Knowledge Graph-Infused Fine-Tuning for Structured Reasoning in Large Language Models

2025-08-20

AI Summary

To address reasoning chain fragmentation and insufficient entity-level semantic modeling in large language models (LLMs) for structured knowledge reasoning, this paper proposes a knowledge graph (KG)-enhanced fine-tuning framework. The method employs graph neural networks to encode KG topological structure and introduces a dynamic gating mechanism to adaptively fuse contextual representations with structured knowledge. A joint loss function is designed to simultaneously optimize task performance and KG structural alignment, mitigating representation space conflicts. Key contributions include: (i) a learnable, multi-source knowledge fusion gating architecture; and (ii) a joint optimization objective explicitly enforcing structural consistency. Extensive experiments demonstrate significant improvements over baselines on entity recognition, question answering, and text generation tasks. Moreover, the framework exhibits strong robustness and generalization across varying learning rates, KG coverage ratios, and structural perturbations.

Technology Category

• Machine Learning: Structured Learning
• Data Mining & Knowledge Management: Linked Open Data, Knowledge Graphs & KB Completion
• Natural Language Processing: Sentence-level Semantics, Textual Inference, etc.

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

This paper addresses the problems of missing reasoning chains and insufficient entity-level semantic understanding in large language models when dealing with tasks that require structured knowledge. It proposes a fine-tuning algorithm framework based on knowledge graph injection. The method builds on pretrained language models and introduces structured graph information for auxiliary learning. A graph neural network is used to encode entities and their relations, constructing a graph-based semantic representation. A fusion mechanism is then designed to jointly model the knowledge graph embeddings with the contextual representations from the language model. To enhance the robustness of knowledge integration, a gating mechanism is introduced to dynamically balance the contributions of linguistic semantics and structural knowledge. This effectively mitigates conflicts between different representational spaces. During training, a joint loss function is constructed to account for both task performance and structural alignment objectives. This helps improve the accuracy of entity prediction and semantic reasoning. The study also includes a series of systematic sensitivity experiments. It evaluates the effects of learning rate, graph coverage, and structural perturbations on model performance. The results further validate the effectiveness and stability of the proposed method across tasks such as entity recognition, question answering, and language generation. Experimental findings show that the proposed structure-aware fine-tuning framework significantly enhances the model's ability to represent complex semantic units. It demonstrates better semantic consistency and contextual logic modeling in scenarios involving structural reasoning and entity extraction.

Problem

• Addresses missing reasoning chains in LLMs for structured knowledge tasks
• Improves entity-level semantic understanding in language models
• Mitigates conflicts between linguistic and structural knowledge representations

Innovation

• Knowledge graph-infused fine-tuning for structured reasoning
• Graph neural network encodes entities and relations
• Gating mechanism balances linguistic and structural knowledge