An Analysis of Online Review Helpfulness that Integrates Content and Function Words ─ An Application to Online Reviews on Amazon.com ─

Abstract

The helpfulness of online reviews can be driven by several factors: review attributes, reviewer attributes, review contents, review sentiments, and review language styles, among others. Although previous studies identified the importance of these factors, most of them only concentrated on a few of the factors and did not take a holistic view. In the field of computer science, researchers always focused on the review content and sentiment by mining content words, and neglected language styles determined by function words. This article aims to extend existing research on the classification of online reviews and their helpfulness by taking all of these factors into account. A theoretical framework that integrates（ 1） reviewer attribute,（ 2） review attribute,（ 3） review content,（ 4） review sentiment, and （5） review language style is proposed. Based on the framework, this study analyzed online reviews on iPads that are written in English and posted on Amazon.com. For the analysis of review content, topic modeling was applied through the latent Dirichlet allocation（ LDA） technique. For the analysis of review sentiment and language styles, a text processing software named LIWC （Linguistic Inquiry and Word Count） was used. A set of Poisson regression analysis was conducted, taking the number of “helpful” votes to the message as a dependent variable. Since significant relationships with the five factors are obtained, the effectiveness of the proposed framework integrating these was confirmed. This integrated view uncovers novel insights. Theoretical and practical implications are also discussed.