Abstract:Fairness-aware recommendation eliminates discrimination issues to build trustworthy recommendation systems.Explaining the causes of unfair recommendations is critical, as it promotes fairness diagnostics, and thus secures users' trust in recommendation models. Existing fairness explanation methods suffer high computation burdens due to the large-scale search space and the greedy nature of the explanation search process. Besides, they perform score-based optimizations with continuous values, which are not applicable to discrete attributes such as gender and race. In this work, we adopt the novel paradigm of counterfactual explanation from causal inference to explore how minimal alterations in explanations change model fairness, to abandon the greedy search for explanations. We use real-world attributes from Heterogeneous Information Networks (HINs) to empower counterfactual reasoning on discrete attributes. We propose a novel Counterfactual Explanation for Fairness (CFairER) that generates attribute-level counterfactual explanations from HINs for recommendation fairness. Our CFairER conducts off-policy reinforcement learning to seek high-quality counterfactual explanations, with an attentive action pruning reducing the search space of candidate counterfactuals. The counterfactual explanations help to provide rational and proximate explanations for model fairness, while the attentive action pruning narrows the search space of attributes. Extensive experiments demonstrate our proposed model can generate faithful explanations while maintaining favorable recommendation performance.

What problem does this paper attempt to address?

### Problems Addressed by the Paper The paper aims to address the issue of fairness in recommendation systems, particularly by providing counterfactual explanations to diagnose the reasons for unfair recommendations. Specifically, the paper focuses on how to improve the fairness of recommendation systems by minimizing attribute changes and proposes a new method—**Counterfactual Explanation for Fairness (CFairER)**. #### Main Issues 1. **High Computational Cost**: Existing fairness explanation methods lead to high computational costs due to the large search space and greedy explanation search process. 2. **Not Suitable for Discrete Attributes**: Existing methods are mainly based on continuous value optimization and cannot handle discrete attributes (such as gender, race, etc.) well. 3. **Reliability of Explanations**: Existing methods may introduce pseudo-explanations because they rely on greedy selection mechanisms. #### Solutions 1. **Adopting Counterfactual Explanations**: The paper utilizes counterfactual explanations from causal inference to find minimal attribute changes that alter the model's fairness. 2. **Using Heterogeneous Information Network (HIN)**: By extracting real-world attributes from HIN to handle discrete attributes, it overcomes the limitations of score-based optimization methods. 3. **Reducing Search Space**: Proposes an attention action pruning method to reduce the search space of candidate counterfactual explanations. 4. **Reinforcement Learning Framework**: Optimizes the explanation strategy in an off-policy reinforcement learning environment to generate high-quality counterfactual explanations. #### Method Overview - **Counterfactual Fairness Explanation Model (CFE)**: Optimizes the explanation strategy through reinforcement learning to generate attribute-level counterfactual explanations. - **Graph Representation Module**: Embeds users, items, and their attributes into dense vectors for state representation learning and action selection. - **Recommendation Model**: Uses Matrix Factorization (MF) to generate the original recommendation results. #### Contributions - Proposes using rich attributes in HIN to provide attribute-level counterfactual explanations. - Designs an off-policy learning framework to identify optimal counterfactual explanations. - Proposes a counterfactual risk minimization method for unbiased policy optimization. - Experiments demonstrate that the method maintains good recommendation performance while generating credible explanations. In summary, the paper aims to address the fairness issue in recommendation systems through counterfactual explanation methods and proposes a series of innovative solutions.

Counterfactual Explanation for Fairness in Recommendation

Counterfactual Explanation for Fairness in Recommendation

Explainable Fairness in Recommendation

Configurable Fairness for New Item Recommendation Considering Entry Time of Items

Personalized Counterfactual Fairness in Recommendation

Learning-based Counterfactual Explanations for Recommendation

Reinforced Path Reasoning for Counterfactual Explainable Recommendation

Fairness and Explainability: Bridging the Gap Towards Fair Model Explanations

Toward Explainable Recommendation Via Counterfactual Reasoning.

Fairness-Aware Explainable Recommendation over Knowledge Graphs

Recommendation with Causality Enhanced Natural Language Explanations

On the Relationship Between Explanations, Fairness Perceptions, and Decisions

GNNUERS: Fairness Explanation in GNNs for Recommendation via Counterfactual Reasoning

Constrained Off-policy Learning over Heterogeneous Information for Fairness-aware Recommendation

Interact with the Explanations: Causal Debiased Explainable Recommendation System.

Explanations, Fairness, and Appropriate Reliance in Human-AI Decision-Making

FairGap: Fairness-aware Recommendation via Generating Counterfactual Graph

On the Relationship Between Counterfactual Explainer and Recommender

Average User-side Counterfactual Fairness for Collaborative Filtering

A Counterfactual Explanation Method Based on Modified Group Influence Function for Recommendation

Counterfactual Graph Augmentation for Consumer Unfairness Mitigation in Recommender Systems