Abstract:Although recommenders can ship items to users automatically based on the users' preferences, they often cause unfairness to groups or individuals. For instance, when users can be divided into two groups according to a sensitive social attribute and there is a significant difference in terms of activity between the two groups, the learned recommendation algorithm will result in a recommendation gap between the two groups, which causes group unfairness. In this work, we propose a novel recommendation algorithm named Diffusion-based Fair Recommender (DifFaiRec) to provide fair recommendations. DifFaiRec is built upon the conditional diffusion model and hence has a strong ability to learn the distribution of user preferences from their ratings on items and is able to generate diverse recommendations effectively. To guarantee fairness, we design a counterfactual module to reduce the model sensitivity to protected attributes and provide mathematical explanations. The experiments on benchmark datasets demonstrate the superiority of DifFaiRec over competitive baselines.

What problem does this paper attempt to address?

### Problems the Paper Attempts to Solve The paper aims to address the issue of group unfairness in recommendation systems. Specifically, when users can be divided into two groups based on certain sensitive social attributes (such as gender, age, etc.), and there is a significant difference in activity levels between these two groups, traditional recommendation algorithms may lead to recommendation gaps, resulting in group unfairness. To tackle this problem, the authors propose a new recommendation algorithm—Fair Recommender based on Diffusion Model (DifFaiRec), to provide fairer recommendation results. ### Key Points Summary 1. **Problem Background**: - Recommendation systems can automatically recommend items to users based on their preferences, but they often lead to unfairness towards certain groups or individuals. - For example, when users are divided into two groups based on a sensitive social attribute, and there is a significant difference in activity levels between these two groups, the recommendation algorithm will create a recommendation gap between these groups, leading to group unfairness. 2. **Solution**: - A new recommendation algorithm, DifFaiRec, is proposed, which is based on a conditional diffusion model. - DifFaiRec has a strong ability to learn the distribution of user preferences from user ratings and generate diverse recommendation results. - To ensure fairness, a counterfactual module is designed to reduce the model's sensitivity to protected attributes, and a mathematical explanation is provided. 3. **Method Characteristics**: - **Counterfactual Module**: Ensures group fairness by mapping each user to different groups to obtain counterfactual users. - **Conditional Diffusion Model**: Reconstructs observed interactions based on the counterfactual module and predicts unknown interactions in a generative manner. - **Objective Compression**: Compresses the two optimization objectives of fairness and accuracy into one, simplifying the optimization process. 4. **Experimental Validation**: - Experiments on benchmark datasets show that DifFaiRec outperforms existing competitive baseline methods on multiple metrics. - The experimental results indicate that DifFaiRec is both accurate and fair. ### Conclusion DifFaiRec effectively addresses the issue of group unfairness in recommendation systems by combining a conditional diffusion model and a counterfactual module, while maintaining recommendation accuracy. This approach provides new insights and solutions for the study of fairness in recommendation systems.

DifFaiRec: Generative Fair Recommender with Conditional Diffusion Model

FairRec: Fairness Testing for Deep Recommender Systems

How Fair is Your Diffusion Recommender Model?

Configurable Fairness for New Item Recommendation Considering Entry Time of Items

Distributional Fairness-aware Recommendation

MultiFDF: Multi-Community Clustering for Fairness-Aware Recommendation

DiffuRec: A Diffusion Model for Sequential Recommendation

Diffusion Recommender Model

A Survey on Diffusion Models for Recommender Systems

Ensuring User-side Fairness in Dynamic Recommender Systems

FairDgcl: Fairness-aware Recommendation with Dynamic Graph Contrastive Learning

EDiffuRec: An Enhanced Diffusion Model for Sequential Recommendation

RecDiff: Diffusion Model for Social Recommendation

Fair Recommendations with Limited Sensitive Attributes: A Distributionally Robust Optimization Approach

Dual constraints and adversarial learning for fair recommenders

Improving Recommendation Fairness via Data Augmentation

Personalized Counterfactual Fairness in Recommendation

Self-supervised learning for fair recommender systems

Towards Long-term Fairness in Recommendation

Preference Diffusion for Recommendation

Dual Conditional Diffusion Models for Sequential Recommendation