Deep Convolutional Neural Network For Knowledge Graph Completion

Knowledge graphs provide structured information of entities and relations,generally encoded in the form of directed multi-relations graphs.This kind of relation knowledge representation can be used in several domains including artificial intelligence and information retrieval.With the large volume of new information created every day,determining the validity of information in a knowledge graph and filling in its missing parts are crucial tasks for many researchers and practitioners.To address this challenge,researchers deal with embedding a large scale knowledge graph composed of entities and relations into a continuous vector space.Recently,many research proposed many meaningful models to overcome this issue.According to the difference of the definition of score functions,embedding based approaches can be roughly divided into two groups,compositional methods and non-compositional methods.Non-compositional methods,such as Trans E,Trans R,can be computed efficiently but are generally consi dered to be less expressive.Compositional methods such as Rescal allows to capture rich interaction,but it requires a large number of parameters.In this paper,we use deep convolutional neural network to capture rich interaction between embedding vectors.Different from other compositional or non-compositional methods which use hand-crafted operator to calculate the interactions of embedding vectors,we utilize deep convolutional neural network to compute the degree of association between latent features of entities.In additon,we find similarity between entities as auxiliary information to improve link prediction.Finally we investigate the application of margin-based pairwise method to the entity and relation linking task.In experiment,our approach i s both effective and outperform well than most of baseline models.