Sentimental Analysis on Social Media Comments with Recurring Models and Pretrained Word Embeddings in Portuguese

Abstract

Natural Language Processing (NLP) techniques are increasingly powerful for interpreting a person’s feelings and reaction to a product or service. Sentiment analysis has become a fundamental tool for this interpretation, and it has studies in languages other than English. This type of application is uncommon and unheard of in Portuguese. This article presents a sentiment analysis classification based on Portuguese social media comments. Representation of word embeddings with both pre-trained Glove and Word2Vec models were generated through a corpus entirely in Portuguese. This article presents a set of results with different models of pre-trained layers and deep learning models exclusive to the Portuguese language on social networks. Two classification models were used and compared: (i) Bidirectional Long Short-Term Memory (BI-LSTM) and (ii) Bidirectional Gated Recurrent Unit (BI-GRU), achieving accuracy results of 99.1