面向互联网评论情感分析的中文主观性自动判别方法研究

Automatically Measuring Subjectivity of Chinese Sentences for Sentiment Analysis to Reviews on the Internet

作为非结构化信息挖掘的一个新兴领域,网络评论情感分析引起了人们的极大兴趣。利用对互联网上客户评论信息的挖掘与分析结果,消费者可以了解其他用户的态度倾向分布,做出更好的购买决策。销售商和生产商可以获得用户对其商品和服务的反馈,了解用户对自己和对竞争对手的评价,从而改进产品、改善服务,获得竞争优势。实现情感分析的前提是有效识别文本中表达主观感情、态度和观点的内容,对文本中的主观性成分进行判断。目前,这一领域的研究主要针对英文评论进行。随着中文网络信息的不断发展,如何对中文网络评论进行自动情感分析已经成为一个迫切需要解决的问题。中文评论情感分析中的一个基础性的问题就是如何对中文评论的主观性进行判断。本文提出了一种根据连续双词词类组合模式(2-POS)自动判断句子主观性程度的方法。在对主观句与客观句的分类试验中,两类语句的查准率和查全率都已接近目前英文同类研究的结果,初步表明了该方法的可行性。

As a new domain of unstructured data mining,online sentiment analysis has aroused great interest recently. Through automatically mining online reviews to certain products,consumers could know the distribution of attitudes of other consumers to this product before making a buying decision.Meanwhile,manufacturers and retailers would get consumers’feedback about their products or services,as well the opinion of customers to their competitors,which is useful for them to improve the products or services and gain competitive advantages.A crucial step before sentiment analysis is to identify subjective expressions in the context.Subjective sentences are usually the parts expressing sentiment,attitude or opinions in text.The existing researches on sentiment analysis mainly focus on English reviews. Few studies have been conducted to Chinese texts.As Chinese information has increased dramatically in cyber space, how to automatically analyze opinions of reviews in Chinese on the Internet has become urgent.One basic and important task is to establish a method to identify subjective expressions in Chinese reviews.This paper proposed an approach to measure subjective strength of Chinese sentences using patterns of continuous two words combination,2-POS model.In experiments of subjectivity classification to Chinese sentences,both subjective and objective sentences achieved high precision and recall.The results show that the performances of the proposed approach are comparable to existing studies in English.