Ontology Matching Using Convolutional Neural Networks

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 5648–5653 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 5648 Ontology Matching Using Convolutional Neural Networks Alexandre Bento, Amal Zouaq, Michel Gagnon ´Ecole Polytechnique de Montr´eal {alexandre.bento, amal.zouaq, michel.gagnon}@polymtl.ca Abstract In order to achieve interoperability of information in the context of the Semantic Web, it is necessary to find effective ways to align different ontologies. As the number of ontologies grows for a given domain, and as overlap between ontologies grows proportionally, it is becoming more and more crucial to develop accurate and reliable techniques to perform this task automatically. While traditional approaches to address this challenge are based on string metrics and structure analysis, in this paper we present a methodology to align ontologies automatically using machine learning techniques. Specifically, we use convolutional neural networks to perform string matching between class labels using character embeddings. We also rely on the set of superclasses to perform the best alignment. Our results show that we obtain state-of-the-art performance on ontologies from the Ontology Alignment Evaluation Initiative (OAEI). Our model also maintains good performance when tested on a different domain, which could lead to potential cross-domain applications. Keywords: ontology matching, machine learning, convolutional neural networks 1. Introduction An ontology is a set of structural rules designed to repre- sent concepts in order to perform logic-based operations to retrieve or infer new information. As Semantic Web tech- nologies are expanding and becoming more popular, the amount of data that needs to be represented grows propor- tionally, as does its complexity. Since web technologies are designed to be decentralised, redundancy inevitably occurs among knowledge bases. For this reason, finding methods to align equivalent data or their model is crucial. This is the goal of our research, which is described in this paper. Traditional ontology matching methods are mainly based on string similarity and structure analysis. In this paper we propose another approach using machine learning tech- niques. The core of our method is based on the idea that semantic information can be retrieved automatically from text information within ontologies, and within their struc- ture. To achieve this task, convolutional neural networks were used. 2. Previous Works As indicated earlier, the usual way to align ontologies gen- erally consists in standard string similarity measurement techniques, and the analysis of the ontologies’ structure to identify equivalent classes and properties. There are three main string analysis techniques: 1) string distance metrics (Levenshtein, Jaccard, string equality, etc.), 2) syntactic transformations (tokenization, lemmati- zation, stop word removal, etc.) and 3) semantic opera- tions (finding synonyms, translating, categorization, etc.) (Cheatham and Hitzler, 2013). Although using only string metrics can lead to significant results for the ontology matching problem, it is crucial to choose the right metric to obtain the best alignment. Some advanced string metrics can also be defined and show powerful results, such as met- rics that take both the similarities and the differences be- tween two labels into account (Stoilos …

Content truncated to 500 words