一种实例库与义原关系相结合的概念消歧算法  

A Concept Disambiguation Algorithm Based on the Combination of Instance Base and Sememe Relations

作　　者：高璐[1] 赵小兵[1,2] 

机构地区：[1]中央民族大学信息工程学院,北京100081 [2]国家语言资源监测与研究中心少数民族语言分中心,北京100081

出　　处：《首都师范大学学报（自然科学版）》2016年第3期7-10,共4页Journal of Capital Normal University：Natural Science Edition

摘　　要：本文提出了一种利用知网的实例库与知网关系进行词义消歧算法.该方法首先利用知网提供的实例库进行初步的匹配;若在实例库中没有完全匹配,则利用上下文搭配关键词与知网中的实例搭配词进行相似度计算,若相似度大于给定阈值,则消歧结束.否则,我们再判断歧义词的义原与关键词的义原是否具有某种关系,根据义原权值调节算法调整义原权值.调整后的义原权值大小不一,按照事先的约定,我们选取综合权值最大的义项.我们发现,该方法能够弥补仅依靠实例库的覆盖率低的问题,又能减少仅依靠统计方法产生的噪音,从而提高词义消歧的正确率.This paper proposes a WSD algorithm based on How Net. The method firstly uses case base that How Net provides for preliminary matching; if the case library do not exactly match,then we compute the similarity between the context matching keywords and How Net examples collocations. If the similarity is greater than a given threshold,then the end of disambiguation. Otherwise,we judge that if there has some relations between sememes of keywords and sememes of case library and adjust the original weight value according to the original weight adjustment algorithm. After the adjustment,the weights of sememes are not the same size. In accordance with the prior agreement,we selected the largest comprehensive weight items. We find that this method can make up for the problem that only depend on the low coverage of the instance library and can also reduce the noise generated by the statistical method,thus improving the correct rate of word sense disambiguation.

关 键 词：HOWNET 实例库 关系网络 权重调节 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]