基于句法分析的评价搭配抽取及其倾向性分析

Polarity Collocation Extraction and Sentiment Classification Based on Syntactic Analysis

提出一种新的基于句法分析的方法可以提取出评价搭配。通过对大量的商品评论进行句法分析,得到了常用的评论模式集合。依靠这些模式对评价搭配进行抽取。同时,构建了一种映射集合,可以映射出隐式评价对象。时于评价对象的倾向性判别,通过hownet构建了情感种子词典,采用SO-PMI方法计算出其极性,并计算出极性强度。实验结果表明,该方法对于评价搭配的抽取和极性分析具有较好的精确率和召回率。

A new method is proposed to extract polarity collocation. Series of templates are generalized using syntactic analysis on lots of prod- uct reviews. Then polarity collocation is extracted via the templates. In order to get implicit comment target, a mapping of comment is built tar- geting onto reviews. At the same time, a seed dictionary is built based on hownet, with SO-PMI used to compute the polarity and the strength. Experiment results show that the method can get a better precision and recall.