Analisis Emosi Multi-Label pada Teks Bahasa Indonesia dengan Implementasi Fine-Tuning BERT

Abstract

In many cases, a single text often contains more than one emotion simultaneously, such as in movie reviews that may express both “joy” and “sadness” at once. To capture this emotional complexity, a multi-label classification approach is considered more representative, as it allows a text to be assigned to multiple emotion categories. This research aims to apply the IndoBERT model for multi-label emotion analysis on Indonesian-language text using a fine-tuning approach. The dataset was collected through web scraping on Twitter using tweet-harvest and annotated into eight basic emotion categories based on Plutchik’s theory, namely anger, anticipation, disgust, fear, joy, sadness, surprise, and trust. The outcome of retraining the fine-tuned IndoBERT model successfully achieved 93% in precision, recall, and F1-score on micro average, and 95% on samples average. The findings suggest that the model performs exceptionally well in identifying multiple emotions in a single text, making it a suitable option for natural language processing tasks that require a thorough understanding of emotions in Indonesian.