基于万有引力改进的TextRank关键词提取算法  

AN IMPROVED TEXTRANK KEYWORD EXTRACTION ALGORITHM BASED ON GRAVITY

作　　者：孙福权[1,2] 张静静[2] 刘冰玉[1,2] 姜玉山[1,2] 多允慧 Sun Fuquan;Zhang Jingjing;Liu Bingyu;Jiang Yushan;Duo Yunhui(Notheastern University at Qinhuangdao,Qinhuangdao 066004,Hebei,China;Notheastern University,Shenyang 110819,Liaoning,China)

机构地区：[1]东北大学秦皇岛分校,河北秦皇岛066004 [2]东北大学,辽宁沈阳110819

出　　处：《计算机应用与软件》2020年第7期216-220,295,共6页Computer Applications and Software

基　　金：国家重点研发计划项目(2018YFB1402800);教育部科技发展中心科研创新项目(2018A03031);全国教育信息技术研究规划课题重点项目(16222874);医学影像智能计算教育部重点实验室资助项目。

摘　　要：为了提高文本关键词提取的准确性,提出基于万有引力改进的TextRank关键词提取算法GtextRank。利用万有引力模型对词语在文档中的主题影响力、词语间距离和词语间共现频率进行有效融合,构建新的TextRank转移概率实现关键词的提取。实验结果表明,与传统关键词提取方法相比,该算法具有显著的优越性,能够完成对关键词的相对正确的提取;同时考虑了文本中词语的语义关系和主题影响度,可以提高关键词的提取精度。In order to improve the accuracy of text keyword extraction,we propose an improved TextRank keyword extraction algorithm GtextRank based on universal gravitation.The universal gravity model was used to effectively fuse the theme influence,the distance between words and the co-occurrence frequency of words in documents,and a new TextRank transition probability was constructed to extract keywords.The experimental results show that compared with the traditional keyword extraction method,our algorithm has significant advantages and can complete the relatively correct extraction of keywords.It shows that the accuracy of keyword extraction can be improved by considering both the semantic relationship of words and the degree of topic influence.

关 键 词：关键词 主题影响度 词向量 TextRank 万有引力 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]