This paper proposes a knowledge graph construction framework based on news articles to facilitate effective road-accident information management. The implementation of the proposed framework included three main stages. First, the news articles containing the keyword ‘traffic accident’ were collected, retaining only those that contain detailed accident description. Second, to effectively handle linguistic variations and ambiguities, large language models with chain-of-thought prompting were employed to extract structured accident information from news texts. Third, semantic embeddings were utilized to analyze similarities among accidents, thereby creating “SimilarTo” relationships to gain deeper insights and support enhanced analytical capabilities. The structured data thus obtained were converted into a knowledge graph (KG). Using the Neo4j graph database, a KG was constructed to represent traffic accidents that occurred in Korea during 2022. The practical relevance of the KG was verified through complex query tests across seven different scenarios with diverse combinations of entities and relationships. The results demonstrated that the proposed framework effectively structured detailed accident-related information extracted from news articles, demonstrating its potential to augment data sources for accident management, analysis and decision-making.
