Abstract: To address the problems of over-reliance on field experience, insufficient matching accuracy between failure characteristics and handling measures, and delayed decision-making responses in pipe sticking handling during drilling engineering, an intelligent generation method for pipe sticking handling measures based on semantic retrieval and graph reasoning was proposed. First, a pipe sticking knowledge graph incorporating failure characteristics, handling measures, and handling effects was constructed by integrating multi-source drilling engineering data, enabling structured organization and relational representation of domain knowledge.On this basis, a five-step retrieval process of “segmentation–matching–identification–retrieval–generation” was proposed. The user’s unstructured failure description was segmented into semantic short sentences. BGE-M3 vector encoding and Chroma indexing were employed for short sentence-to-node semantic matching. Hybrid classification combining feature engineering and a large language model was used to infer the pipe sticking type. Multi-hop retrieval was performed in the Neo4j graph to obtain a structured handling subgraph, which was then fed into a large language model to generate handling recommendations. Meanwhile, a three-dimensional evaluation model encompassing “operational rationality–hierarchical rationality–completeness” was established to quantitatively assess the generated measures and provide graded recommendations. Evaluation results from field pipe sticking cases indicated that the handling measures generated by this method achieved an accuracy of 94.5%, which represented a 29.0 percentage point improvement over the pure large language model approach, thereby validating the effectiveness and engineering feasibility of the method. The study demonstrates that the deep integration of semantic retrieval and graph reasoning can reduce the dependence of pipe sticking handling on field experience, improve the matching accuracy between failure characteristics and handling measures, and enhance decision-making response efficiency, providing a novel knowledge-traceable and engineering-applicable technical solution for intelligent emergency management in drilling engineering.