Effectiveness Analysis of RoBERTa and DistilBERT in Emotion Classification Task on Social Media Text Data
Abstract
The development of social media provides various benefits in various ways, especially in the dissemination of information and communication. Through social media, users can express their opinions, or even their feelings. In this regard, sometimes users also convey information or opinions according to the user's feelings or emotions. This triggers the impact of aggressive online behavior, including cyberbullying, which triggers unhealthy debates on social media. The development of deep learning models has also been developed in several ways, especially emotion classification. In addition to using deep learning models, the development of classification tasks has also been carried out using transformer architectures, such as BERT. The development of the BERT model continues to be carried out, so this study will analyze and explore the application of BERT model development, such as RoBERTa and DistilBERT. The optimal result of this study is with an accuracy value of 92.69% using the RoBERTa model.