Counterfactual Explainable Recommendation (2108.10539v3)

Abstract: By providing explanations for users and system designers to facilitate better understanding and decision making, explainable recommendation has been an important research problem. In this paper, we propose Counterfactual Explainable Recommendation (CountER), which takes the insights of counterfactual reasoning from causal inference for explainable recommendation. CountER is able to formulate the complexity and the strength of explanations, and it adopts a counterfactual learning framework to seek simple (low complexity) and effective (high strength) explanations for the model decision. Technically, for each item recommended to each user, CountER formulates a joint optimization problem to generate minimal changes on the item aspects so as to create a counterfactual item, such that the recommendation decision on the counterfactual item is reversed. These altered aspects constitute the explanation of why the original item is recommended. The counterfactual explanation helps both the users for better understanding and the system designers for better model debugging. Another contribution of the work is the evaluation of explainable recommendation, which has been a challenging task. Fortunately, counterfactual explanations are very suitable for standard quantitative evaluation. To measure the explanation quality, we design two types of evaluation metrics, one from user's perspective (i.e. why the user likes the item), and the other from model's perspective (i.e. why the item is recommended by the model). We apply our counterfactual learning algorithm on a black-box recommender system and evaluate the generated explanations on five real-world datasets. Results show that our model generates more accurate and effective explanations than state-of-the-art explainable recommendation models.

The paper "Counterfactual Explainable Recommendation" discusses the introduction of a novel framework named CountER for generating explainable recommendations using counterfactual reasoning from causal inference. The primary objective of CountER is to provide simplified yet effective explanations for recommendation decisions in black-box recommender systems. These explanations focus on user and system perspectives to improve understanding, transparency, and debugging.

Key Contributions and Methodologies:

1. Counterfactual Framework: CountER leverages counterfactual reasoning to generate explanations by modifying item aspects to observe changes in recommendation outcomes. The method employs a joint optimization approach, targeting a balance between complexity (minimal aspect changes) and strength (significant impact on decision reversal).
2. Complexity and Strength:

The framework mathematically defines two core properties for explanations: - Explanation Complexity (EC): Defined as the number of aspects altered and the magnitude of these alterations. It is quantified using a combination of $\ell_2$-norm and $\ell_0$-norm. - Explanation Strength (ES): Measures the extent to which an explanation influences the change in the recommendation decision, assessed via ranking score differences.

1. Optimization Scheme: Due to the non-differentiable nature of the original formulation, the authors introduce a relaxable optimization problem substituting $\ell_0$-norm with $\ell_1$-norm for complexity and utilize a hinge loss for enforcing explanation strength constraints.
2. Standard Evaluation Metrics:
   • User-Oriented Evaluation: Utilizes aspects positively mentioned in user reviews as ground truth to assess precision, recall, and $F_1$ scores of the explanations.
   • Model-Oriented Evaluation: Introduces metrics for Probability of Necessity (PN) and Probability of Sufficiency (PS) to quantitatively evaluate how explanations correlate with the model’s actual reasoning behind recommendations.
3. Extensive Experimentation: CountER is evaluated against three baselines across five datasets, showing superior performance in generating precise and effective explanations both from user and model perspectives.
4. Findings and Discussions:
   • Explanations of items higher on the recommendation list showed greater complexity, aligning with the notion that stronger recommendations require more robust justifications.
   • There exists an intricate relationship between explanation complexity and user-oriented performance, contrasting with explanation strength's impact on model-oriented evaluation.

Future Directions:

The paper suggests expanding CountER to consider more diverse data forms, such as visual or textual features, and applying counterfactual reasoning in other complex frameworks like knowledge graphs or graph neural networks.

CountER represents an advancement in creating interpretable recommendation systems, pushing towards more transparent AI solutions by integrating causal inference with recommendation models. This method not only aids users in understanding recommendations but also provides system designers with tools for insightful model diagnostics.