Matching ontology using machine learning

Abstract

Ontology matching, a field within data management and information retrieval, utilizes machine learning to integrate and link different data sources by identifying equivalences and relationships between concepts across various ontologies. This report provides an overview of key techniques and methods in ontology matching, including vectorization and embeddings such as Word2Vec, similarity measure like Levenshtein Distance and Jaccard Index, and machine learning models that improve accuracy in classifying and matching concepts. These models use classification and regression tasks to categorize concept pairs based on numerical features. To handle heterogeneous data and minimize overfitting, multiple decision trees are employed, offering robust methods for predicting equivalences between concepts and addressing nonlinear relationships. While ontology matching has diverse applications, from data integration to intelligent search, the field still faces challenges such as semantic differences and data complexity. This report highlights emerging research trends and technologies, as well as potential future developments.