Understanding and Predicting Users’ Rating Behavior: A Cognitive Perspective

1 Oct 2020

Information Systems

Li, Qiudan; Zeng, Daniel Dajun; Xu, David Jingjun; Liu, Ruoran; Yao, Riheng

Published in INFORMS Journal on Computing, October 2020  

With the popularity and success of e-commerce, online consumer reviews have become a valuable source for consumers, businesses, and organizations across multiple stages of decision processes. During the process, review content, aspect ratings, and overall rating provide product evaluation at multiple levels of granularity. Review content includes the semantics and users’ opinion and sentiment; Aspect ratings reflect users’ feelings about more specific attributes of a product; and the overall rating score is a mixture of product quality and the customer’s overall interest in the product. Understanding and predicting users’ rating behavior will enable the market to estimate why and how satisfied (or not) a customer will be with a product, thus providing great opportunities for users and organizations to make better decisions.  

To this end, based on the part-whole model of learning theory and cognitive load theory in cognitive psychology literature, David Jingjun Xu, Associate Professor at the Department of Information Systems and co-authors develop a hierarchical attention-based neural network mechanism to unify the explicit aspect ratings and review contents to predict the overall ratings.  

“The findings suggest that our proposed approach can better capture the sentiment information of the review and associate aspect ratings with important words and sentences in review content, which helps better explain the detailed reasons why users like/dislike each aspect,” says Xu.  

  Experiments on two real-world data sets show that the performance advantage of the proposed approach mainly comes from the high-quality representation of review content and the effective integration of aspect ratings.  

“Our user study shows that both written reviews and aspect ratings influence users’ perceived review helpfulness and help users understand the overall score given to the reviewed restaurants,” Xu adds. 

Aspect ratings were further found to reduce users’ cognitive effort in understanding the overall score given for a product. 

  There are several practical implications. First, users and merchants can better understand why certain overall ratings are given for a product by examining the most relevant aspect ratings that have the effects on the overall ratings. Aspect ratings were found to reduce consumers’ cognitive effort in understanding the overall score given for a product. Users/businesses can consider this information comprehensively and efficiently when tracking product quality and making a choice. Second, identifying important words and sentences related to aspect ratings helps users/businesses to better understand the detailed reasons why users like/dislike the product so that they can quickly digest them without actually reading through large amounts of reviews.