过程,以及运用χ\+2统计的思想来综合评价模型性能的方法.该模型可以有效地对专题事件进行子主题分析,进而使人们能够更直接和快速地了解主题事件的进展.实验结果表明提出的方法使子主题演化分析的性能有了显著的提高.

关键词: 演化分析, 子主题, 动态, 增量式, 主题事件

Abstract: There has been increasing interest in the follow-up progress of events because of sustainability and mutual influence of events. Meanwhile, more and more emergent events make it necessary to follow events in an intuitive and efficient way. However, the majority of traditional event analysis is sentence-oriented or topic-oriented which is event extraction or topic detection and tracking. A hierarchical structure of the topic event is constructed according to the research objects and the scope. A dynamic incremental model is presented for analyzing sub-topic dynamics in the topic event, which borrows the ideas of single-pass clustering, multi-category and dynamic incremental model. It is document-oriented and built on temporal property of the topic event, including dynamic threshold selection, similarity smoothing and dynamic incremental strategy. Meanwhile, overall evaluation criteria combinied with χ\+2-test is served for performance analysis. The algorithm is effective and facilitates users to follow the topic event explicitly. Experimental results reported for four well-known topic events in China show that the performance of sub-topic evolution analysis is improved significantly.

Key words: evolution analysis, sub-topic, dynamic, incremental, topic event