基于Gensim的摘要自动生成算法研究与实现  

RESEARCH AND IMPLEMENTATION OF GENSIM-BASED ABSTRACT AUTOMATIC GENERATION ALGORITHM

作　　者：肖元君 吴国文[1] Xiao Yuanjun;Wu Guowen(School of Computer Science and Technology,Donghua University,Shanghai 201600,China)

机构地区：[1]东华大学计算机科学与技术学院

出　　处：《计算机应用与软件》2019年第12期131-136,共6页Computer Applications and Software

基　　金：国家自然科学基金项目(61472075)

摘　　要：为了让计算机能够对中文文章提取摘要,提出一种中文摘要自动生成算法。该算法基于Gensim自然语言处理框架实现,并在原有的基础上做出了改进,算法主要分为两个阶段。关键句生成阶段,对中文语料进行预处理,并放入Gensim框架中的Word2vec模型进行训练,修改TextRank算法使其能够接受词向量的输入生成无向图从而找到关键句;摘要生成框架构建阶段,根据文章结构与Gensim框架中的LDA主题模型所提取的关键词,赋予句子不同的权值,将分数高的几个句子组合生成文章摘要。Rouge摘要评测结果表明,该算法生成的摘要能够包含文章关键信息,相比于其他自动文摘算法,句意通顺程度得到了提升。In order to extract abstracts from Chinese articles by computer, this paper presents a Chinese abstracts automatic generation algorithm. The algorithm was realized based on natural language processing framework Gensim, and improved on the original basis. The algorithm was mainly divided into two stages. In the key sentence generation stage, the Chinese corpus was preprocessed and then put them into the Word2vec model of Gensim framework for training. The Textrank algorithm was modified to accept the input of word vector and generate undirected graph to find the key sentence. In the construction stage of abstracts generation framework, according to the keywords extracted from the article structure and the LDA topic model in the Gensim framework, the sentences with different weights were given, and several sentences with high scores were combined to generate the article abstracts. The evaluation results of Rouge show that the abstracts generated by the algorithm can contain the key information of the article,and the sentence meaning is more smooth than other automatic abstract algorithms.

关 键 词：Gensim框架 Word2vec模型 TextRank算法 摘要生成框架 LDA主题模型 Rouge摘要评测 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]