cs.IRFeb 24, 2026

A Counterfactual Approach for Addressing Individual User Unfairness in Collaborative Recommender System

Authors: Nikita Baidya, Bidyut Kr. Patra, Ratnakar Dash

Organizations: Pattern Recognition Laboratory, Department of Computer Science and Engineering, National Institute of Technology Rourkela, Odisha, India. · Department of Computer Science and Engineering, Indian Institute of Technology (BHU) Varanasi, Uttar Pradesh, India.

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Recommender Systems (RSs) are exploited by various business enterprises to suggest their products (items) to consumers (users). Collaborative filtering (CF) is a widely used variant of RSs which learns hidden patterns from user-item interactions for recommending items to users. Recommendations provided by the traditional CF models are often biased. Generally, such models learn and update embeddings for all the users, thereby overlooking the biases toward each under-served users individually. This leads to certain users receiving poorer recommendations than the rest. Such unfair treatment toward users incur loss to the business houses. There is limited research which addressed individual user unfairness problem (IUUP). Existing literature employed explicit exploration-based multi-armed bandits, individual user unfairness metric, and explanation score to address this issue. Although, these works elucidate and identify the underlying individual user unfairness, however, they do not provide solutions for it. In this paper, we propose a dual-step approach which identifies and mitigates IUUP in recommendations. In the proposed work, we counterfactually introduce new interactions to the candidate users (one at a time) and subsequently analyze the benefit from this perturbation. This improves the user engagement with other users and items. Thus, the model can learn effective embeddings across the users. To showcase the effectiveness of the proposed counterfactual methodology, we conducted experiments on MovieLens-100K, Amazon Beauty and MovieLens-1M datasets. The experimental results validate the superiority of the proposed approach over the existing techniques.

Figures & tables

Figure 1: Imputation of items with highest preference scores for a probable candidate user to obtain modified training data.
Figure 2: Process of finding user candidates for final imputation consideration.
Figure 3: Imputation of items with highest preference scores for a candidate user to obtain modified training data.
 Before pre-processingAfter pre-processing
Dataset#Users#Items#Ratings#Users#Items#Ratings
ML-100K9431682100,000943115297,953
A-Beauty1,210,271249,2742,023,0701,34073328,798
ML-1M6,0403,7061,000,2096,0403,260998,539
Table 1: Statistics of the three datasets before and after pre-processing.
  Overall Users
DatasetModelF1@10N@10F1@15N@15F1@20N@20
ML-100KAntidote Rastegarpanah et al. (2019)0.01710.02690.02270.02890.02630.0305
 OCEF Do et al. (2022)0.03550.04790.03970.05100.04450.0562
 DEE-KDE Cho et al. (2022)0.06550.09600.07450.09810.08020.1022
 ACFR-User u Li et al. (2024)0.03380.05650.03860.05500.04220.0558
 Proposed0.14590.19930.16250.20530.17100.2139
Table 2: Recommendation performance of proposed and fairness models for overall users.
  Candidate Users
DatasetModelF1@10N@10F1@15N@15F1@20N@20
ML-100KAntidote Rastegarpanah et al. (2019)0.01720.02850.01790.02900.01530.0277
 OCEF Do et al. (2022)0.05280.06440.05570.07740.05710.0847
 DEE-KDE Cho et al. (2022)0.06980.07500.08450.09140.09760.1102
 ACFR-User u Li et al. (2024)0.01760.03010.01920.03050.02550.0369
 Proposed0.14380.21630.15550.19920.17500.2072
Table 3: Recommendation performance of proposed and fairness models for candidate users.
  F1@10N@10F1@15N@15F1@20N@20
DatasetMethodOverall Users
ML-100KLGCN He et al. (2020)0.13420.18430.15050.18960.16250.2002
 Proposed0.14590.19930.16250.20530.17100.2139
  Candidate Users
 LGCN He et al. (2020)0.10990.16290.14130.16630.16090.1774
 Proposed0.14380.21630.15550.19920.17500.2072
Table 4: Recommendation performance on three datasets for the baseline and proposed model on all users and candidate users.
   #UsersBeforeAfterPGain
kMetricProposed WorkCandidate Users
10F1without counterfactual3720.13460.1210-0.0136
 with counterfactual270.10990.14380.0339
NDCGwithout counterfactual3720.16000.1395-0.0205
 with counterfactual270.16290.21630.0534
  Remaining Users
Table 5: The recommendation performance of proposed work on MovieLens 100K with and without the counterfactual segment, for k = 10.
   #UsersBeforeAfterPGain
kMetricMethodCandidate Users
15F1without counterfactual3720.14110.1302-0.0109
 with counterfactual270.14130.15550.0142
NDCGwithout counterfactual3720.17850.1585-0.02
 with counterfactual270.16630.19920.0329
  Remaining Users
Table 6: The recommendation performance of proposed work on MovieLens 100K with and without the counterfactual segment, for k = 15.
   #UsersBeforeAfterPGain
kMetricMethodCandidate Users
20F1without counterfactual3720.14500.1299-0.0151
 with counterfactual270.16090.17500.0141
NDCGwithout counterfactual3720.19990.1739-0.0260
 with counterfactual270.17740.20720.0298
  Remaining Users
Table 7: The recommendation performance of proposed work on MovieLens 100K with and without the counterfactual segment, for k = 20.
More in this topic

