Triple Context-Based Knowledge Graph Embedding

Abstract

Knowledge graph embedding aims to represent entities and relations of a knowledge graph in continuous vector spaces. It has increasingly drawn attention for its ability to encode semantics in low dimensional vectors as well as its outstanding performance on many applications, such as question answering systems and information retrieval tasks. Existing methods often handle each triple independently, without considering context information of a triple in the knowledge graph, such an information can be useful for inference of new knowledge. Moreover, the relations and paths between an entity pair also provide information for inference. In this paper, we define a novel context-dependent knowledge graph representation model named triple-context-based knowledge embedding, which is based on the notion of triple context used for embedding entities and relations. For each triple, the triple context is composed of two kinds of graph structured information: one is a set of neighboring entities along with their outgoing relations, the other is a set of relation paths which contain a pair of target entities. Our embedding method is designed to utilize the triple context of each triple while learning embeddings of entities and relations. The method is evaluated on multiple tasks in the paper. Experimental results reveal that our method achieves significant improvements over the state-of-the-art methods.

Keywords

• Knowledge graph embedding
• semantic Web
• link prediction