基于层叠条件随机场的微博商品评论情感分类

商品评论是消费者针对某一个商品的主观议论。针对微博中商品的评论文本短小、结构多样等特征,在仅使用现有的微博级情感标注的条件下,提出了一种基于层叠条件随机场模型。以中文小句中枢说为理论基础,将商品评论的句子划分为若干小句,使用微博内小句序列的各种特征训练粗粒度的随机条件场情感分类模型,同时使用小句内汉字序列的各种特征来训练细粒度的随机条件场情感分类模型。实验结果表明,本文提出的方法优于传统的情感分类方法。

Product reviews sentiment classification in Micro-blog based on cascaded conditional random field

Product reviews are subjective comments submitted by customers. Nowadays, product reviews are in the form of Micro-blog text which is typically very short but with varied structures. We proposed a novel sentiment classification method for product reviews from Micro-blog based on cascaded Conditional Random Field(CRF). First, review sentences were divided into a number of clauses based on the theory of clausal pivot. Then, features of the Chinese clause sequences were exploited to train a coarse-grained CRF sentiment classification model. Meanwhile, features of the Chinese character sequences within clauses were exploited to train a fine-grained CRF sentiment classification model. The experimental evaluation shows that the proposed method is better than the state-of-the-art ones.