基于正则化主题建模的隐式产品属性抽取

为实现产品意见挖掘中的隐式产品属性抽取,在传统主题建模思想的基础上,通过分析评论信息中不同产品属性对应的意见词分布以及意见词的主题依赖性假设,提出一种基于正则化思想的新主题建模框架.在该框架下,评论信息中的意见词特征,通过定义在不同评论中意见词的使用相似度上的正则化因子,纳入到传统的主题建模框架中.正则化的基本思想为：若2条评论在意见词的使用模式上相似,则它们评论相同的产品属性的概率越高.定性和定量2种实验结果均表明,本文的正则化主题模型较传统的主题模型算法有更高的准确率,说明本文的正则化思想是有效的.

Implicit product feature extraction through regularized topic modeling

To solve the implicit product feature extraction task in product opinion mining, we proposed a novel regularized topic modeling framework based on the classical topic modeling through the analysis of the distribution of opinion words for different product features in reviews and also the assumption of topic dependency of opinion words. In the new framework, we took into consideration the opinionated information by defining a regularizer based on the similarity in opinion word usage of different reviews. The basic idea of the regularization was that if two reviews were similar in the usage of opinion words, they were more likely to comment on the same features. The qualitative and quantitative experiments both show that the novel framework outperforms classical topic modeling algorithms in accuracy and thus indicate the effectiveness of the regularization.