VeeAlign: Multifaceted Context Representation Using Dual Attention for Ontology Alignment

Abstract

Ontology Alignment is an important research problem applied to various fields such as data integration, data transfer, data preparation, etc. State-of-the-art (SOTA) Ontology Alignment systems typically use naive domain-dependent approaches with handcrafted rules or domainspecific architectures, making them unscalable and inefficient. In this work, we propose VeeAlign, a Deep Learning based model that uses a novel dual-attention mechanism to compute the contextualized representation of a concept which, in turn, is used to discover alignments. By doing this, not only is our approach able to exploit both syntactic and semantic information encoded in ontologies, it is also, by design, flexible and scalable to different domains with minimal effort. We evaluate our model on four different datasets from different domains and languages, and establish its superiority through these results as well as detailed ablation studies. The code and datasets used are available at https://github.com/Remorax/VeeAlign.

Enhancing Data Space Semantic Interoperability through Machine Learning: a Visionary Perspective

“…Machine learning-based methods for ontology alignment [16] and ontology matching [7] can be applied to automatically map concepts and terms from one ontology or vocabulary to another. These algorithms use techniques such as semantic similarity measures [25], graph-based methods [24], and deep learning models [4,10,11] to identify correspondences between concepts in different ontologies or vocabularies. The goal is to produce a mapping that enables data exchange between systems using different ontologies or vocabularies while preserving the meaning of the data.…”

Machine Learning-Friendly Biomedical Datasets for Equivalence and Subsumption Ontology Matching

“…IDFSample (text similarity-based). This strategy is to introduce hard negative candidates that are ambiguous to the ground truth class at text level (i.e., with similar labels 14 ). We first build a sub-word inverted index [12] for the labels of all the classes of O ′ using a sub-word tokenizer pre-trained on biomedical texts [1].…”

“…Recently, machine learning (ML)-based OM systems have become increasingly popular as they can go beyond surface-form string comparison by encoding ontology entities into vectors. For example, DeepAlignment [17] adopts counter-fitting to refine word embeddings for better representation of class labels; VeeAlign [14] proposes a dual encoder to encode both textual and path information of classes; and BERTMap [12] derives mappings through dynamic contextual text embeddings from the pre-trained language model BERT.…”

Machine Learning-Friendly Biomedical Datasets for Equivalence and Subsumption Ontology Matching

“…Recently, machine learning (ML)-based OM systems have become increasingly popular as they can go beyond surface-form string comparison by encoding ontology entities into vectors. For example, DeepAlignment [17] adopts counter-fitting to refine word embeddings for better representation of class labels; VeeAlign [14] proposes a dual encoder to encode both textual and path information of classes; and BERTMap [12] derives mappings through dynamic contextual text embeddings from the pre-trained language model BERT.…”

“…2. LogMap 13 & AML14 . LogMap and AML are two classical OM systems based on lexical matching, mapping extension and repair.…”