Selecting Cover Images for Restaurant Reviews: AI vs. Wisdom of the Crowd
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Problem definition: Restaurant review platforms, such as Yelp and TripAdvisor, routinely receive large numbers of photos in their review submissions. These photos provide significant value for users who seek to compare restaurants. In this context, the choice of cover images (i.e., representative photos of the restaurants) can greatly influence the level of user engagement on the platform. Unfortunately, selecting these images can be time consuming and often requires human intervention. At the same time, it is challenging to develop a systematic approach to assess the effectiveness of the selected images. Methodology/results: In this paper, we collaborate with a large review platform in Asia to investigate this problem. We discuss two image selection approaches, namely crowd-based and artificial intelligence (AI)-based systems. The AI-based system we use learns complex latent image features, which are further enhanced by transfer learning to overcome the scarcity of labeled data. We collaborate with the platform to deploy our AI-based system through a randomized field experiment to carefully compare both systems. We find that the AI-based system outperforms the crowd-based counterpart and boosts user engagement by 12.43%–16.05% on average. We then conduct empirical analyses on observational data to identify the underlying mechanisms that drive the superior performance of the AI-based system. Managerial implications: Finally, we infer from our findings that the AI-based system outperforms the crowd-based system for restaurants with (i) a longer tenure on the platform, (ii) a limited number of user-generated photos, (iii) a lower star rating, and (iv) lower user engagement during the crowd-based system. Funding: The authors acknowledge financial support from the Social Sciences and Humanities Research Council [Grant 430-2020-00106]. Supplemental Material: The online appendix is available at https://doi.org/10.1287/msom.2021.0531 .
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.001 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it