Abstract

Knowledge graph-based semantic similarity measures have been used in several applications. Although knowledge graphs typically describe entities according to different semantic aspects modeled in ontologies, state-of-the-art semantic similarity measures are general-purpose since they consider the whole graph or depend on expert knowledge for fine-tuning.

We present a novel toolkit that can tailor aspect-oriented semantic similarity measures to fit a particular view on similarity. It starts by identifying the semantic aspects, then computes similarities for each semantic aspect, and finally uses a supervised machine learning method to learn a supervised semantic similarity according to the similarity proxy. The toolkit combines six taxonomic semantic similarity and four embedding similarity measures and provides baseline evaluation approaches.

This extended abstract is related to the paper “Towards Supervised Biomedical Semantic Similarity” accepted to the SeWeBMeDA 2022 but focuses on our work’s technical contribution whereas the workshop submission focuses on the use case for biomedical informatics.