Mining associative and comparative patterns for Thai sentiment analysis

The aspect-based sentiment analysis has been popularly applied for analyzing product reviews. The results from the analysis could help summarize the customer satisfaction towards the products. Previous aspect-based sentiment analysis only focuses on relating sentiment polarity with product aspect. In a competitive market, it is more important to gain some insight of the sentiment and aspect towards a product brand and in comparison with other brands. To enhance the existing aspect-based sentiment analysis, we propose the following extensions, (1) association of sentiment and aspect with the product brand, and (2) comparison between two product brands. Our proposed approach is based on the pattern analysis of simplified patterns with dynamic filler terms. The simplified patterns help increase the effectiveness of pattern extraction. To evaluate the proposed approach, we performed experiments using product reviews in the smartphone domain. The results show that the performance of product brand and aspect extraction is significantly improved with the simplified patterns for both associative and comparative pattern minings.