BERT: a sentiment analysis odyssey

摘要

The study investigates relative effectiveness of four sentiment analysis techniques: (1) unsupervised lexicon-based model using SentiWordNet, (2) traditional supervised machine learning model using logistic regression, (3) supervised deep learning model using Long Short-Term Memory (LSTM), and (4) advanced supervised deep learning model using Bidirectional Encoder Representations from Transformers (BERT). Publicly available labeled corpora of 50,000 movie reviews originally posted on Internet movie database (IMDB) were analyzed. Sentiment classification performance was calibrated on accuracy, precision, recall, and F1 score. The study puts forth two key insights: (1) relative efficacy of four sentiment analysis algorithms and (2) undisputed superiority of pre-trained advanced supervised deep learning algorithm BERT in sentiment classification from text. The study is of value to analytics professionals and academicians working on text analysis as it offers critical insight regarding sentiment classification performance of key algorithms, including the recently developed BERT.