SRREN:Self-aware and Relation-aware Recurrent Evolution Network for Temporal Knowledge Graph Reasoning

Knowledge Graph (KG) reasoning for predicting incomplete KGs has been extensively studied. However, the Temporal Knowledge Graph (TKG), which consists of KGs at various timestamps, still has significant potential for improvement in forecasting future facts. A comprehensive understanding of historical facts is crucial for accurately predicting future events. For the same timestamp, the concurrent facts have structural dependencies, and the facts have a certain order in adjacent times. At the same time, for different timestamps, the attribute features of KGs in the current timestamp have a strong correlation with the attribute features of KGs in the historical timestamp. To effectively capture and comprehend these relational features and enhance the accuracy of TKG predictions , we propose a new recursive algorithm, the Self-aware and Relation-aware Recurrent Evolution Network (SRREN). SRREN learns an evolutionary representation of time-stamped entities and relations by recursively modeling the sequence of KGs. Specifically, for evolutionary units, structural dependencies are captured through relationship awareness. At each timestamp within the TKG, the sequential character of the facts is captured by a self-aware mechanism of regression modeling and a recursive component of the gate. In addition, the addition of entity static attributes can also achieve better entity representation. Experiments on six benchmark datasets show that SRREN can improve the accuracy of predicting future facts, and its performance is better than most other current models.