Knowledge graph embedding model for unknown facts

Abstract

Knowledge graph completion is a task that aims to predict missing triples by computing embeddings for entities and relations observed during learning. In the real-world task, new entities and relations are dynamically added to knowledge graphs. However, most knowledge graph completion methods cannot infer missing triples for the added entities and relations if they are less or non-existent in the knowledge graph. In this research, we constructed a model that can be trained on a small sample of data and can make embeddings for new entities and relations, which can also perform knowledge graph completion. Experimental results showed that the performance of the model was not as good as the existing knowledge completion model, but the modeling of multi-hop relations might lead to improved results.