2023年2月25日第7卷第4期现代信息科技ModernInformationTechnologyFeb.2023Vol.7No.436362023.022023.02收稿日期：2022-10-13基金项目：甘肃省农业大学盛彤笙科技创新基金（GSAU-STS-2021-15）；国家自然基金（32060437）；甘肃农业大学省级大学生创新创业训练计划项目（202216018）基于Word2Vec及TextRank算法的长文档摘要自动生成研究朱玉婷，刘乐，辛晓乐，陈珑慧，康亮河（甘肃农业大学，甘肃兰州730070）摘要：近年来，如何从大量信息中提取关键信息已成为一个急需解决的问题。针对中文专利长文档，提出一种结合Word2Vec和TextRank的专利生成算法。首先利用PythonJieba技术对中文专利文档进行分词，利用停用词典去除无意义的词；其次利用Word2Vec算法进行特征提取，并利用WordCloud对提取的关键词进行可视化展示；最后利用TextRank算法计算语句间的相似度，生成摘要候选句，根据候选句的权重生成该专利文档的摘要信息。实验表明，采用Word2Vec和TextRank生成的专利摘要质量高，概括性也强。关键词：Jieba分词；关键词提取；Word2Vec算法；TextRank算法中图分类号：TP391.1文献标识码：A文章编号：2096-4706（2023）04-0036-04ResearchonAbstractAutomaticGenerationofLongDocumentBasedontheWord2Vec+TextRankAlgorithmZHUYuting,LIULe,XINXiaole,CHENLonghui,KANGLianghe(GansuAgriculturalUniversity,Lanzhou730070,China)Abstract:Inrecentyears,howtoextractcriticalinformationfromlargeamountsofinformationhasbecomeaproblemwhichneedstobesolvedurgently.ForChinesepatentlongdocuments,apatentgenerationalgorithmcombiningWord2VecandTextRankisproposed.Firstly,PythonJiebatechnologyisusedtosegmentwordsinChinesepatentdocuments,andmeaninglesswordsareremovedbyusingthestopdictionary.Secondly,theWord2Vecalgorithmisusedforfeatureextraction,andtheextractedkeywordsarevisuallydisplayedbyWordCloud.Finally,theTextRankalgorithmisusedtocalculatethesimilaritybetweensentences,generateabstractcandidatesentences,andgenerateabstractinformationofthepatentdocumentsaccordingtotheweightofcandidatesentences.ExperimentsshowthatpatentabstractsgeneratedbyWord2VecandTextRankareofhighqualityandhavestronggeneralization.Keywords:Jiebawordsegmentation;keywordextraction;Word2Vecalgorithm;TextRankalgorit...
