IME: Integrating Multi-curvature Shared and Specific Embedding for Temporal Knowledge Graph Completion

TL;DR

This work tackles temporal knowledge graph completion by addressing the limitations of single-space embeddings in capturing complex, evolving geometry. It proposes IME, a framework that learns multi-curvature embeddings across hyperspherical, hyperbolic, and Euclidean spaces, augmented with space-shared and space-specific representations and Adjustable Multi-curvature Pooling. The model introduces similarity, difference, and structure losses to align cross-space information and preserve geometric structure, achieving competitive improvements on ICEWS14, ICEWS05-15, and GDELT. By enabling richer geometric modeling and cross-space fusion, IME advances TKGC accuracy and robustness for real-world temporal reasoning tasks.

Abstract

Temporal Knowledge Graphs (TKGs) incorporate a temporal dimension, allowing for a precise capture of the evolution of knowledge and reflecting the dynamic nature of the real world. Typically, TKGs contain complex geometric structures, with various geometric structures interwoven. However, existing Temporal Knowledge Graph Completion (TKGC) methods either model TKGs in a single space or neglect the heterogeneity of different curvature spaces, thus constraining their capacity to capture these intricate geometric structures. In this paper, we propose a novel Integrating Multi-curvature shared and specific Embedding (IME) model for TKGC tasks. Concretely, IME models TKGs into multi-curvature spaces, including hyperspherical, hyperbolic, and Euclidean spaces. Subsequently, IME incorporates two key properties, namely space-shared property and space-specific property. The space-shared property facilitates the learning of commonalities across different curvature spaces and alleviates the spatial gap caused by the heterogeneous nature of multi-curvature spaces, while the space-specific property captures characteristic features. Meanwhile, IME proposes an Adjustable Multi-curvature Pooling (AMP) approach to effectively retain important information. Furthermore, IME innovatively designs similarity, difference, and structure loss functions to attain the stated objective. Experimental results clearly demonstrate the superior performance of IME over existing state-of-the-art TKGC models.

Figures (5)

• Figure 1: A brief description of IME. Learning multi-curvature representations through space-shared and space-specific properties. These features are later utilized for subsequent predictions by the adjustable multi-curvature pooling.
• Figure 2: The framework of IME. Specifically, IME models the query (Albert Einstein, Born In, ?, 1879-3-14) in multi-curvature spaces through information aggregation and information distribution. Subsequently, IME explores space-shared and space-specific properties to learn the commonalities and characteristics across different curvature spaces, effectively reducing spatial gaps among them. Finally, these identified features are employed for adjustable multi-curvature pooling in subsequent predictions.
• Figure 3: Comparison of different pooling approaches.
• Figure 4: H@1 with varying loss weights $\alpha$, $\beta$ and $\gamma$ on ICEWS14.
• Figure 5: Comparison of MRR performance with different embedding dimensions on ICEWS14.