A Study on Vietnamese Semantic Analysis using BERT-Based PreTrained Language Model

Abstract

One of the most significant NLP tasks is sentiment analysis, in which machine learning models are taught to identify text based on polarity of opinion. Many suggested models have produced cutting-edge results for sentiment analysis in English corpora. However, there have not been many investigations of this technique for Vietnamese corpus, which has resulted in several limitations in Vietnamese study. In this paper, we suggested a sentiment analysis technique for Vietnamese utilizing the PhoBERT pretrained model. PhoBERT is based on RoBERTa, a robust Vietnamese optimization of the well-known BERT model. Our technique produces quite good performance on the given dataset with an AUC score of 86%. This is anticipated to provide the groundwork for future study in Vietnamese, which is a language with limited resources.