Knowledge Graph Self-Supervised Rationalization for Recommendation (2307.02759v1)

Abstract: In this paper, we introduce a new self-supervised rationalization method, called KGRec, for knowledge-aware recommender systems. To effectively identify informative knowledge connections, we propose an attentive knowledge rationalization mechanism that generates rational scores for knowledge triplets. With these scores, KGRec integrates generative and contrastive self-supervised tasks for recommendation through rational masking. To highlight rationales in the knowledge graph, we design a novel generative task in the form of masking-reconstructing. By masking important knowledge with high rational scores, KGRec is trained to rebuild and highlight useful knowledge connections that serve as rationales. To further rationalize the effect of collaborative interactions on knowledge graph learning, we introduce a contrastive learning task that aligns signals from knowledge and user-item interaction views. To ensure noise-resistant contrasting, potential noisy edges in both graphs judged by the rational scores are masked. Extensive experiments on three real-world datasets demonstrate that KGRec outperforms state-of-the-art methods. We also provide the implementation codes for our approach at https://github.com/HKUDS/KGRec.

An Assessment of Knowledge Graph Self-Supervised Rationalization for Recommendation

This paper introduces KGRec, a recommendation framework based on integrating self-supervised learning into knowledge graph-augmented recommender systems. The work explores the rationale behind user-item interactions, leveraging a comprehensive self-supervised approach to enhance recommendation accuracy by discerning pertinent knowledge graph (KG) connections.

Research Contributions and Methodology

KGRec distinguishes itself by substituting random augmentation for a more nuanced approach in self-supervised learning paradigms, notably employing an innovative method called rational masking. This involves generating rationale scores for knowledge triplets to identify and represent informative connections. KGRec encapsulates two primary tasks as part of this method: a generative task and a contrastive learning task.

The generative component employs a masking-reconstructing strategy where knowledge triplets with high rationale scores are masked during training. The model is subsequently tasked with reconstructing these masked connections, effectively highlighting key knowledge paths within the KG. The resulting self-supervised learning process strengthens the model's ability to discern significant relational patterns that contribute most to recommendation tasks.

Complementing the generative task, KGRec incorporates a contrastive learning framework that aligns collaborative signals from user-item interactions with those from the KG. This task involves masking potential noise within both graphs to ensure robust contrastive learning, thereby enhancing the quality of entity representations.

Experimental Validation

The authors assert that KGRec achieves significant improvements over state-of-the-art methods, citing experiments conducted on three widely recognized datasets: Last-FM, MIND, and Alibaba-iFashion. By implementing the rationale-aware masking mechanism and cross-view contrastive tasks, KGRec is demonstrated to outperform other knowledge-aware recommenders in both recall and normalized discounted cumulative gain (NDCG) metrics.

The evaluation results are particularly noteworthy in cold-start and long-tail item scenarios, typical areas of weakness for many recommendation models. KGRec effectively manages data sparsity by leveraging knowledge graphs efficiently while maintaining interpretability in its recommendation process.

Theoretical and Practical Implications

The research underscores the necessity of integrating knowledge graphs into recommendation engines with a fine-grained, rationale-driven approach. The authors effectively demonstrate that self-supervised learning, when appropriately guided by task-related rationales, can markedly improve recommendation accuracy. Practically, this implies that organizations can achieve better alignment of recommendation results with user preferences, a critical factor in increasing user engagement and satisfaction.

Future developments in AI applications can draw from this work to explore more sophisticated graph representation learning strategies. The potential to personalize interaction-based services across various sectors, like e-commerce and content streaming, augments the significance of such methodological innovations.

Conclusion

KGRec presents a compelling advancement in knowledge-enhanced recommender systems by embedding self-supervised rationalization into their core functionality. The integration of rational masking and rationality-aware contrastive learning defines new horizons for achieving efficient and interpretable recommendation systems. The paper serves as a methodological reference for researchers aiming to advance the state-of-the-art in knowledge-aware recommendation frameworks, making a pivotal contribution to the field of recommender systems.