Attention Is Not the Only Choice: Counterfactual Reasoning for Path-Based Explainable Recommendation

Yicong Li,Xiangguo Sun,Hongxu Chen,Sixiao Zhang,Yu Yang,Guandong Xu

DOI: https://doi.org/10.1109/TKDE.2024.3373608

2024-03-03

Abstract:Compared with only pursuing recommendation accuracy, the explainability of a recommendation model has drawn more attention in recent years. Many graph-based recommendations resort to informative paths with the attention mechanism for the explanation. Unfortunately, these attention weights are intentionally designed for model accuracy but not explainability. Recently, some researchers have started to question attention-based explainability because the attention weights are unstable for different reproductions, and they may not always align with human intuition. Inspired by the counterfactual reasoning from causality learning theory, we propose a novel explainable framework targeting path-based recommendations, wherein the explainable weights of paths are learned to replace attention weights. Specifically, we design two counterfactual reasoning algorithms from both path representation and path topological structure perspectives. Moreover, unlike traditional case studies, we also propose a package of explainability evaluation solutions with both qualitative and quantitative methods. We conduct extensive experiments on three real-world datasets, the results of which further demonstrate the effectiveness and reliability of our method.

Information Retrieval

What problem does this paper attempt to address?

### Problems Addressed by the Paper

The paper primarily focuses on enhancing the interpretability of recommendation systems. Specifically, the authors point out that current attention mechanism-based recommendation systems, while performing well in terms of accuracy, have significant shortcomings in interpretability. The main issues include:

1. **Unreliable Attention Mechanism**: Existing attention mechanisms exhibit unstable weight distributions across different experimental runs, making it difficult for users to accept the explanations of the recommendation results.
2. **Frequent Path Preference**: Attention mechanisms tend to assign higher weights to common frequent paths, ignoring unique paths that carry important explanatory information.

To address these issues, the paper proposes a new method based on counterfactual reasoning for explaining path-based recommendation systems. The specific methods are as follows:

1. **Counterfactual Reasoning Framework**: By adding slight perturbations to observe changes in recommendation scores, the framework identifies paths that significantly impact the recommendation results.
2. **Two Counterfactual Reasoning Algorithms**: Two counterfactual reasoning algorithms are designed from the perspectives of path representation and path topology.
3. **Comprehensive Evaluation Scheme**: A comprehensive evaluation scheme, including both qualitative and quantitative evaluation methods, is proposed to verify the effectiveness and reliability of the new method.

Through these methods, the paper aims to provide a more reliable and effective interpretability framework for recommendation systems.