【作者】 贾候萍； 万小军； 黄小江； 杨建武； 肖建国；

【Author】 Houping Jia Xiaojun Wan Xiaojiang Huang Jianwu Yang Jianguo Xiao Institute of Computer Science and Technology,Peking University,Beijing 100871

【机构】 北京大学计算机科学技术研究所；

【摘要】 多文档摘要系统中旬子捧序问题一直是文档自动摘要系统中一项重要而艰巨的任务。传统的摘要句排序方法假设各句子在语义上是相互独立的,只简单利用句子的时间特征和在原文档中的位置特征来排序,效果较差。为了获取两个句子之间的语义先后次序,我们定义了一套排序标准。基于这套标准,反复地比较两个句子或两个文本段之间的次序关系,构成一个新文本段,直到我们得到包含所有摘要句的文本段。通过与其它两种传统排序方法比较实验,此方法与人工排序方法的结果最接近,证明了该方法的有效性。更多还原

【Abstract】 Sentence ordering is an important but difficult task for multi.document summarization.Traditional sentence ordering methods usually assume sentences to be semantically independent and make only use of the time feature or the position feature in their original documents,which usually leads to unsatisfactory results.To capture the semantic-level order information between two sentences,we defme a set of ordering criteria.We iteratively concatenate two sentences or two text segments into a new text segment based on these criteria,until we obtain the total segnient containing all sentences to he arranged.Compared with other two traditional ordering methods,the proposed method’s result is most correlated with human-labeled result,which demonstrates the good effectiveness of our proposed method.更多还原