基于句子评分的多文档摘要提取  

Multi-document Summarization Based on Sentence Score

作　　者：张世博[1] 刘博爱 

机构地区：[1]北京邮电大学计算机学院,北京100876 [2]北京石油化工学院计算机系,北京102600

出　　处：《山西大学学报（自然科学版）》2015年第3期406-412,共7页Journal of Shanxi University(Natural Science Edition)

基　　金：北京市自然科学基金(4142015);北京石油化工学院国家级大学生创新训练计划资助(No.2015J00093)

摘　　要：技术越来越得到关注,文章设计了从多文档中生成摘要的方法。基于多种语义特征和语义相似度计算句子的重要性评分值,使用stack decoder算法利用多个堆栈生成摘要;基于分团问题的生成方式构建了句子的非相似性图,通过求解非相似图中有边的子集合顶点选取合适的句子。实验以DUC2004语料集为数据集,通过上述方法产生摘要,ROUGE评测指标达0.069 7,与已有方法相比在ROUGE值上有明显提高,并对归一化方法进行了讨论,探索了适用的归一化措施。Automatic summarization technology is getting more and more attention,so we designed methods for generating summarization from multi-document.First,the importance scores are computed based on various semantic features and a semantic similarity score.We use stack-decoder algorithm as a template and build on it to produce summaries.Second,summaries are generated based on the cliques found in the constructed graph.The graph is generated by building edges between sentences which talk about similar topics but not similar semantic.These approaches are used to generate summaries for the DUC 2004 dataset.ROUGE score is 0.069 7which has significant improvement comparable to present method.We also discussed different normalized methods and designed the suitable normalized function.

关 键 词：文档摘要 语义相似度 语义特征 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]