Fairness in Recommender Systems

See all 15 papers

Explore similar work

CardsList
Jun 27, 2026

Fairness Attacks on Recommender Systems

The unfairness of recommender systems has become a topic of concern due to its significant social and ethical implications. Although existing works have shown the effectiveness of attacks on the performance of recommender systems (e.g., promotion and demotion attack), the study of fairness attacks on recommender systems remains largely under-explored. To this end, we propose a novel structure-aware reinforcement learning-based fairness attack method designed to exacerbate the unfairness of target recommender systems. Specifically, we first employ a graph-based structure encoder to model the structural dependencies among the generated fake user-item interactions and the original user-item interactions. Then, we model the sequential dependency of the injected fake items using a recurrent neural network. Based on the learned structure-aware and sequence-aware representations of the fake user and item, the item selection policy attentively decides the next injected fake item. Since the target recommender system may employ fairness-aware training and leverage the user's sensitive attribute information, such as gender, we further designed a gender selection policy to decide the gender of the entire fake user profile. Both the item selection and gender selection policy are learned jointly in our proposed method. Finally, experimental results on four types of target recommendation models and two real-world datasets demonstrate the effectiveness of the proposed attack method in exacerbating the unfairness of recommender systems.

Yanan Wang, Yong Ge

Department of Information Systems and Operations Management, College of Business, The University of Texas at Arlington · Department of Management Information Systems, Eller College of Management, the University of Arizona

Adversarial AttacksAlgorithmic FairnessRecommender SystemsData Poisoning AttacksFairness in Recommender Systems
AskPDF
Aug 21, 2026

Why This, Not That? Mining User Profiles for Pair-wise Counterfactuals

The topic of explanation in recommender systems has seen steady research attention since the earliest days of the field. With some exceptions, this work has focused on the explanation of single items in a recommendation list and, especially recently, has emphasized approaches that are decoupled from the logic of the recommendation algorithm itself. Based on findings in the psychology of interpersonal communication, we propose a new task, pairwise interpretation of item rankings, asking the comparative question ``Why is item A ranked higher than item B?''. An effective solution to this task, we argue, is inherently grounded in the operation of the recommendation algorithm. We propose a class of techniques based on counterfactual learning to uncover the items in a user's profile that have contributed to the relative ranking of items. Using multiple datasets, we show that it is possible to identify such items as potential basis for comparative explanation.

Meysam Varasteh, Veronika Bogina, Noam Koenigstein +1

Department of Computer Science, University of Colorado Boulder, Boulder, CO 80309, USA · Tel Aviv University, Tel Aviv, Israel · Department of Information Science, University of Colorado Boulder, Boulder, CO 80309, USA

Explainable RecommendationCounterfactual Explanations
AskPDF
Apr 2, 2026

Modeling User Exploration Saturation: When Recommender Systems Should Stop Pushing Novelty

Fairness-aware recommender systems often mitigate bias by increasing exposure to under-represented or long-tail content, commonly through mechanisms that promote novelty and diversity. In practice, the strength of such interventions is typically controlled using global hyperparameters, fixed regularization weights, heuristic caps, or offline tuning strategies. These approaches implicitly assume that a single level of exploration is appropriate across users, contexts, and stages of interaction. In this work, we study exploration saturation as a user-dependent phenomenon arising from fairness- and novelty-driven recommendation strategies. We define exploration saturation as the point at which further increases in exploration no longer improve user utility and may instead reduce engagement or perceived relevance. Rather than proposing a new fairness-aware algorithm or optimizing a specific objective, we empirically analyze how increasing exploration affects users across varied recommendation models. Through longitudinal experiments using MovieLens-1M and Last.fm datasets, our results indicate that fairness-induced exploration exhibits diminishing or non-monotonic returns and varies substantially across users. In particular, users with limited interaction histories tend to reach saturation earlier, suggesting that uniform fairness or novelty pressure can disproportionately disadvantage certain users. These findings reveal a trade-off between fairness and user experience, suggesting that recommendation systems should adapt not only to relevance but also to the amount of fairness-driven exploration applied to individual users.

Enock O. Ayiku, Evelyn Osei, Emebo Onyeka

University of Massachusetts Boston, Boston, MA, USA · Kansas State University, Manhattan, KS, USA · Virginia Tech, Blacksburg, VA, USA

Recommender SystemsFairness in Recommender SystemsExploration-Exploitation TradeoffAlgorithmic Fairness
AskPDF