Abstract:Large language model (LLM) demonstrates strong semantic understanding and knowledge generalization capabilities in natural language processing. However, when applied to specialized questions in vertical domains, general-purpose LLMs often exhibit deficiencies such as insufficient domain expertise, susceptibility to hallucinated facts, and limited interpretability. To address these challenges, this study proposes a knowledge-enhanced question-answering solution for vertical domains. In the data processing stage, a text segmentation strategy based on multi-feature semantic segmentation is adopted to construct a high-quality domain knowledge base. Subsequently, information filtering is designed to constrain the response scope of the question-answering model, and an entity extraction module is deeply integrated with a knowledge graph to achieve effective fusion of structured and unstructured knowledge, thereby improving the accuracy and professionalism of generated answers. The effectiveness of the segmentation strategy and the question-answering model is evaluated through both subjective and objective assessments, as well as ablation experiments. Experimental results show that the proposed approach achieves significant improvements in domain-specific reasoning capability, answer professionalism, and interpretability of generated results.