基于领域知识的图模型词义消歧方法  

Word Sense Disambiguation with Graph Model Based on Domain Knowledge

作　　者：鹿文鹏[1,2] 黄河燕[1] 吴昊[1] 

机构地区：[1]北京理工大学计算机学院 北京市海量语言信息处理与云计算应用工程技术研究中心,北京100081 [2]齐鲁工业大学理学院,济南250353

出　　处：《自动化学报》2014年第12期2836-2850,共15页Acta Automatica Sinica

基　　金：国家重点基础研究发展计划(973计划)(2013CB329303);国家自然科学基金(61132009);山东省高等学校科技计划(J12LN09)资助~~

摘　　要：对领域知识挖掘利用的充分与否,直接影响到面向特定领域的词义消歧(Word sense disambiguation,WSD)的性能.本文提出一种基于领域知识的图模型词义消歧方法,该方法充分挖掘领域知识,为目标领域收集文本领域关联词作为文本领域知识,为目标歧义词的各个词义获取词义领域标注作为词义领域知识;利用文本领域关联词和句子上下文词构建消歧图,并根据词义领域知识对消歧图进行调整;使用改进的图评分方法对消歧图的各个词义结点的重要度进行评分,选择正确的词义.该方法能有效地将领域知识整合到图模型中,在Koeling数据集上,取得了同类研究的最佳消歧效果.本文亦对多种图模型评分方法做了改进,进行了详细的对比实验研究.Whether domain knowledge is fully utilized would impact the performance of word sense disambiguation （WSD） on a specific domain. A WSD method with graph model based on domain knowledge is proposed in the paper. The method makes full use of domain knowledge： first, the keywords related with target text domain are collected as text domain knowledge, and domain annotations of each sense of target ambiguous word are obtained as sense domain knowledge; second, a disambiguation graph is constructed with text domain knowledge and sentence context words~ thirdly, the disambiguation graph is adjusted based on sense domain knowledge; finally, the sense nodes in the graph are scored with an improved evaluation method to judge the right sense. This WSD method effectively integrates domain knowledge with graph model. Evaluation is performed on Koeling dataset. Compared with similar methods, the WSD method yields state-of-the-art performance. Besides, multiple graph evaluation models are improved and compared in detail.

关 键 词：词义消歧 领域知识 图模型 词义领域 文本领域 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]