A Novel LLM-Powered Enterprise-Specific Conversational Search Framework

Organizations are increasingly adopting chatbots to deliver enterprise-specific search services, providing conversational and contextually appropriate natural language responses to user queries. To meet complex demands and enhance operational efficiency, while overcoming the limitations of traditional retrieval methods, we present an advanced approach for developing a Large Language Model (LLM)-Powered Enterprise Conversational Search Framework (LECS). This framework employs a sophisticated mixedgranularity chunking strategy for processing both short and long texts, an advanced question-rewriting mechanism, and a comprehensive multi-source ranking system. Extensive experiments have demonstrated the superior performance of LECS, particularly in terms of recall@5 and answer accuracy.