基于网络化数据挖掘策略的中文多文档自动文摘研究.pdfVIP

基于网络化数据挖掘策略的中文多文档自动文摘研究 胡珀 何婷婷 张勇 (华中师范大学计算机科学系武汉430079) (phu宜mail．ccnu．edu．cn) 摘要多文档自动文摘是近期自然语言处理领域的一项热点研究课题．文中提出了一种基于网络化 数据挖掘策略的中文多文档自动文摘方法，该方法的特色在于生成的摘要具有全面而简洁的特性，其内容 能同时获得主题覆盖和冗余之间的平衡。考虑到同一主题的文档集合其潜在子主题的分布往往符台段落密 度特性，提出了以段落为中心的研究方法．利用网络模型来表达段落并模拟段落间的语义相关性联系，利 用网络化数据挖掘策略来自适应地划分段落语义网络中的各抱团结构并依此识别文档集合中的各潜在子主 题，通过对子主题中心句子单元的抽取及排列生成最终的摘要．此外，一种基于句子向量相似度的方法被 用来自动评价生成摘要的覆盖度和冗余度，通过与几种典型的多文档摘要方法的实验比较及对自动评价和 人工评价结果的统计相关性分析验证了该方法的可行性和有效性． 关键词多文档自动文摘网络化数据挖掘子主题识别 Chinese Multi．document Summarization Based on Networked Data Mining Strategy Po Hu Tingting He Yong Zhang (DepartmentofCompu忙rScience,HuazhongNormalUniversity,Wuhan，430079，China) Abstract Multi-documant summarization bacom％a hot topic in recant natural language processing research nIis paper proposes a novel approach for Chinese multi-document summarization based 011 networked data mining strategy．Thc characteristic of the method is that the generated summary Can be both complete and brief：It can cOVeT diverse subtopics and reduce its rcdund¨cy obviously at the same time．To maintain the summary content’S balance betwcan cuveragc and redundancy,a paragraph similarity nctwonk is constructed to model the semantic relationship among all paragraphs in a Chinese document set with same topic and a networked data mining stralc酣is cmploycd to adaptively analyze the network and antomatically detect the different communities for subtopics discovery,then the centroid sentence is extracted from different subtopic and is ordercd to generate final summary．In addition．an evaluation method based OU santance-vectot similarity calculation is used for automatic summarization evaluation．Experimental results and comparisons with ot}lcr typical methods indicate that this method iS effective and e柏ciant． 基盒项目：强家自然科学基金：国家社会科学基金(06BYY029)；教育部科学技术研究重点项目基iz(iosll71 —36l一 Key words multi—document summarization,networked data mining,subtopic discovery 1前言 随着全球信息化进程的加快及信息竞争的趋势日盛，搜索浏览各类需求信息已成为当今 人们日常生活的必需成分。然而，不断增加的海量信息却导致了信息资源的丰富化与信息获 取的