TouriER: Temporal Knowledge Graph Completion by Leveraging Fourier Transforms

Abstract

In recent times, numerous studies on static knowledge graphs have achieved significant advancements. However, when extending knowledge graphs with temporal information, it poses a complex problem with larger data size, increased complexity in interactions between objects, and a potential for information overlap across time intervals. In this research, we introduce a novel model called TouriER, based on the MetaFormer architecture, to learn temporal features. We also apply a data preprocessing method to integrate temporal information in a reasonable manner. Additionally, the utilization of Fourier Transforms has proven effective in feature extraction. Through experiments on benchmark datasets, the TouriER model has demonstrated better performance compared to well-known models based on standard metrics.