基于增强输出和同义拓展的无监督关键词抽取模型构建  

Construction of Unsupervised Keyword Extraction Model Based on Output Enhancement and Synonymous Expansion

作　　者：莫纯扬 MO Chunyang(School of Information Management,Sun Yat-sen University,Guangzhou 510006,China)

机构地区：[1]中山大学信息管理学院,广东广州510006

出　　处：《晋图学刊》2025年第1期1-9,18,共10页Shanxi Library Journal

基　　金：广东省基础与应用基础研究基金重点项目“面向东盟国家的多语言信息处理关键技术研究”(项目编号:2019B1515120085)。

摘　　要：文章构建了基于增强输出和同义拓展的无监督关键词抽取模型KEOE(Keyphrase Extraction Output Enhancer,增强输出器),用于提高无监督关键词抽取模型的可靠性与可用性。首先,以现具有无监督关键词抽取模型的YAKE!、KP-Miner和MultipartiteRank为对象,融合多模型结果进行输出优化;其次,结合WordNet同义词知识库对关键词进行同义词拓展,扩大关键词输出的语义范围;最后,在前述基础上搭建一套可复用的无监督关键词抽取算法。实验结果表明,利用KEOE模型在各实验数据集的平均抽取准确率达到最优,证明了该模型的有效性。This paper presents an unsupervised keyphrase extraction model named KEOE(Keyphrase Extraction Output Enhancer),which is designed to enhance the reliability and usability of unsupervised keyphrase extraction models.Initially,the model integrates the results of existing unsupervised keyphrase extraction models such as YAKE!,KP-Miner,and MultipartiteRank to optimize the output.Subsequently,it leverages the WordNet thesaurus to expand the keyphrases with synonyms,thereby broadening the semantic scope of the keyphrases.Finally,a reusable unsupervised keyphrase extraction process is constructed based on the aforementioned approach.Experimental results demonstrate that the KEOE model achieves the best average extraction precision across various experimental datasets,thereby validating the effectiveness of the model.

关 键 词：无监督学习 关键词抽取 增强输出 同义拓展 

分 类 号：G350.7[文化科学—情报学]