A Multi-Strategy Retrieval-Augmented Generation Approach for Enhancing Network Management

With the fast growth of Large Language Models (LLMs), their use in network management has received more attention. While LLMs are good at understanding natural language, they still have clear limits in real-world use, especially in retrieving domain-specific knowledge and generating accurate responses. These problems make it hard for them to work well in intelligent network operations and maintenance (O&M). To address these problems, this study proposes a MultiStrategy RAG (Retrieval-Augmented Generation) approach that mixes hybrid retrieval, reranking methods, and real-time OSS. This method enhances the knowledge retrieval capabilities and response quality of LLMs in network management tasks by combining full-text and vector retrieval, using reranking to optimize candidate passages, and adding real-time data from the OSS system. Also, the method is tested using real network management datasets, and its result is evaluated by the RAGAS framework. Result shows that the method makes the retrieved knowledge more useful and the answers more correct. This study gives a useful way to build intelligent O&M systems in network management.