Knowledge entity learning and representation for ontology matching based on deep neural networks

We study the task of ontology matching that is used mainly for solving the semantic heterogeneity problems, which concentrates on finding semantically related entities between different ontologies. Many previous works exploit the character-level or token-level information of the descriptions of an entity in ontology directly when applying the string-based matcher or token based matcher to find the corresponding entities. They ignored the higher level correlations between different descriptions of an entity. To address this problem, we propose a representation learning method based on deep neural networks which aim at learning the high level abstract representations of the input entity. Particularly, the representations of the entities are learned in an unsupervised way firstly, and then fine-tuned in a supervised manner with the training data. The experiment results show that our approaches can learn useful representations for entities from its descriptive information to better measure the similarity between entities.