[Objective] Earthquakes pose a major threat to the safety of human life and the integrity of property, while they are also considered to be one of the main catalysts of various geological disasters. Given their potential impact, analysing and integrating information related to earthquake disaster has become an important task in supporting effective decision-making processes. However, earthquake disaster observation data are usually heterogeneous from multiple sources, which contain different scientific knowledge and low correlation between the data. This makes it difficult to use the information effectively for integration and querying. [Methods] A promising approach to these challenges is the use of the Semantic Web, of which the most exemplary approach is Knowledge Graph. Knowledge graph offer significant advantages in fusing and representing heterogeneous data from multiple platforms and disciplines. Considering the need for consistency between entity structures, while recognizing the diversity of information in the earthquake disaster domain, a structured approach was adopted. Firstly, a top-down approach was used to provide a systematic overview of concepts related to the earthquake disaster domain. This included the construction of a comprehensive ontology with several key components: earthquake disaster data, geological and geographic contexts, specific earthquake disaster events, emergency response tasks associated with these disasters, and models relevant to understanding and predicting earthquake behaviors. An earthquake disaster ontology layer was created. A bottom-up approach was adopted, aiming to create a high-quality data layer. In order to analyze the surface changes caused by earthquake activity before and after the disaster, a convolutional neural network was utilized to achieve a complex transformation from image data to structured textual knowledge based on remote sensing data. In addition, a fine-tuned Universal Information Extraction (UIE) pre-trained model was used. The model helps in recognizing named entities and extracting relational attributes from unstructured text data with extraction accuracies of 82.04% and 70.66% respectively. After the extraction phase, data fusion and unified representation is achieved by evaluating the semantic similarity of word vectors corresponding to the extracted entities and attributes. [Results] Taking the earthquake of 18 December 2023 in Jieshishan County, Linxia Prefecture, Gansu Province as an example, a high-quality earthquake disaster knowledge graph is formed through ontology construction, data extraction, and unified expression, and the transformation from multi-source heterogeneous seismic data of earthquake disaster to unified knowledge expression is achieved. [Conclusion] By building a knowledge graph of earthquake disaster, functions such as querying disaster loss data, providing decision support for the entire emergency response chain, and facilitating reasoning and querying about potential secondary hazards were realized. This reasoning capability is enhanced by integrating relevant geological data. The innovative approach presented in this paper utilizes deep learning techniques and pre-trained models to effectively fuse multimodal data and ultimately support the construction of an earthquake disaster knowledge graph, which helps to improve the ability to quickly and accurately query earthquake disaster information, thereby supporting proactive measures to respond to the occurrence of secondary disasters. This integrated approach not only facilitates immediate post-disaster analyses, but also lays the foundation for future advances in disaster management and response methods.
DownLoad: