基于主题模型的汉语动词隐喻识别  

Chinese Verb Metaphor Recognition Based on Topic Model

作　　者：白振凯 黄孝喜[1] 王荣波[1] 谌志群[1] 王小华[1] 

机构地区：[1]杭州电子科技大学认知与智能计算研究所,浙江杭州310018

出　　处：《计算机技术与发展》2016年第11期67-71,76,共6页Computer Technology and Development

基　　金：国家自然科学基金资助项目(61103101;61202281);教育部人文社会科学研究项目(10YJCZH052)

摘　　要：隐喻是人类语言不可缺少的组成部分,隐喻处理的好坏将直接影响到自然语言处理和机器翻译的发展,其中隐喻识别作为隐喻处理中基础性的工作,越来越得到研究者们的关注。目前,汉语隐喻识别的研究大部分都集中在短语级别的名词性隐喻的识别上,然而,实际文本中动词性隐喻出现的频率更高,更应该受到更多中文隐喻研究者们的重视。为了提高汉语隐喻的识别率,针对句子级别的汉语动词性隐喻,提出了基于主题模型的识别方法,将主题模型LDA(Latent Dirichlet Allocation)应用于汉语的动词隐喻识别过程中。该方法利用句子的主题分布作为特征,结合机器学习的方法对动词隐喻进行识别,得到的平均正确率为76.46%,在加入主题标注特征后,平均正确率达到80.42%。实验结果表明,基于主题模型的识别方法是有效的。Metaphor is an integral part of human language,and the quality of metaphor processing will directly affect the effectiveness of natural language processing and machine translation. Metaphor recognition is an essential task in metaphor processing as a foundational work and has got the attention of the researchers. At present,most Chinese metaphors recognition has focused on identifying the phrase level of noun metaphor, however, verbal metaphors has higher frequency in the actual text, which should be paid attention by more Chinese metaphor researchers. In order to improve the recognition rate of Chinese metaphor, in view of the Chinese verb metaphor, an approach to metaphor recognition is proposed based on topic model. In this method, sentence topic distribution generated through LDA mod- el is used as a feature, and the metaphor recognition is implemented with SVM. The average accuracy of the method is 76.46%, after further joined the feature of topic annotation,the average accuracy of the method is 80.42%. The experimental results show that the method is effective.

关 键 词：隐喻识别 主题模型 LDA 机器学习 自然语言处理 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]