EOAC-LLM: An LLM-driven event ontology automatic construction system

As a core knowledge carrier describing the dynamic behaviors, temporal correlations, and causal logic of event entities, event ontology is indispensable for intelligent decision-making, public opinion analysis, and emergency response. However, existing domain-specific event ontology construction methods face critical limitations: low automation efficiency, lack of a systematic end-to-end framework, and insufficient dynamic update capabilities, failing to adapt to the high-timeliness requirements of complex scenarios like emergency response. To address these gaps, this study aims to develop a fully automated, LLM-driven event ontology construction approach that integrates deep semantic modeling and dynamic evolution. The proposed method encompasses five core functional modules—event ontology modeling corpus generation, event element-relation extraction, deep semantic representation and aggregation, event ontology generation, and dynamic update—forming a closed-loop systematic framework. Its key novelties include: (1) a full-process automated framework eliminating reliance on manual annotation or expert knowledge. (2) a multi-dimensional dynamic aggregation method for semantic-temporal-relation to achieve accurate semantic mapping. (3) an event gap identification and incremental update mechanism enabling self-driven dynamic evolution. Validated in the emergency disaster relief domain with three mainstream LLMs (Deepseek, Doubao, ERNIE Bot), the approach outperforms baseline methods by over 15% in event extraction F1-score, achieves superior clustering performance, and realizes fully automated ontology construction with dynamic adaptability. This work provides high-quality, efficient ontology support for dynamic knowledge-intensive scenarios. • Propose a large-model-driven full-process automated framework for event ontology. • Design a semantic temporal relationship multi space fusion model, combined with dynamic weighting to enhance the completeness of event feature characterization. • Innovatively propose a temporal relationship-aware hierarchical clustering algorithm, which addresses the uneven granularity of event clusters in traditional hierarchical clustering by integrating temporal consistency scores and dynamic weight adaptation. • Build an event gap identification and incremental update mechanism to ensure the dynamic adaptation of the ontology.