Machine Reading for Notion-Based Sentiment Mining

Although several sentiment classification methods have been proposed, rare are the ones that provide a solid link between human analysis of a sentiment text and machine analysis of the same text. In this paper, we investigate the automation of human's reading and building of notions. We show that the proposed process of automated machine reading and notion extraction can be used for sentiment mining and the identification of product characteristics. We then focus on the issue of sentiment classification of online reviews which have been receiving an increased level of attention by online users as well as data mining researchers in the past few years. Our proposed method bases itself on the human's thought process in extracting notion and sentiment from text. The approach first employs part of speech (POS) tagging, then learns the product features and characteristics that form the preconceived thoughts or notions about a topic's sub features. The idea is similar to how humans learn and build preconceived notions based on the review's topic, and then use it when reading a new review. Experiments show the success of the method in sentiment mining and in extracting a product's desirable features.