A Semi-Supervised Approach for User Reviews Topic Modeling and Classification

Nowadays, many people prefer to purchase through online websites. Usually, those people start with reading user reviews, i.e., comments, before making a purchase decision. The user reviews are considered powerful sources of information about products, in which users share opinions and previous experiences on using these products. However, these reviews are mostly textual and uncategorized. Thus, new customers need to read a massive amount of reviews, one by one, in order to make a decision. Topic modeling is a common approach to categorize reviews so that the customers only read the reviews in a certain topic of interest, i.e., category. In literature, most related studies applies unsupervised learning for topic modeling and reviews classification. This paper proposes a combination of both unsupervised and supervised learning, i.e., semi-supervised learning, for classifying the reviews. The proposed approach was applied on a set of Amazon reviews and results showed that this approach is more efficient as compared to the unsupervised learning approach solely.