Temporal knowledge graph diffusion model for open-world reasoning

Yuhan Wang, Xiang Zhao, Zhen Tan, Weidong Xiao, Xueqi Cheng

Abstract: In recent years, temporal knowledge graphs (TKGs) have emerged as a prominent research area in artificial intelligence and knowledge engineering, offering notable potential for various applications. However, as the scope of TKG applications continues to expand, distinct challenges arise in specific contexts, especially in open-world scenarios. Traditional representation and reasoning models struggle to capture the associations of unseen entities within the knowledge graph, making it difficult to accurately represent these entities and perform unseen entity prediction tasks in open-world settings. To address these challenges, this paper proposes a novel reasoning framework, the diffusion model-based unseen entity prediction method (DM-UEP). This innovative approach generates virtual representations for unseen entities using a diffusion model and establishes a two-phase reasoning framework. The framework includes a graph extension and a transformer with reference to human cognitive reasoning. In addition, we reconstruct three test datasets tailored specifically for unseen entity prediction by leveraging existing public datasets. These datasets demonstrate the superiority of the proposed DM-UEP method in tackling the specialized task of unseen entity prediction in open-world scenarios.

External IDs:dblp:journals/chinaf/WangZTXC25