Perbandingan IndoBERT dan IndoRoBERTa Untuk Analisis Sentimen Pada Film Dokumenter Dirty Vote

Abstrak

Sentiment analysis is a technique in Natural Language Processing (NLP) used to identify and categorize opinions or emotions in text. This study compares the performance of two Transformer-based models, IndoBERT and IndoRoBERTa, in analyzing sentiment toward the documentary film Dirty Vote. The research process includes data collection, text preprocessing, lexicon-based sentiment labeling, and model evaluation using K-Fold Cross-Validation. The results show that IndoBERT achieved an average accuracy of 99%, higher than IndoRoBERTa, which achieved 94%. IndoBERT also demonstrated better alignment with lexicon-based labeling in classifying positive, negative, and neutral sentiments. In terms of architecture, IndoBERT employs static masking, while IndoRoBERTa applies dynamic masking, leading to differences in the models' sensitivity to textual meaning. IndoBERT tends to provide more definitive classifications for opinions or strong criticisms, whereas IndoRoBERTa more frequently categorizes ambiguous comments as neutral sentiment. The conclusion of this study indicates that IndoBERT outperforms IndoRoBERTa in sentiment analysis of the documentary film Dirty Vote, both in terms of accuracy and consistency with lexicon-based labeling. These findings provide insights into the effectiveness of Transformer-based models for sentiment analysis in the Indonesian language and can serve as a reference for further NLP model development.