基于TextRank的产品评论关键词抽取方法研究  

Research on Keyword Extraction Method for Product Reviews Based on TextRank

作　　者：尤苡名 YOU Yi-ming(School of Zhejiang Sci-Tech University,Information Academy,Hangzhou 310018,China)

机构地区：[1]浙江理工大学信息学院,浙江杭州310018

出　　处：《软件导刊》2020年第4期229-233,共5页Software Guide

摘　　要：关键词抽取技术能从海量产品评论文本中挖掘出用户关注的焦点,方便后续为用户推荐合适的产品。经典关键词抽取算法TextRank在迭代计算词汇节点的重要性得分时,忽略了邻近词汇节点的影响力差异。为此,提出一种融合TFIDF与TextRank算法(简称TFTR)抽取评论中的关键词。首先,通过引入用户浏览评论后给出的评论有用性反馈,提高有效评论中出现的重要词语权重,对TFIDF算法进行改进。然后将改进后的词频逆文档频率作为词节点特征权重引入到TextRank算法中,以改进词汇节点的重要性得分分配过程。实验结果表明,相比传统的TextRank算法,TFTR算法提取出的产品评论关键词准确性在P@10标准下提高了15.7%,证明了该算法的有效性。In order to recommend proper products for users,it’s essential to make use of keyword extraction techniques to mine out what users really focus on first. As a typical keyword extraction algorithm,the TextRank suffers from the prolbem that that it ignores the different importance between adjacent nodes when computing the score of origin vocabulary nodes iteratively. Based on TFIDF and TextRank algorithms,we propose TFTR model to extract keywords from reviews. First,we take account of the helpful feedback of reviews to improve TFIDF algorithm by raising the weight of important words. Then,the improved word frequency inverse document frequency is introduced to the TextRank algorithm as the feature weight of vocabulary nodes,which enhances the importance score distribution process. The experimental result shows that compared with the traditional TextRank algorithm,the accuracy of keywords extracted by TFTR algorithm is increased by 15.7% under the P@10 standard for product reviews. This result illustrates the effectiveness of our proposed algorithm.

关 键 词：关键词抽取 TFIDF TextRank TFTR 评论有用性反馈 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]