↓Skip to main content

An Interpretable Data-Driven Unsupervised Approach for the Prevention of Forgotten Items

·2 mins

TL;DR #

An Interpretable Data-Driven Unsupervised Approach for the Prevention of Forgotten Items introduces methods for reminding shoppers about products they may have intended to buy but omitted.

Problem #

Shopping-recommendation systems often focus on suggesting new products. A different and practical problem is predicting a forgotten item: a product a person is likely to need or buy regularly but did not include in the current basket.

What did we do? #

We formalized forgotten-item prediction as an unsupervised learning problem. The proposed methods, XMT and TXMT, combine signals such as purchase recurrence, time between purchases, seasonality, co-purchases, and sequential shopping patterns.

The approach is designed to remain interpretable. Instead of returning only a score, it can expose the behavioral signals that support a reminder, such as a product’s usual purchase interval or its relationship with the current basket.

Main findings #

The experiments compare the proposed methods with next-basket and forgotten-item baselines on real retail data. The data-driven models improve prediction quality while retaining human-readable reasons for their recommendations.

Resources #

Cite this paper #

@inproceedings{corbucci2025forgotten,
  title = {An Interpretable Data-Driven Unsupervised Approach for the Prevention of Forgotten Items},
  author = {Corbucci, Luca and Borges Legrottaglie, Javier Alejandro and Spinnato, Francesco and Monreale, Anna and Guidotti, Riccardo},
  booktitle = {Proceedings of the 28th European Conference on Artificial Intelligence},
  year = {2025},
  doi = {10.3233/FAIA250912}
}