Deep reinforcement learning approach for ontology matching problem

The ontology matching is an active field of research, which is considered as a key solution to solve the semantic heterogeneity problem. Given two ontologies, the alignment process produces a set of matches each linking two entities. To address this issue, we have reformulated the alignment problem as a binary classification problem using different similarity measures between different ontologies entities as features. In our approach, we propose a model based on deep reinforcement learning, using deep Q-learning network. In this model, the classification action is carried out by the agent at each time step, and the environment evaluates the agent's decisions and returns a reward to this latter. In order to improve the performance of our model, we have made a comparative study with state-of-the-art approaches tackling the ontology matching problem. The approach is evaluated on two datasets from the Ontology Alignment Evaluation Initiative campaign (OAEI). Our experiments demonstrate that our proposed models outperform some machine learning-based approaches and are comparable in performance to other existing systems that use different techniques.