Mining Multifaceted Customer Opinions from Online Reviews

Abstract

Online reviews are a valuable source for studying customer needs and preferences. Previous studies focus on extracting a set of a priori defined constructs such as product attribute perception or explicit customer needs from reviews. Such a priori focus circumvents the limitations of certain natural language processing algorithms but discards valuable information in reviews that are not in the scope of the predefined construct. This study proposes a new method of extracting customer opinions and opinion targets from reviews with the Aspect Sentiment Triplet Extraction (ASTE) algorithm and then identifying theoretical constructs critical for product development with a posteriori interpretation method. We demonstrate the value of our proposed method by identifying granular opinion targets and expressions to find infrequent but important phenomena such as user innovations and delights.