Tokenization, Fusion and Decoupling: Bridging the Granularity Mismatch Between Large Language Models and Knowledge Graphs

Researchers propose KGT, a novel framework that bridges the gap between Large Language Models and Knowledge Graph Completion by using dedicated entity tokens for full-space prediction. The approach addresses fundamental granularity mismatches through specialized tokenization, feature fusion, and decoupled prediction mechanisms.

• →KGT framework introduces dedicated entity tokens to enable efficient full-space prediction in Knowledge Graph Completion tasks.
• →The approach combines structural and textual features through relation-guided gating mechanisms without requiring training from scratch.
• →Decoupled prediction uses independent heads to separate semantic and structural reasoning processes.
• →Experimental results show KGT consistently outperforms existing state-of-the-art methods across multiple benchmarks.
• →The solution addresses the fundamental granularity mismatch between token-based LLMs and entity-based knowledge graphs.