基于多词典融合的词汇语义倾向判别  

Lexical Semantic Tendency Determination Based on Multi-dictionary Strategy

作　　者：刘清松[1] 张仰森[1] 

机构地区：[1]北京信息科技大学人工智能实验室,北京100192

出　　处：《计算机技术与发展》2015年第5期104-109,共6页Computer Technology and Development

基　　金：国家自然科学基金资助项目(61370139);北京市属高等学校创新团队建设与教师职业发展计划项目(IDHT20130519);北京市教委专项基金(PXM2013_014224_000042;PXM2014_014224_000067)

摘　　要：文本情感倾向性判别是情感分析的重要组成部分,而精确的词汇情感计算是文本情感倾向性判别的基础。基于词汇知识库的情感词倾向判别计算引起了学者们的广泛关注与研究。文中融合国内知名的三大词典:How Net、同义词词林、情感词汇本体,重新对基准词对做进一步的归纳与总结,从不同的角度构建三类等价情感倾向集合并提出两种处理集合的策略,建立了待定情感词与特定等价情感倾向集合的情感映射关系。实验结果表明:该方法获得的最高准确率可达91.62%,平均正确率85.31%,符合人们对词语的情感倾向认识。Text sentiment orientation analysis is an important part of textual emotion classification,and to determine the emotional bias for words precisely is the very foundation of text sentiment orientation analysis. Nowadays,dictionary-based methods of sentiment orientation computing for words have been widely concerned and researched. In this paper,summarize and conclude benchmark words considering knowledge from HowNet,Tongyici-Cilin and affective lexicon ontology,then build three emotion-equal sets from different perspective and propose two strategies of set processing,finally establish the mapping between the sentiment orientation of a word to be computed and a specific emotion-equal set. Experimental results has generally agreed with the sentiment orientation felt by human,and achieved maxi-mum correctness 91. 62% and 85. 31% in average.

关 键 词：HOWNET 情感词汇本体 同义词词林 等价情感倾向集合 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]