用于网络评论分析的主题-对立情感挖掘模型

Topic-Opposite Sentiment Mining Model for Online Review Analysis

为了挖掘网络评论中的产品主题和主题的对立情感信息,以帮助生产商和服务商改进产品和服务质量,帮助消费者做出购买决策,基于LDA(latent Dirichlet allocation)提出了一个用于网络评论分析的主题-对立情感挖掘模型(topic-opposite sentiment mining model,TOSM),模型中假设句子为分配主题和情感的最小单位。该模型在LDA的基础上增加情感层,将LDA的三层结构拓展为四层,能同时得到主题以及主题的对立情感信息。为了使对立情感的描述更准确,在情感层中融入了情感词典先验信息。在Amazon网站的电子产品评论和Yelp网站的饭店评论数据集上进行了三组实验,实验表明,TOSM挖掘到的观点主题与评论中有价值的细节描述相匹配,TOSM模型的情感分类结果优于其他模型。

At mining product topics and opposite sentiment information of the topics from online reviews, in order to help manufactures and service providers improve their products and services, and help customers make decisions, this paper proposes a topic model called topic-opposite sentiment mining model （TOSM） based on latent Dirichlet allocation （LDA）, which assumes that all words in a single sentence are generated from one topic and one sentiment. This paper extends LDA to TOSM with adding the sentiment layer, so that TOSM can detect topics and opposite sen- timent of topic simultaneously from reviews. Moreover, this paper uses sentiment lexicon in TOSM to make the opposite sentiment represented clearly. This paper does three experiments with the reviews of electronic devices from Amazon and the reviews of restaurants from Yelp. The experimental results show that the topic-sentiment found by TOSM matches evaluative details of the reviews, and TOSM outperforms other generative models.