Predicting continued review activity in grocery e-commerce with a pretrained language model
Introduction Predicting whether a reviewer will contribute again in grocery e-commerce requires an evaluation aligned with the timing of available information. This study evaluates a pretrained language model for continued review activity, defined as another verified-purchase-tagged grocery review by the same user within 180 days. Methods Using Amazon Reviews 2023, the design combines 300,000 training anchors, 100,000 development-validation anchors, mature training labels, product-snapshot exclusion, matched information sets, and five fitting seeds. Frozen models underwent one-time temporal confirmation on 100,000 later anchors from the same platform and category. Results RoBERTa achieved a mean average precision (AP) of 0.599751, compared with 0.595339 for LightGBM and 0.584889 for joint text-structured logistic regression. The respective AP gains were 0.004412 and 0.014862, with multiplicity-adjusted user-cluster 97.5% confidence intervals of [0.001847, 0.006733] and [0.011646, 0.017903]. Both adjusted lower bounds exceeded zero, and all five paired RoBERTa-LightGBM differences were positive. The exploratory low-history-user AP gain was 0.019731, with unadjusted subgroup intervals and no interaction test. Discussion The time-aligned comparison shows a modest ranking advantage for the evaluated RoBERTa configuration. Practical utility, calibration, and causal effects require separate evaluation. Applications to intelligent language services, cross-border enterprises, and sustainable food-system outcomes require separate validation.
Authors
- Leiyi Fang
Institutions
- Hunan University (CN)
- Department of Science and Technology of Hunan Province (CN)
- Hunan University of Information Technology (CN)
- Hunan University of Technology (CN)
- Changsha University of Science and Technology (CN)
Publication Details
- Journal
- Frontiers in Sustainable Food Systems
- Published
- 2026-11-18
- DOI
- https://doi.org/10.3389/fsufs.2026.2006716
- Primary Topic
- Sentiment Analysis and Opinion Mining
- Type
- article
- Field-Weighted Citation Impact
- 0.00