AI vs. Wisdom of the Crowds in Selecting Cover Images for Restaurant Reviews
42 Pages Posted:
Date Written: October 31, 2020
Restaurant review platforms such as Yelp and TripAdvisor have received an increasing amount of photos in the review submissions. These photos provide additional values to the reviews as previous literature has shown that platform users generally perceive them as being useful. As such, the choice of the "cover images" (i.e., the representative images of each restaurant) can potentially influence the level of user engagement in the platform. However, the selection of such images is time consuming and often requires human intervention. Meanwhile, it is particularly challenging to develop a quantitative approach to systematically assess the images on their effectiveness in increasing user engagement. In this study, we collaborate with a large-scale review platform in Asia to investigate this issue. We discuss two different image selection designs, namely the crowd-based and AI-based systems. The AI-based system, which is used to learn complex latent image features, is further enhanced through the feature representation transfer process to overcome lack of high-quality labeled data. In addition to the holdout evaluation method, we conduct a randomized field experiment to objectively evaluate the effectiveness of both designs and show that the AI-based system can outperform the crowd in selecting images that stimulate user interactions. Post-hoc analyses using observational data are conducted to identify underlying mechanisms that drive the superior performance of the AI-based system.
Suggested Citation: Suggested Citation