Mitigating Popularity Bias in Fair and Explainable Recommender Systems Using SHAP

Abstract

Popularity bias is a prevalent issue in recommendation systems, where popular items dominate recommendation lists, leading to reduced diversity and fairness. Traditional methods evaluate popularity bias based on overall item frequency, disregarding individual user tendencies. This study introduces a novel post-processing ranking method called Dynamic User Tendency Re-ranking (DUTR) to mitigate popularity bias in multi-criteria recommendation systems by incorporating user-specific preferences. DUTR leverages SHAP (SHapley Additive exPlanations) analysis to determine the influence of different criteria on user decision-making. Unlike conventional methods, which classify item popularity based on general trends, DUTR dynamically assesses each user's priority preferences. It then classifies items as popular or less popular based on individual preference patterns. This approach ensures that recommendation lists align more closely with user-specific interests while maintaining a balance between popular and less popular items. To validate the effectiveness of DUTR, extensive experiments were conducted on the YM10 and YM20 datasets. The results show that DUTR significantly reduces popularity bias while improving diversity and fairness in recommendations. Moreover, the integration of SHAP values enhances the explainability of the recommendation process, providing users with personalized and transparent suggestions. In conclusion, comparative analysis with existing techniques demonstrates that DUTR outperforms traditional methods in balancing popularity and personalization.

Keywords

• Recommender systems
• Popularity bias
• User tendency
• SHAP
• Multi-criteria systems

Adil ve Açıklanabilir Öneri Sistemlerinde SHAP ile Popülerlik Yanlılığını Giderme

Abstract

Popülerlik yanlılığı, öneri sistemlerinde yaygın bir sorundur; popüler öğeler öneri listelerine hakim olur ve bu durum çeşitliliğin ve adaletin azalmasına neden olur. Geleneksel yöntemler popülerlik yanlılığını genel öğe sıklığına göre değerlendirirken, bireysel kullanıcı eğilimlerini göz ardı etmektedir. Bu çalışma, çok kriterli öneri sistemlerinde popülerlik yanlılığını azaltmak amacıyla, kullanıcıya özgü tercihler içeren yeni bir son işlem sıralama yöntemi olan Dinamik Kullanıcı Eğilimi Yeniden Sıralama (DUTR) yöntemini önermektedir. DUTR, kullanıcıların karar verme süreçlerinde farklı kriterlerin etkisini belirlemek için SHAP (SHapley Additive exPlanations) analizinden yararlanmaktadır. Geleneksel yöntemler öğelerin popülerliğini genel eğilimlere göre sınıflandırırken, DUTR her kullanıcının öncelikli tercihlerini dinamik olarak değerlendirmektedir. Daha sonra, bireysel tercih kalıplarına göre öğeleri popüler veya daha az popüler olarak sınıflandırmaktadır. Bu yaklaşım, öneri listelerinin kullanıcıların özel ilgi alanlarıyla daha iyi örtüşmesini sağlarken, popüler ve daha az popüler öğeler arasında bir denge oluşturmayı hedeflemektedir. DUTR'nin etkinliğini doğrulamak için YM10 ve YM20 veri setleri üzerinde kapsamlı deneyler gerçekleştirilmiştir. Sonuçlar, DUTR'nin popülerlik yanlılığını önemli ölçüde azalttığını ve önerilerin çeşitliliğini ve adaletini artırdığını göstermektedir. Ayrıca, SHAP değerlerinin entegrasyonu, öneri sürecinin açıklanabilirliğini geliştirerek kullanıcılara kişiselleştirilmiş ve şeffaf öneriler sunmaktadır. Sonuç olarak, mevcut tekniklerle yapılan karşılaştırmalı analizler, DUTR'nin popülerlik ve kişiselleştirme arasında denge sağlamada geleneksel yöntemlerden daha başarılı olduğunu ortaya koymaktadır.

Keywords

• Öneri sistemleri
• Popüler yanlılığı
• Kullanıcı eğilimi
• SHAP
• Çok ölçütlü sistemler

References

1. [1] F. O. Isinkaye, Y. Folajimi and B. A. Ojokoh, "Recommendation systems: Principles, methods and evaluation," Egyptian Informatics Journal, vol. 16, pp. 261–273, 2015.
2. [2] H. Abdollahpouri, M. Mansoury, R. Burke and B. Mobasher, "The unfairness of popularity bias in recommendation," RMSE Workshop 13th ACM Conferenceon Recommender Systems (RecSys), Copenhagen, Denmark, 2019.
3. [3] L. Boratto, G. Fenu and M. Marras, "Connecting user and item perspectives in popularity debiasing for collaborative recommendation," Information Processing and Management, vol. 58, 2021, Art. no. 102387.
4. [4] I. Covert and S.-I. Lee, "Improving KernelSHAP: Practical shapley value estimation using linear regression," in Proceedings of The 24th International Conference on Artificial Intelligence and Statistics, San Diego, California, USA, 2021, pp. 3457–3465.
5. [5] C. Anderson, The Long Tail: Why the Future of Business is Selling More For Less. New York, USA: Hyperion, 2006.
6. [6] E. Brynjolfsson, Y. J. Hu and M. D. Smith, "From niches to riches: Anatomy of the long tail," Sloan Management Review, vol. 47, no. 4, pp. 67–71, 2006.
7. [7] Ò. Celma and P. Cano, "From hits to niches?: Or how popular artists can bias music recommendation and discovery," in NETFLIX '08: Proceedings of the 2nd KDD Workshop on Large-Scale Recommender Systems and the Netflix Prize Competition, Las Vegas, Nevada, USA, 2008, pp. 1-8.
8. [8] Y. J. Park and A. Tuzhilin, "The long tail of recommender systems and how to leverage it," in RecSys’08: Proceedings of the 2008 ACM Conference on Recommender Systems, Lausanne, Switzerland, 2008, pp. 11-18.

