Topic: Heterogeneous Information Fusion for Recommender Systems: Knowledge, Modalities, and Behaviors

Recommender systems have become fundamental tools for mitigating information overload in the digital era. Despite their success, traditional collaborative filtering methods frequently encounter significant challenges, including data sparsity and the cold-start problem, due to their reliance on limited user-item interaction records. To address these bottlenecks, contemporary research has increasingly shifted towards heterogeneous information fusion, a paradigm that integrates diverse auxiliary data to enhance the quality of representation learning.

This special issue aims to explore advanced methodologies for synthesizing complex data sources beyond simple identifier matching. We specifically focus on the integration of knowledge graphs to capture semantic item relationships, multi-modal features to enrich content understanding, and multi-behavior data to model fine-grained user interests. Additionally, we encourage investigations into cross-domain transfer learning and sequential modeling that leverage these heterogeneous signals to improve accuracy and robustness. We invite researchers to submit original contributions that propose novel architectures and algorithmic frameworks to effectively fuse knowledge, modalities, and behaviors, thereby advancing the theoretical and practical frontiers of recommendation technology.

Knowledge-enhanced recommendation, multi-behavior recommendation, multi-modal recommendation, cross-domain recommendation; sequential recommendation, graph neural networks, collaborative filtering, heterogeneous information network