Online review systems

Abstract

Consumers refer to the reviews to collect information for their purchasing decisions, despite the questions about the accuracy of the ratings. Potential problems (e.g., manipulated reviews) lead consumers to question the trustworthiness of the ratings. Therefore, in the first essay, we empirically investigate the effect of valence on sales before and after the verified purchase scheme was introduced. Our results reveal that the introduction of the verified purchase scheme increases the predictive power of valence on sales and increases the reviewers’ rating value at the same time. These findings indicate that ratings biases or manipulations may restrict the value of the open review system. In addition, in the second essay, we investigate the reviewers and their reviews that can decrease the predictive power of valence. Our findings indicate that the predictive power of valence on sales in the mixed review systems is limited because of non-verified purchase reviews. To improve the predictive power of valence, we define a small subset of reviewers who have an extreme ratings pattern. Based on clustering analysis, text mining, and topic modeling, these reviewers have a possibility of degrading the predictive power of valence. Removing small percentage of these reviewers can improve the predictive power of valence. Our proposed method suggests the new perspective of manipulation and its management.