9. [9] C. Chen, M. Zhang, Y. Liu and S. Ma, “Missing data modeling with user activity and item popularity in recommendation,” in Information Retrieval Technology: 14th Asia Information Retrieval Societies Conference, AIRS 2018, Y.H. Tseng et al., Eds., Taipei, Taiwan, 2018, pp. 113-125.
10. [10] T. Kamishima, S. Akaho, H. Asoh and J. Sakuma, "Correcting popularity bias by enhancing recommendation neutrality," Proceedings of the 8th ACM Conference on Recommender Systems (RecSys 2014), Foster City, Silicon Valley, USA, 2014.
11. [11] H. Abdollahpouri, R. Burke and B. Mobasher, "Controlling popularity bias in learning-to-rank recommendation," in RecSys 2017 - Proceedings of the 11th ACM Conference on Recommender Systems, Como, Italy, 2017, pp. 42–46, 2017.
12. [12] H. Abdollahpouri, R. Burke and B. Mobasher, "Popularity-aware ıtem weighting for long-tail recommendation," arXiv preprint arXiv:1802.05382, 2018.
13. [13] H. Abdollahpouri, R. Burke and B. Mobasher, "Managing popularity bias in recommender systems with personalized re-ranking," in Proceedings of the 32nd International Florida Artificial Intelligence Research Society Conference, FLAIRS 2019, Florida, USA, 2019.
14. [14] H. Abdollahpouri, M. Mansoury, R. Burke, B. Mobasher and E. Malthouse, "User-centered evaluation of popularity bias in recommender systems," in UMAP'21: Proceedings of the 29th ACM Conference on User Modeling, Adaptation and Personalization, Utrecht, Netherlands, 2021, pp. 119-129.
15. [15] E. Yalcin and A. Bilge, "Evaluating unfairness of popularity bias in recommender systems: A comprehensive user-centric analysis," Information Processing & Management, vol. 59, no. 6, 2022, Art. no. 103100.
16. [16] E. Yalcin and A. Bilge, "Blockbuster: A new perspective on popularity-bias in recommender systems," in 2021 6th International Conference on Computer Science and Engineering (UBMK), Ankara, Türkiye, 2021, pp. 107–112.
17. [17] E. Yalcin and A. Bilge, "Popularity bias in personality perspective: An analysis of how personality traits expose individuals to the unfair recommendation," Concurrency and Computation: Practice and Experience, vol. 35, no. 9, 2023, Art. no. e7647.
18. [18] Y. Tacli, E. Yalcin and A. Bilge, "Novel approaches to measuring the popularity inclination of users for the popularity bias problem," in 2022 International Symposium on Multidisciplinary Studies and Innovative Technologies (ISMSIT), Ankara, Türkiye, 2022, pp. 555–560.
19. [19] R. Sanders, "The Pareto principle: its use and abuse," Journal of Services Marketing, vol. 1, no. 2, pp. 37–40, 1987.
20. [20] M. Gulsoy, E. Yalcin, Y. Tacli and A. Bilge, "DUoR: dynamic user-oriented re-ranking calibration strategy for popularity bias treatment of recommendation algorithms," International Journal of Human-Computer Studies, vol. 203, 2025, Art. no. 103578.
21. [21] K. Lakiotaki, N. F. Matsatsinis and A. Tsoukias, "Multicriteria user modeling in recommender systems," IEEE Intelligent Systems, vol. 26, pp. 64–76, 2011.
22. [22] H. Wang, Y. Lu and C. Zhai, "Latent aspect rating analysis without aspect keyword supervision," in Proceedings of the 17th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, San Diego, California, USA, 2011, pp. 618–626.
23. [23] A. M. Turk and A. Bilge, "Robustness analysis of multi-criteria collaborative filtering algorithms against shilling attacks," Expert Systems with Applications, vol. 115, pp. 386–402, 2019.
24. [24] J. L. Herlocker, J. A. Konstan, A. Borchers and J. Riedl, "An algorithmic framework for performing collaborative filtering," in Proceedings of the 22nd Annual International ACM SIGIR Conference on Research and Development in Information Retrieval, SIGIR 1999, Bberkeley, CA, USA, 1999, pp. 230-237.
25. [25] L. Zhang, "The definition of novelty in recommendation system," Journal of Engineering Science & Technology Review, vol. 6, pp. 141–145, 2013.