.png){kind=icon}
Return
Utilizing temporal knowledge graphs for disaster detection and analysis
.png){kind=icon}
PRE
AI
DOI：10.1016/j.knosys.2026.117058.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
As various types of disasters occur more frequently, leading to increased casualties and property damage, it is essential to analyze large volumes of disaster data to enable rapid, accurate detection and response. However, disaster-related data comes from diverse sources and is distributed across different agencies, making real-time integrated analysis difficult and limiting both initial response and situational awareness. This article presents Temporal Knowledge Graphs (TKGs) constructed from social media, public, and news data to represent relationships among disaster-related data. The proposed approach captures temporal changes and extracts meaningful insights from evolving disaster data. We employ social data subgraphs to evaluate disaster detection capabilities, using four Graph Neural Network (GNN) models: Graph Convolutional Network (GCN), SAmple and aggreGatE (GraphSAGE), Graph Attention Network (GAT), and Gated Graph Convolutional Network (GGCN). The experimental results are promising, achieving an F1-score of approximately 82%. In addition, performance evaluation with K-means clustering as a baseline confirms the superior disaster detection capability of the GNN models. To extend the analysis beyond short-term detection, we integrate multiple knowledge graphs (KGs) for long-term analysis, forming a unified KG. This approach aggregates news data subgraphs to construct merged graphs and apply the Named Entity Recognition (NER) model to extract disaster-related entities. The evaluation process assesses the NER model’s performance using the extracted entities, demonstrating its ability to identify unseen entities and automatically extract key information. The results demonstrate that the proposed approach effectively supports disaster situation analysis and identifies cascading disaster events, providing a comprehensive understanding of disaster dynamics.
Keywords:
Temporal knowledge graphs
Graph neural networks
Graph integration
Disaster detection
Situation analysis
Journal
K
IF:
7.6
Papers:
1.3W
Citations:
4.5W
Organization
No organization information available
Cited Papers
No cited papers available