Cross-Domain Sentiment Analysis of Product Reviews by Combining Lexicon-Based and Learn-Based Techniques

Product reviews can direct consumers' purchasing behavior and sellers' marketing strategy. Therefor, in this paper, we propose a novel method which combines lexicon-based and learn-based techniques in order to solve cross-domain sentiment analysis of product reviews. We first create three domain lexicons based on the basic lexicon and corpus from three domains containing book, hotel and electronics. Furthermore, we use four kinds of features to build classifiers. Besides, we conduct a series of experiments to evaluate our method by using different lexicons and different classifiers. Experimental results on data of product reviews show that domain lexicons outperform the basic lexicon no matter in which domain. What's more, our method performs much better than the existing state-of-the-art method especially in domains of book and hotel, and is slightly inferior in the domian of electronics.