ECS-KG: An event-centric semantic knowledge graph for event-related news articles

Abstract

Recent advances in deep learning techniques and contextual understanding render Knowledge Graphs (KGs) valuable tools for enhancing accessibility and news comprehension. Conventional and news-specific KGs frequently lack the specificity for efficient news-related tasks, leading to limited relevance and static data representation. To fill the gap, this study proposes an Event-Centric Semantic Knowledge Graph (ECS-KG) model that combines deep learning approaches with contextual embeddings to improve the procedural and dynamic knowledge representation observed in news articles. The ECS-KG incorporates several information extraction techniques, a temporal Graph Neural Network (GNN), and a Graph Attention Network (GAT), yielding significant improvements in news representation. Several gold-standard datasets, comprising CNN/Daily Mail, TB-Dense, and ACE 2005, revealed that the proposed model outperformed the most advanced models. By integrating temporal reasoning and semantic insights, ECS-KG not only enhances user understanding of news significance but also meets the evolving demands of news consumers. This model advances the field of event-centric semantic KGs and provides valuable resources for applications in news information processing.