Implementation of Text Mining to Detect Emotions of Fuel Price Increase using BERT-LSTM Methods

Abstract

Fuel is crucial for everyday life, especially as a primary source of transportation fueled by oil. In early April 2022, Indonesia experienced a significant event that deeply affected its populace: a surge in fuel prices. Addressing this pressing issue, this study employs emotion classification utilizing BERT and LSTM methods on social media data, particularly from platforms like YouTube, to categorize emotional responses to governmental decisions. This research aims to classify social media discourse surrounding fuel-related topics, notably the increases in fuel prices. The highest accuracy, at 95%, was achieved with oversampling techniques, contrasting with a mere 47% accuracy without oversampling. Surprisingly, experiments indicate that employing oversampling and BERT for emotion classification results in reduced accuracy during testing phases.