Abstract

Large language models (LLMs) have made significant progress ingeneral-purpose natural language processing tasks. However, LLMs are stillfacing challenges when applied to domain-specific areas liketelecommunications, which demands specialized expertise and adaptability toevolving standards. This paper presents a novel framework that combinesknowledge graph (KG) and retrieval-augmented generation (RAG) techniques toenhance LLM performance in the telecom domain. The framework leverages a KG tocapture structured, domain-specific information about network protocols,standards, and other telecom-related entities, comprehensively representingtheir relationships. By integrating KG with RAG, LLMs can dynamically accessand utilize the most relevant and up-to-date knowledge during responsegeneration. This hybrid approach bridges the gap between structured knowledgerepresentation and the generative capabilities of LLMs, significantly enhancingaccuracy, adaptability, and domain-specific comprehension. Our resultsdemonstrate the effectiveness of the KG-RAG framework in addressing complextechnical queries with precision. The proposed KG-RAG model attained anaccuracy of 88% for question answering tasks on a frequently usedtelecom-specific dataset, compared to 82% for the RAG-only and 48% for theLLM-only approaches.