Abstract—A large number of fact triples are contained in knowledge graph. However, the knowledge is not completed in large-scale knowledge graph at present, which calls for knowledge graph completion urgently. Most proposed approaches embed entities and relationships into a continuous low-dimensional vector space through a single factual triple information, the knowledge graph completion is achieved through vector calculation. Besides, features are extracted from the relation paths between entities, which are used for training classifiers to predict relationships between entities. In this work, we designed a model, which not only utilize the information of relation paths among fact triples and entities in knowledge graph, but also the description text information of entities in knowledge graph were utilized. Through these information of entities and the distributed expression of the relationships, the task of knowledge graph completion was achieved. Through extensive experimental evaluations, the proposed model proves to be with better retrieval performance compared to state-of-the-arts.
Index Terms—Knowledge graph, knowledge graph completion, link prediction, relation path.
The authors are with the Computer science and technology Department, Shandong University, Shandong, China (e-mail: luoqi0918@gmail.com, hxw@sdu.edu.cn).
[PDF]
Cite: Qi Luo and Xingwei Hao, "Relation Path Modeling with Entity Description for Knowledge Graph Completion," International Journal of Knowledge Engineering vol. 4, no. 2, pp. 76-80, 2018.
