《文本挖掘技术》课堂报告-万小军20080522.pdf

《文本挖掘技术》课堂报告 万小军 万小军 wanxiaojun@ wanxiaojun@ May 23, 2008 May 23, 2008 主要内容 主要内容 小组工作介绍 小组工作介绍 SIGIR2007论文 SIGIR2007论文 SIGIR2008论文 SIGIR2008论文 摘要方向展望 摘要方向展望 一些体会和建议 一些体会和建议 小组工作介绍（略） 小组工作介绍（略） SIGIR2007 Paper： CollabSum: Exploiting Multiple Document Clustering for Collaborative Single Document Summarizations Outline Introduction Related Works Collaborative Summarization Experiments and Results Conclusions and Future Work Introduction(1/2) What is summarization? Document (set) - 〉concise description; Single vs. Multi-document; Generic vs. Topic-focused; Single document summarization 只利用了单文档内部的信息; 假定文档之间是相互独立的; 单文档摘要过程没有考虑文档之间的相互作用； Introduction(2/2) Motivation 文档类簇中文档之间的相互影响; 对句子评价过程中更多的相关文档能够提供更 多的知识和线索; Collaborative summarization 利用多个文档改善单文档摘要; Related Works(1/3) Summarization methods: Abstraction vs. Extraction; Extraction-based methods Rank sentences or passages; Features: Term frequency; Sentence position; Cue words; Stigma words; Topic signature; Related Works(2/3) Extraction-based methods (cont’d) Machine learning classification, clustering, HMM, CRF, Mutual reinforcement learning; IR techniques MMR, LSA; Graph-ranking TextRank [Mihalcea and Tarau 2004，2005] LexPageRank [ErKan and Radev, 2004 ]; Related Works(3/3) Limit