DSE-RE: Dynamic Schema Evolution with LLM for Iterative Relation Extraction

Knowledge graphs serve as pivotal tools for organizing fragmented information into structured representations. Relation extraction (RE) is a key technology for knowledge graph construction that identifies semantic connections between entities. In RE methodologies, supervised learning is often constrained in low-resource or rapidly evolving domains because it requires extensive labeled datasets. Large Language Models (LLMs) exhibit strong generalization capabilities; however, they remain susceptible to output inconsistency, hallucination, and schema drift. To address these issues, this paper proposes DSE-RE (Dynamic Schema Evolution with LLM for Iterative Relation Extraction). DSE-RE dynamically adapts to diverse domains for iterative relation extraction and knowledge graph construction. Rather than relying on manual ontology design or extensive model training, DSE-RE employs a closed-loop feedback mechanism. It dynamically constructs and optimizes entity type tables and relation type tables, and embeds them into LLM prompts as contextual guidance. A graph-based semantic clustering algorithm is designed to merge synonymous redundancies and control schema expansion. Experiments on the public FOBIE benchmark and the self-constructed Huizhou-style Architecture dataset (HSA) demonstrate the effectiveness of the proposed framework. DSE-RE achieves competitive performance with minimal manually annotated data. After incorporating DSE-RE, the baseline LLMs achieved improved performance, with F1 scores increasing by approximately 3% on FOBIE and 6.8% on HSA. Compared to supervised deep learning models, DSE-RE exhibits improved adaptability in low-resource scenarios. However, its performance is bounded by the quality and coverage of the underlying LLM, and the current approach has not been evaluated on highly heterogeneous multi-lingual corpora. This study also demonstrates practical utility by constructing a domain-specific knowledge graph for Huizhou-style architecture.