A User-Centric Review and Comparative Analysis of Explainable AI Methods for Social Media Recommender Systems

Alkhateeb, B; Solaiman, E

doi:10.1109/ICMI68585.2026.11539924

A User-Centric Review and Comparative Analysis of Explainable AI Methods for Social Media Recommender Systems

Lookup NU author(s): Dr Ellis Solaiman ORCiD

Downloads

Accepted version [.pdf]

Licence

This is the authors' accepted manuscript of a conference proceedings (inc. abstract) that has been published in its final definitive form by IEEE, 2026.

For re-use rights please refer to the publisher's terms and conditions.

Abstract

As algorithmic transparency becomes increasingly important, social media platforms have integrated explainability features into their recommender systems, following a one-size-fits-all approach that fails to address the diverse explainability needs of different users. This survey presents a comparative synthesis of existing literature on five XAI methods (LIME, SHAP, Anchors, Counterfactual Explanations, and Concept Bottleneck Models) and evaluates their suitability for three types of social media users (developers, domain experts, and lay users). To support this analysis, we conducted a systematic search across three major academic databases using relevant keywords, filtering for publications from 2018 to 2025. A total of 30 papers met the inclusion criteria and were selected for review. Our analysis demonstrates that no single XAI method works equally well for all user tiers or all social-media contexts. We therefore advocate tier-aware explainability, explicitly matching XAI techniques to user types and platform contexts, to maximize clarity, trust, and accountability.