Evaluation of graph embedding based reasoning over knowledge bases

Knowledge graphs (KG) include large amounts of structured data in many different domains. Knowledge or information is captured by entities and relationships among them in KG . One of the open problems in the knowledge graphs area is "link prediction", that is predicting new relationships or links among entities, given the existing entities and links in KG . A recent approach in graph-based learning problems is "graph embedding", in which graphs are represented as low-dimensional vectors. It is easier to make link predictions using these vector representations using this method. We also use graph embedding for graph representations. A sub-problem of link prediction in KG is link prediction in the presence of literal values, and specifically numeric values, on the receiving end of links. This creates a difficult situation as the numeric literal values take arbitrary values. There are several studies in this area, but they are all complex approaches. In this study, we propose a novel approach for link prediction in the presence of numerical values. We cluster the numerical values in graphs to enhance the prediction rates. We evaluated our method on Freebase knowledge graph, which includes entities, relations, and numeric literals. Test results show that a considerable increase in link prediction rate can be achieved in comparison to the other work.