Aspect And Sentiment Extraction Based On Semi-supervised Topic Model

With the rapid development of the Internet, people are becoming more and more frequent to behave online, such as expressing their opinions or views on products. In general, when people comment on the products or services they consume on E-commerce websites, these comments always contain sentiment expression on detail aspects of products or services. Thesentimentalinformation is valuable for both users and business organizations. However, most of the reviews are plain text and thus require much effort to obtain information about relevant details.In sentiment analysis, aspect extraction is a key problem. Existing methods either generate multiple fine-grained aspects without proper categorization or categorize semantically unrelated product aspects, such as by unsupervised topic modeling. By categorizing, we mean that the synonymous aspects should be clustered into the same category.In order to overcome the limitations of existing methods, a novel semi-supervised model for product aspect extraction and sentiment analysis with a Word2 Vec hybrid(WSAS) is then proposed. In this research, we tackle the problem of automatically detecting what aspects are evaluated in reviews and how sentiments for different aspects are expressed. This proposed model incorporates aspect and sentiment together to model sentiments toward different aspects. WSAS discovers pairs of {aspects, sentiment} which we call senti-aspects. More specifically, WSAS first makes sentiment analysis on reviews of products or services, and treats these sentiments information as prior knowledge of model's sentiment distribution. Then, WSAS adds some seeding aspects and general sentiment words. These seeding aspects words and sentiment words are applied to guide the model to discover user sentiment words that are aspect-specific. Our experiment results show that the proposed model outperformsstate-of-the-art models forsenti-aspects extraction on some benchmarkhotel reviews data.