词句协同自动摘要提取方法  

Method of automatic summarization algorithm based on word-sentence co-ranking

作　　者：吴云 杨长春[1] 梅佳俊 顾寰 WU Yun;YANG Chang-chun;MEI Jia-jun;GU Huan(School of Information Science and Engineering,Changzhou University,Changzhou 213164,China)

机构地区：[1]常州大学信息工程学院,江苏常州213164

出　　处：《计算机工程与设计》2018年第9期2776-2779,2810,共5页Computer Engineering and Design

基　　金：赛尔网络下一代互联网技术创新基金项目(NGII20160703)

摘　　要：为提高自动文摘的质量,提出一种词句协同的自动摘要提取算法(F-CoRank)。在传统词频的基础上,提高与标题相似的特征词的词频,得出提高后的词频矩阵和句子之间的相似度后,构建无向网络图,根据词句协同算法,得到各个节点的权重,对得到的粗文摘进行冗余处理,根据相应的需求,选择权重较高的前几个句子作为摘要。在哈工大的单文本文档语料上进行实验,实验结果表明,提高词频权重在一定程度上改进了文摘的质量,相比词句协同算法(Co-Rank)在覆盖率上有了较大提高。To improve the quality of automatic summarization,a word-sentence co-ranking based on the word frequency(F-CoRank)was proposed.Based on the traditional word frequency,the word frequency of the characteristic word which was similar to the title was improved,and after obtaining the improved word frequency matrix and the similarity between the sentences,the undirected network graph was constructed,and the weights of nodes were obtained according to the word-sentence co-ranking algorithm.The rough abstracts were redundantly processed,and the first few sentences with higher weights were selected according to the corresponding requirements.Experimental results on the HIT’s single document show that improving the frequency of word frequency exactly improves the quality of the abstract to a certain degree,and compared with the word-sentence co-ranking(Co-Rank),it improves the coverage rate.

关 键 词：词权重 标题 词句协同 单文本文档 覆盖率 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]