Numerical Knowledge Representation Learning and Link Prediction over Knowledge Graph

Link prediction has been extensively studied in knowledge graph (KG) completion tasks, whereas numerical attribute relation prediction studies are still relatively scarce due to the non-discrete challenges. Numerical attribute relations are pervasive in real-world KGs (e.g., relations in product graphs such as “car, hasMaxPower, 103”), which imbue entities with rich semantics. This paper presents a novel framework for numerical relation prediction. The framework comprises a numerical relation knowledge representation method (NumKR) and a numerical relation prediction framework based on GEN. The aim is to enhance the accuracy of the numerical link prediction model in a realworld product knowledge graph. The experimental results demonstrate that our model outperforms the simple KGE model in terms of link prediction performance.