融合核心句与依存关系的评价搭配抽取  

Extraction of Evaluation Collection of Merging Kernel Sentence and Dependency Relation

作　　者：陶新竹[1,2] 赵鹏[1,2] 刘涛[1,2] 

机构地区：[1]安徽大学计算机科学与技术学院,安徽合肥230601 [2]安徽大学计算智能与信号处理教育部重点实验室,安徽合肥230601

出　　处：《计算机技术与发展》2014年第1期118-121,共4页Computer Technology and Development

基　　金：安徽省教育重点资助项目(KJ2009A001Z);安徽省重大科技专项资助项目(08010201002);安徽大学青年科学研究基金资助项目(2009QN004A)

摘　　要：评价搭配抽取是情感分析的基础任务之一。目前大部分抽取方法都是以依存句法分析为基础,但依存分析对中文评论文本的分析结果不稳定。针对此问题,提出了融合核心句抽取与依存关系的评价搭配抽取方法。该方法利用核心句抽取规则简化评论句结构,在此基础上进行依存句法分析,根据人工构建的依存关系模板进行评价搭配的抽取,并引入潜在评价搭配抽取规则抽取文本中省略评价对象的评价搭配。在中文酒店评论语料中进行试验,与基于依存分析的方法相比,该方法的F值提高约7%,证明了该方法的有效性。Extraction of evaluation collection is one of basic tasks in the area of sentiment analysis. Currently, most extraction methods are based on dependency parsing,but the result of dependency analysis is unstable. For this problem, a novel method merging kernel sentence and dependency relation is proposed. This method used kernel sentence extraction role to simplify the sentence structure, on this basis, did the dependency parsing and extracted evaluation collection according to artificially constructed dependencies template, in addition, intro- duced the extraction rules of implied evaluation collection to extract the evaluation collection which omitted the opinion object. Experi- mental results on Chinese hotel reviews domains show that compared with the method based on the dependency parsing, F-value in- creased by about 7% to verify the validity of the method.

关 键 词：核心句抽取 依存关系 评价搭配 潜在评价搭配 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]