Comparação extrínseca de algoritmos de word embedding na simplificação léxica de texto

Abstract

The advent of Artificial Intelligence has provided the advance and the creation of solutions applied to the most diverse areas. Within Natural Language Processing this has not been different, in the last five years, the studies of algorithms for vector representation and semantic retrieval of words have shown great advances. Also called word embeddings, these algorithms add benefits that earlier methods did not provide. Aiming at the need to further study these new algorithms, such as Skip-Gram, Glove and CBOW, and at the same time, noting the importance of the automation of lexical simplification for the benefit of Portuguese learners, dyslexics, aphasia, among others, this work proposes the development of a lexical simplifier using these representations. This simplifier also used a Artificial Neural Network and some dictionaries to create simplifications. There were three main contributions observed in the experiments carried out: a simplifier capable of assisting a proficient speaker in the lexical simplification process, an artificial neural network structure with a tendency to automated learning and the extrinsic comparison of the algorithms. The algorithm Wang2vec Continuous Bag-of-Words performed the best results for the lexical simplification activity during this work’s experiments.