Enhancing BERT Classification by Increasing Training Corpus Using Paraphrasing

Atencio-Torres C.E.; Vera-Sancho J.A.; Calcina-Ccori P.C.; Mamani-Aliaga A.H. · Communications in Computer and Information Science (2025) · DOI: 10.1007/978-3-031-91428-7_16

Resumen

Sentiment Analysis usually classifies a sentence by label it with positive or negative. Previous to apply a classification model, we need a big corpus with many diverse cases. In this paper we run different paraphrasing techniques over the IMDB reviews dataset increasing the number of sentences. We tested such augmented corpus using a BERT classification model. The results: without paraphrasing, the BERT model obtained 90% of f-measure; and with paraphrasing we obtained between 99.56% and 99.86%. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

Instituciones: Universidad Nacional de San Agustín de Arequipa