This work focuses on modeling multi-relational data of knowledge graphs in Neural Network, with the goal of providing an efficient model to complete them by automatically adding missing types of entities, without requiring extra knowledge. This neural-based representation learning method represents knowledge graph data with into low-dimensional vector space representations of both entities and relations where the vector similarity can be regarded as semantic similarity. The novel part of this work would be relation-based neighbourhood clustering into consideration the neural network model. For this the model able to utilize inference patterns of the type object by learning semantic relationships between type object and relation present in knowledge graphs for effective and efficient type inference. We will define margin-based score function as objective for training, and learning process will be carried out using stochastic gradient descent (SGD).
mzkhan2000/TypeNeuralModel
This work focuses on modeling multi-relational data of knowledge graphs in Neural Network, with the goal of providing an efficient model to complete them by automatically adding missing types of entities, without requiring extra knowledge. This neural-based representation learning method represents knowledge graph data with into low-dimensional vector space representations of both entities and relations where the vector similarity can be regarded as semantic similarity. The novel part of this work would be relation-based neighbourhood clustering into consideration the neural network model. For this the model able to utilize inference patterns of the type object by learning semantic relationships between type object and relation present in knowledge graphs for effective and efficient type inference. We will define margin-based score function as objective for training, and learning process will be carried out using stochastic gradient descent (SGD).
Python