基于条件随机场的商品评论信息抽取研究

CRFs-based Information Extraction of Product Reviews

采用条件随机场(CRFs)算法,以商品属性为中心,挖掘出消费者对商品的情感观点以及观点态度的强弱。通过对商品评论进行标注学习,实现了商品属性和相应的评价词的自动抽取,从而识别出评论文本中的关键信息。研究中抽取的三个维度的关键信息包括商品特征属性,与之相关的评论情感观点,以及情感程度的强弱。仿真实验表明,借助词本身和词性特征,以及上下词的位置关系特征,CRFs算法对商品评论信息抽取有着较高的查准率和召回率。

Focusing on commodity attributes, this paper adopts conditional random fields （CRFs） method to mine the consumerst emotional point of view and the attitude strength toward products. Through the learning of manual tags on product reviews, the automatic extraction of commodity attributes and corre- sponding assessments was realized. On this account, the key information in product reviews was identified. The three dimensions of key information in this study include attributes of goods, and related comments emotional point of view, and the degree of strength of emotion. Simulation results show that, with the help of the word itself and part of speech features, as well as the characteristic of the relationship between the location of the word, CRFs method has a higher precision and recall rate of information extraction on product reviews.