An Intelligent Question-Answering System Based on Deep Hybrid Search and Knowledge Graph for the Ecological Knowledge

As a critical ecological security barrier in China, the Sanjiangyuan region has accumulated vast amounts of multi-source heterogeneous ecological data over the long term. This data holds significant scientific value for studying regional ecosystem changes and assessing the effectiveness of ecological protection. To address the issues of dispersed and inefficient access to the ecological data of Sanjiangyuan, as well as the limitations of traditional retrieval methods (which fail to uncover complex relationships between entities), this study designed and implemented an intelligent question-answering system based on deep hybrid search and knowledge graph for the ecological knowledge of Sanjiangyuan. The system builds an intelligent dual-path retrieval framework based on the ecological data of the Sanjiangyuan region from 2013 to 2023. On the one hand, for questions with clear entity relationship query intent, the system utilizes a large language model (DeepSeek) to automatically convert natural language into Cypher query statements, and performs precise structured queries in the pre-constructed ecological knowledge graph of the Sanjiangyuan to reveal the deep associations among entities. On the other hand, for open or factual questions, a deep hybrid search strategy is employed. This strategy performs semantic vector recall through the Bi-Encoder model, keyword recall through the BM25 algorithm, merges the results of the two paths using the reciprocal rank fusion algorithm, expands and optimizes the candidate set, and finally introduces the Cross-Encoder model to perform deep contextual relevance ranking on candidate answers, thereby ensuring answer precision. The core advantage of the system lies in its intent recognition and intelligent scheduling mechanism, which automatically determines the optimal query path based on the user's input question. Experimental tests show that the system is able to efficiently integrate and deeply analyze the complex ecological data in the Sanjiangyuan area, which provides useful exploration and reference for improving the efficiency of digital ecological protection work and intelligent information service in Sanjiangyuan.