Constructing Spatio-temporal Disaster Knowledge Graph from Social Media

Abstract. Social media enables the disclosure of real-time crowd situations and provides high accessibility to the public. Thus, social media has emerged as a promising resource for discovering and managing disasters. This study aims to realize an effective process of constructing a spatio-temporal disaster knowledge graph (ST-DKG) from social media for disaster semantic interpretation and achieving an effective solution for disaster management. ST-DKG is constructed using PTT, one of Taiwan’s most popular social network platforms, and natural language processing with artificial intelligence (AI) models based on Bidirectional Encoder Representations from Transformers (BERT). ST-DKG addresses issues such as ellipsis and coreference resolution, entity recognition, relation extraction, and the identification of subject-predicate-object triples. Our method not only enhances the access efficiency and interoperability of social awareness information, but also provides bottom-up spatio-temporal disaster relevant knowledge for disaster management.