Paper Number
ICIS2025-1117
Paper Type
Complete
Abstract
Modern review-based recommender systems achieve high accuracy by increasingly adopting advanced deep learning techniques. However, this makes their recommendation processes intransparent, diminishing user experience. Existing methods to increase transparency struggle to provide intuitive, user-centric explanations. To address this challenge, we propose C-ReX, a novel, model-agnostic approach for concept-based variation of review texts, enabling the generation of user-centric explanations for review-based recommender systems. C-ReX identifies human-relevant concepts within reviews, structures them meaningfully, and systematically perturbs the texts to quantify their influence on recommendations. This approach is compatible with any perturbation-based Explainable Artificial Intelligence method and any black-box recommender system. Our user study reveals that users consistently prefer C-ReX’s explanations over the state-of-the-art LIME method. These findings establish C-ReX as a promising approach to improve transparency in review-based recommender systems while maintaining model performance.
Recommended Citation
Glatzel, Anna-Lena; Habla, Maximilian; Züllig, Kilian; and Zimmermann, Steffen, "From Reviews to Insights: Concept-based Variations of Textual Reviews for Explanations in Recommender Systems" (2025). ICIS 2025 Proceedings. 2.
https://aisel.aisnet.org/icis2025/user_behav/user_behav/2
From Reviews to Insights: Concept-based Variations of Textual Reviews for Explanations in Recommender Systems
Modern review-based recommender systems achieve high accuracy by increasingly adopting advanced deep learning techniques. However, this makes their recommendation processes intransparent, diminishing user experience. Existing methods to increase transparency struggle to provide intuitive, user-centric explanations. To address this challenge, we propose C-ReX, a novel, model-agnostic approach for concept-based variation of review texts, enabling the generation of user-centric explanations for review-based recommender systems. C-ReX identifies human-relevant concepts within reviews, structures them meaningfully, and systematically perturbs the texts to quantify their influence on recommendations. This approach is compatible with any perturbation-based Explainable Artificial Intelligence method and any black-box recommender system. Our user study reveals that users consistently prefer C-ReX’s explanations over the state-of-the-art LIME method. These findings establish C-ReX as a promising approach to improve transparency in review-based recommender systems while maintaining model performance.
When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.
Comments
16-UserBehavior