OAEI Machine Learning Dataset for Online Model Generation

This paper is a technical report that introduces a new dataset designed for machine learning systems participating in the Ontology Alignment Evaluation Initiative (OAEI). The dataset includes training, validation, and test splits for various OAEI tracks, facilitating online model learning, which allows systems to adapt to input alignments without human intervention. The authors discuss the limitations of traditional offline model generation, where models are pre-packaged and may not perform well on new datasets. The dataset aims to enable fair comparisons among machine learning-based matching systems by providing a structured approach to training and validation. The report also details the stratification process used to create the dataset, ensuring a balanced representation of different types of correspondences. Additionally, the usefulness of the dataset is demonstrated through experiments that fine-tune confidence thresholds for popular matching systems, showcasing potential improvements in precision and recall metrics. The authors express hopes for further advancements in matching system development using this dataset.