Enhancing the Comprehensiveness of Criteria-Level Explanation in Multi-Criteria Recommender System

Background: The explainability of recommender systems (RSs) is currently attracting significant attention. Recent research mainly focus on item-level explanations, neglecting the need to provide comprehensive explanations for each criterion. In contrast, this research introduces a criteria-level explanation generated in a content-based pardigm by matching aspects between the user and item. However, generation may fall short when user aspects do not match perfectly with the item, despite possessing similar semantics. Objective: This research aims to extend the aspect-matching method by leveraging semantic similarity. The extension provides more detail and comprehensive explanations for recommendations at the criteria level. Methods: An extended version of the aspect matching (AM) method was used. This method identified identical aspects between users and items and obtained semantically similar aspects with closely related meanings. Results: Experiment results from two real-world datasets showed that AM+ was superior to the AM method in coverage and relevance. However, the improvement varied depending on the dataset and criteria sparsity. Conclusion: The proposed method improves the comprehensiveness and quality of the criteria-level explanation. Therefore, the adopted method has the potential to improve the explainability of multi-criteria RSs. The implication extends beyond the enhancement of explanation to facilitate better user engagement and satisfaction.