Abstract

Sentiment analysis problems have frequently used machine learning models and lexicon-based methods. In this paper, we explore the use of a state-of-the-art model called BERT — Bidirectional Encoder Representations from Transformers — following a comprehensive review of the sentiment analysis literature.

We compared the performance of BERT in sentiment analysis to NLTK’s VADER, TextBlob’s Sentiment Analyzer, and Logistic Regression. BERT had the highest accuracy and recall score, but also took the longest runtime.

Keywords: Sentiment analysis, VADER, BERT

Full text article and source codes can be provided upon request.