Constructing Multimodal Wireless Knowledge Graphs for Large Language Model–Based Network Reasoning

The increasing complexity of wireless networks and the proliferation of heterogeneous data sources pose significant challenges to intelligent network management and decision-making. Although Large Language Models (LLMs) have demonstrated strong capabilities in natural language understanding and reasoning, their direct application in wireless network tasks is limited due to the lack of structured domain knowledge and reliable contextual foundations. This paper proposes an intelligent knowledge graph enhanced retrieval-enhanced generation (KG-RAG) framework for multimodal wireless networks. First, through structured entity and relationship modeling, multimodal wireless data including network measurements, configuration parameters and domain knowledge are transformed into a unified knowledge graph. The constructed knowledge graph, as a clear semantic backbone, supports efficient knowledge retrieval and reasoning. By integrating graph-based retrieval results with LLMs, the proposed framework achieves context-aware and interpretable network reasoning. The effectiveness of this method is evaluated through a case study based on lightweight network slicing. The experimental results show that in different LLMs, compared with the traditional text-based RAG method, the KG-enhanced RAG method improves the reasoning accuracy, reasoning consistency and interpretive integrity. These findings indicate that structured knowledge representation plays a crucial role in enhancing the reliability and interpretability of LLM-driven wireless network intelligence.