Domain-specific Word2vec-based trained models - Chem300, Phys300, MatSci200, MatSci300, and Mixed300

Abstract

The repository contains materials science, chemistry, and physics-specialized unsupervised trained models. Word embeddings are generated by means of the Word2vec, a natural language processing technique comprised of language model architectures for fast and efficient learning of distributed representations of words. Continuous Skip-gram model architecture with a negative sampling strategy, as implemented in the Gensim library, is employed for model training. The word embeddings consisting of 200 and 300 vectorial components for materials science and 300 vectorial components for chemistry, physics, and mixed domain are here provided.