基于概率统计与图的多文档自动文摘研究.pdfVIP

lA 计算机科学2007V01．34No．1 基于概率统计和图的多文档自动文摘研究 刘量谌卫军王建民 (清华大学软件学院 北京100084) 摘要本文介绍了一种新的多文档自动文摘方法。该方法的基本思想是通过估计词在文摘中出现的概率采选 择最能反映多文档集信息内容的核心词，然后根据该概率计算核心词的权值，最后将整个文档集合中的句子及其 信息内容重叠关系表示为一个无向有权图，从中连选出丈摘句。本文利用DUC2003和DUG2004的敦裾集对该方 法进行评测。测试表明，该方法所产生的文摘质量优于其它的一些方法比如LexRank 关键词弧IDF，Graph，LexRank，DUC ResearchOil Document on SummarizationBased Modeland Multiple Probability Graph LIU CHENWei-JunWANG Liang Jian-Min of (School Software。TsinghuaUniversity，Beijing100084) AbstractThe anovel tO a from documents．The paperpresents approachautomaticallyproducesummarymultiple firSt ofour istOselectasetofcentralfeatureswhichcail themainthemeofthedocuments． step approach represent T}地selectiondecisionsaremade thefeatures’distributionsinthefuture the byconsidering summary．1rhenweight of铭chfeatureis whichilleasLlre喳itscontributiontOthe of丑sentenceincludedinthe computed possibility being allundirectedisconstructedto thesentencesetandUSedforthe sentence summary．k峙tly graph represent sulYRnal7 extraction．ThemethodwasevaluatedontheDUCdatasetsof2003and resultsshow proposed 2004．Experimental thatourmethodoffers tOothersummarizersuchasLexRank． promisingperformancecompared systems KeywordsTF．正)F，Graph，LexRank．DUC 别原文档集中最重要的信息，然后利用自