Home // EXPLAINABILITY 2025, The Second International Conference on Systems Explainability // View article

RanXplain: Explaining Rankings in Recommendation Systems

Authors:
Atakan Yilmaz
Nuray Eylul Erler
Emre Atilgan
Melisa Bal Aslan

Keywords: Recommendation System; Explainable AI (XAI); Machine Learning Explainability.

Abstract:
Recommendation systems are designed to rank items according to users' predicted interest. As these systems increasingly affect choices in domains like e-commerce and media, understanding the reasoning behind their rankings becomes essential. However, most existing approaches that explain recommendations focus on individual predictions, rather than explaining why one item is prioritized over another. To bridge this gap, this paper introduces RanXplain, an approach specifically designed to explain the ranking decisions produced by recommendation models. RanXplain operates as a separate machine learning model trained on pairs of items, using features that are derived from the original ranking model. The impact of different feature sets and model architectures on model performance is systematically investigated. Furthermore, a simulation based performance evaluation was presented on different breakdowns, specifically analyzing the proximity of item ranks and whether items belong to the same category to detect scenarios in which RanXplain yields superior performance. A practical insight is discussed regarding instances in which RanXplain fails to identify the ranking model’s prioritization.

Pages: 21 to 27

Copyright: Copyright (c) IARIA, 2025

Publication date: October 26, 2025

Published in: conference

ISBN: 978-1-68558-318-7

Location: Barcelona, Spain

Dates: from October 26, 2025 to October 30, 2025