Dynamic Adaptive Chain Model of Knowledge Graph Representation Learning

Abstract

Knowledge graph representation learning models are mostly used for static data. When the data changes, the models cannot be adjusted dynamically as the data changes. The data is constantly changing in actual usage scenarios. However, most representation learning models focus on prediction accuracy and ignore the dynamic change of knowledge graph. A trained representation learning model is closed, but the facts in the real world are constantly changing. The knowledge graph of the real world should be open to realize the description of reality. This paper proposes a dynamic adaptive chain transfer model aim to deal with changing knowledge graph data. Our model realizes the dynamic increase and decrease of triples without retraining. We designed an experimental method to verify the validity of the model. Experimental results show that our model can achieve dynamic data changes and keep the performance of the original model.