Application Of Transformer Model For Sentiment Detection On Indonesian Twitter Data

Abstract

Social media has become an important platform for people to voice their opinions, aspirations and feelings on various social, economic and political issues. Twitter, as one of the most popular social media platforms, presents a wealth of data for research, especially in the field of sentiment analysis. This research explores the application of the Transformer model, specifically IndoBERT, in detecting sentiment from Indonesian tweets. The dataset used was collected from the Twitter API, processed, and manually labelled into three categories: positive, negative, and neutral. Model evaluation was conducted by comparing IndoBERT's performance with traditional classification methods such as Naïve Bayes and Support Vector Machine (SVM). The results show that IndoBERT significantly outperforms conventional models in terms of accuracy, recall, precision, and F1-score, signalling that the Transformer model is highly effective for sentiment analysis in Indonesian.