DCGL framework merges LLMs and knowledge graphs for smarter recommendations

Current recommendation systems leveraging Knowledge Graphs (KGs) and Large Language Models (LLMs) face three key limitations: inadequate modeling of implicit semantic relationships beyond explicit KG links, suboptimal single-channel fusion of ID and LLM embeddings causing signal interference, and insufficient consideration of user-item interaction frequency variations. To address these, researchers from multiple institutions developed DCGL (Dual-Channel Graph Learning), a novel framework accepted at SIGIR 2026.

DCGL introduces three innovations. First, a dual-channel architecture structurally decouples rich semantic information from user behavioral patterns, preventing early interference. Second, a multi-level contrastive learning mechanism enhances robustness against KG noise through intra-view contrasts and bridges semantic gaps between channels via inter-view alignment. Third, a dynamic fusion mechanism adaptively balances semantic generalization and behavioral specificity based on interaction frequency, resolving the cascading limitation of static fusion. Extensive experiments on four real-world datasets show DCGL consistently outperforms state-of-the-art methods, yielding substantial improvements in sparse scenarios while maintaining precision for active users. The code is publicly available.