基于WordNet的无导词义消歧方法  

Unsupervised word sense disambiguation based on WordNet

作　　者：王瑞琴[1,2] 孔繁胜[1] 潘俊[1] 

机构地区：[1]浙江大学人工智能研究所,浙江杭州310027 [2]温州大学计算机科学与工程学院,浙江温州325035

出　　处：《浙江大学学报（工学版）》2010年第4期732-737,共6页Journal of Zhejiang University：Engineering Science

摘　　要：有导词义消歧机器学习方法由于需要大量人力进行词义标注,难以适用于大规模词义消歧任务.提出一种避免人工词义标注的无导消歧方法.该方法综合利用WordNet知识库中的多种知识源(包括:词义定义描述、使用实例、结构化语义关系、领域属性等)描述歧义词的词义信息,生成词义的"代表词汇集"和"领域代表词汇集",结合词汇的词频分布信息和所处的上下文环境进行词义判定.利用通用测试集Senseval-3对6个典型的无导词义消歧方法进行开放实验,该方法取得平均正确率为49.93%的消歧结果.Word sense disambiguation （WSD） based on supervised machine learning is hard to deal with large-scale WSD because of its big labor cost.To solve this problem,an unsupervised WSD method was provided,which describes the word senses of an ambiguous word via synthesizing multiple knowledge sources in WordNet ontology,including definition glosses,samples,structured semantic relations,domain attributes,etc.From the description,a representative glossary and a domain representative glossary are deduced.The two structures together with the word sense frequency distribution and the context are used for WSD.The average disambiguation accuracy was 49.93% by this method in open test for six representative unsupervised WSD methods with Senseval-3 English lexical sample data set.

关 键 词：词义消歧 WordNet知识库 结构化语义关系 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]