Specialized review selection using topic models

Abstract

Online reviews and comments about a product or service are an invaluable source of information for users to assist them in making purchase decisions. In recent years, the research in review selection has attracted considerable attention. Many of the existing works attempted to identify a number of statistical features related to review text such as word count (Mudambi and Schuff 2010) and hidden relations between these features and review quality by using supervised learning methods such as classification techniques. However, one significant drawback of these works is that they do not take the review content into consideration. A recent work has been proposed to find specialized reviews that focus on a specific feature based on similar words to the feature (Long et al. 2014). In this paper, we propose a topic model based method which selects reviews by considering both similar words and related words from a topic model such as LDA model. The conducted experiment has proven that those related words generated from LDA have a great contribution to the task of finding helpful reviews on a specified feature.