QuatCNNEx: A Knowledge Graph Embedding Model based on Quaternion Convolutional Neural Network

Authors:

Abstract
We can infer the missing links in the Knowledge Graph from the known knowledge when the existing Knowledge Graph is incomplete. Knowledge Graph Embedding is a common method to solve this kind of problem, that is, using low dimensional embedding of entities and relations to infer the missing links. The existing Knowledge Graph Embedding methods either do not make full use of the interactive feature information of entities and relations or have weak modeling ability for composition relation. Therefore, the combination of quaternion and convolution neural network is explored to construct QuatCNNEx which is a new Knowledge Graph Embedding model. The quaternion used in the model encodes the triple of Knowledge Graph, the quaternion embedding of entities and relations are used to the input of convolution neural network, and the model combines the Hadamard product of quaternion to predict the missing links in Knowledge Graph. The results of the experiment show that QuatCNNEx can model the composition relations and predict the missing triples in the Knowledge Graph using the interactive feature information of entities and relations.

This publication has 14 references indexed in Scilit: