Toward Explainable Recommendation via Counterfactual Reasoning

Recently, counterfactual explanation models have shown impressive performance in adding explanations to recommendation systems. Despite their effectiveness, most of these models neglect the fact that not all aspects are equally important when users decide to purchase different items. As a result, the explanations generated may not reflect the users' actual preferences. Furthermore, these models typically rely on external tools to extract aspect-level representations, making the model's explainability and recommendation performance are highly dependent on external tools. This study addresses these research gaps by proposing a co-attention-based fine-grained counterfactual explanation model that uses co-attention and aspect representation learning to directly capture user preferences toward different items for recommendation and explanation. The superiority of the proposed model is demonstrated through extensive experiments.