自适应分支求解策略增强的智能建造政策类文本检索方法OA

Accurate interpretation and application of intelligent construction policies are essential throughout the project lifecycle,as they directly affect compliance,technology selection,and project evaluation.However,policy retrieval in this domain is challenged by lengthy documents,specialized terminology,and regional variations,making conventional retrieval methods insufficient for complex reasoning tasks.Although retrieval-augmented generation(RAG)has shown promise,its direct application to intelligent construction remains limited by weak reasoning ability and difficulty in resolving semantic ambiguity.To address these limitations,this paper proposes Branch-RAG,an improved RAG framework based on an adaptive branch resolution strategy for intelligent construction policy question and answer.By integrating a policy knowledge graph as external knowledge,Branch-RAG dynamically enhances logical reasoning according to question granularity.A policy knowledge base covering 13 provinces and 24 prefecture-level cities in China was constructed,together with a dataset of 3 420 structured question-answer pairs.Experimental results show that Branch-RAG achieves an overall accuracy of 79.6％,outperforming the second-best Graph-RAG by 3.3％,while maintaining an average response time of 14.82 s.It performs particularly well on high-difficulty questions.This paper provides an efficient and accurate solution for complex policy retrieval in intelligent construction and supports the practical deployment of RAG in vertical domains.