Generative artificial intelligence drives large model generation algorithm for enhancing Chinese modern traditional cultural knowledge

To address outdated knowledge, cultural-context mismatch, weak evidence linkage, and unsupported expansion in general-purpose large language model generation of traditional-cultural content, this study proposes a generative AI-driven knowledge-enhanced generation algorithm. The method decouples a frozen Qwen2.5-7B-Instruct generator from an external dynamic knowledge base and uses a four-field semantic tag scheme covering cultural category, historical stage, regional origin, and application scenario. Candidate evidence is ranked by vector similarity, graph relations, tag consistency, source credibility, and conflict penalties, then assembled into a constrained prompt without modifying the internal attention layers of the frozen model. An incremental update mechanism combines source verification, conflict detection, and expert review, while factual consistency, cultural-context adaptability, and evidence coverage are used for post-generation verification. Experiments cover classical texts, cultural heritage objects, intangible cultural heritage crafts, traditional Chinese opera, and folk culture, with 1,080 factual question-answering samples and 360 contextual generation tasks. The proposed system achieves 88.61% factual accuracy, 95.28% answer relevance, 94.17% evidence faithfulness, and 92.50% cultural-context adaptability. After knowledge updating, accuracy on update-dependent questions rises from 58.67% to 91.33%. These results indicate that the proposed framework provides a traceable, context-sensitive generation pipeline while retaining a frozen generation model; between-system differences are interpreted descriptively because complete paired baseline predictions were not retained.