Counterfactual Explanations for Group Recommendations
Mubasher, Farhan (2024)
2024
Tietojenkäsittelyopin maisteriohjelma - Master's Programme in Computer Science
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2024-10-21
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202410179337
https://urn.fi/URN:NBN:fi:tuni-202410179337
Tiivistelmä
This study tackles the challenge of generating counterfactual explanations in group recommender systems by identifying a minimal set of items whose removal excludes a target item from the group’s recommendations. To address the complexity of this task, we introduce the Iterative Heuristic Search for Explanation Generation (IHSEG) technique, which efficiently identifies the minimal subset while accounting for key factors like fairness, user intensity, and item intensity. The IHSEG technique is evaluated using the MovieLens 1M dataset on three group types—homogeneous, heterogeneous, and hybrid—with 4-user and 8-user groups.
Our findings show that homogeneous groups, characterized by aligned preferences, produce shorter and more concise explanations, requiring fewer iterations. On the other hand, heterogeneous and hybrid groups, with more diverse preferences, demand larger explanation sets and additional iterations, due to the greater complexity in balancing varying user preferences. Larger groups also exhibit higher computational complexity, especially within heterogeneous groups where conflicting preferences significantly increase the search space for explanations.
This work demonstrates how preference alignment influences the ease of generating counterfactual explanations and highlights the trade-offs between conciseness, fairness, and computational efficiency. By improving the transparency of group recommendations, the IHSEG technique offers a scalable approach for real-world systems that prioritize fairness and user trust, providing insights into the future of interpretable and fair recommendation systems.
Our findings show that homogeneous groups, characterized by aligned preferences, produce shorter and more concise explanations, requiring fewer iterations. On the other hand, heterogeneous and hybrid groups, with more diverse preferences, demand larger explanation sets and additional iterations, due to the greater complexity in balancing varying user preferences. Larger groups also exhibit higher computational complexity, especially within heterogeneous groups where conflicting preferences significantly increase the search space for explanations.
This work demonstrates how preference alignment influences the ease of generating counterfactual explanations and highlights the trade-offs between conciseness, fairness, and computational efficiency. By improving the transparency of group recommendations, the IHSEG technique offers a scalable approach for real-world systems that prioritize fairness and user trust, providing insights into the future of interpretable and fair recommendation systems.