TabM vs. Traditional ML for e-Commerce Product Ranking: A Multi-Signal Framework for Frequently Bought Together Recommendations

Dilge Karakaş, Enis Teper, Okan Kaya

2025

Abstract

We present a machine learning framework for ranking products in e-commerce recommendation systems, specifically targeting “Frequently Bought Together” scenarios. Leveraging a TabM neural architecture with parameter-efficient BatchEnsemble mechanisms for ensemble learning, our system integrates similarity scores, position signals, and commercial performance metrics to optimize purchase probability predictions. Deployed on a major e-commerce platform, our approach demonstrates improved ranking performance while main- taining computational efficiency through strategic weight sharing across ensemble members. TabM model achieves 23.5% improvement in HR@5 over position-based baseline and 14.5% improvement in NDCG@10 over logistic regression. The model effectively handles class imbalance through diverse ensemble perspectives and significantly outperforms traditional machine learning approaches including gradient boosting and logistic regression.

Paper Citation

in Harvard Style

Karakaş D., Teper E. and Kaya O. (2025). TabM vs. Traditional ML for e-Commerce Product Ranking: A Multi-Signal Framework for Frequently Bought Together Recommendations. In Proceedings of the 2nd International Conference on Advances in Electrical, Electronics, Energy, and Computer Sciences - Volume 1: ICEEECS; ISBN 978-989-758-783-2, SciTePress, pages 243-248. DOI: 10.5220/0014297100004848

in Bibtex Style

@conference{iceeecs25,
author={Dilge Karakaş and Enis Teper and Okan Kaya},
title={TabM vs. Traditional ML for e-Commerce Product Ranking: A Multi-Signal Framework for Frequently Bought Together Recommendations},
booktitle={Proceedings of the 2nd International Conference on Advances in Electrical, Electronics, Energy, and Computer Sciences - Volume 1: ICEEECS},
year={2025},
pages={243-248},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0014297100004848},
isbn={978-989-758-783-2},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 2nd International Conference on Advances in Electrical, Electronics, Energy, and Computer Sciences - Volume 1: ICEEECS
TI - TabM vs. Traditional ML for e-Commerce Product Ranking: A Multi-Signal Framework for Frequently Bought Together Recommendations
SN - 978-989-758-783-2
AU - Karakaş D.
AU - Teper E.
AU - Kaya O.
PY - 2025
SP - 243
EP - 248
DO - 10.5220/0014297100004848
PB - SciTePress