Abstract
The rapid expansion of social networks has made a vast amount of user-generated data readily available for public analysis. Such data can be leveraged for various purposes, including textual analysis of comments and reviews.
This study utilizes a variant of the Bidirectional Encoder Representations from Transformers (BERT) model, specifically designed for Bahasa Indonesia (IndoBERT), to improve performance on the Indonesian natural language understanding benchmark, focusing on sentiment analysis and emotion classification tasks.
A hybrid approach was applied, combining the summation of the IndoBERT model’s last hidden layers with a neural network. The model’s performance was evaluated using the F1-score metric, with experimental results indicating an accuracy of 0.92 for sentiment analysis and 0.76 for emotion classification.
