Generative Artificial Intelligence for Classification, Prediction, and Discovery of Consumer Product Recalls
Journal of Computing and Information Science in Engineering · 01.08.2026 · 26 (8)
DOI: 10.1115/1.4072427 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Computing and Information Science in Engineering, online erschienen am 13.08.2026. Verfasst von Diana Bolanos und 3 weiteren Autor:innen. Die Arbeit nennt 51 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Historical information offers valuable insights throughout product development, and recalled products in particular reveal potential design risks, yet they remain underutilized. We curate RECALL-MM, a multimodal dataset from the US Consumer Product Safety Commission (CPSC) recall database, augmented using generative methods. First, we employ generative artificial intelligence (GenAI) to classify hazards from textual descriptors; using GPT-4o, we achieve a relaxed accuracy of 0.73 against reference hazard labels. Second, we benchmark five state-of-the-art vision–language models (VLMs) on hazard prediction from product images, where Top-1 accuracy ranges from 23% to 40% and Top-3 reaches 67% with GPT-5. Although classical retrieval baselines (Sentence-BERT and ResNet50 nearest-neighbors) exceed VLM accuracy, the foundation models still identify the correct hazard among three plausible guesses well above chance, zero-shot, from a single image. The near-equivalence of image-feature and large language model (LLM)-distilled text retrieval suggests that visual interpretation carries most of the predictive signal, motivating continued development of flexible multimodal models. To complement these metrics, we apply an LLM-as-Judge framework that scores long-form predictions on a two-axis rubric, highlighting GenAI strengths and challenges in capturing hazard mechanisms and modalities. Finally, two application studies present interactive clustering maps that embed all recalls into a shared latent space to support hazard discovery. Together, these studies show how computational and GenAI-driven methods can leverage recall archives to classify, predict, and discover hazards, ultimately supporting risk-informed tools in early-stage engineering design.
Zitieren
APA
Bolanos, D., Ataei, M., Grandi, D., & Goucher-Lambert, K. (2026). Generative Artificial Intelligence for Classification, Prediction, and Discovery of Consumer Product Recalls. Journal of Computing and Information Science in Engineering, 26(8). https://doi.org/10.1115/1.4072427
Vancouver
Bolanos D, Ataei M, Grandi D, Goucher-Lambert K. Generative Artificial Intelligence for Classification, Prediction, and Discovery of Consumer Product Recalls. Journal of Computing and Information Science in Engineering. 2026;26(8). doi:10.1115/1.4072427
BibTeX
@article{Bolanos2026,
author = {Bolanos, Diana and Ataei, Mohammadmehdi and Grandi, Daniele and Goucher-Lambert, Kosa},
title = {Generative Artificial Intelligence for Classification, Prediction, and Discovery of Consumer Product Recalls},
journal = {Journal of Computing and Information Science in Engineering},
year = {2026},
volume = {26},
number = {8},
doi = {10.1115/1.4072427}
}
EUVIMED