Disinformation Detection on 2024 Indonesia Presidential Election using IndoBERT

Social media is not only used for social communication, but also for the comprehensive and effective dissemination of news and information. Twitter is one of the largest social media also used to spread news and information. Information published on Twitter may not always be verifiable. This can lead to disinformation being spread on Twitter. The spread of disinformation on social media has become a growing problem, especially around the time of the presidential election. The purpose of this study is to use the IndoBERT model to identify and minimize the spread of disinformation on Twitter related to the 2024 Indonesian presidential election. This study was conducted in several phases including dataset collection, preprocessing, data labeling, word embedding with Word2Vec, classification with IndoBERT, validation and evaluation with K-Fold Cross Validation. The results show that using IndoBERT in combination with NLTK Tokenizer and BERT AutoTokenizer yields promising results in minimizing the spread of disinformation on social media. Accuracy results achieved were 85% when using IndoBERT with BERT AutoTokenizer and 87% when using IndoBERT with NLTK Tokenizer and BERT AutoTokenizer. Overall, this study demonstrates the effectiveness of using advanced NLP models like IndoBERT in detecting and minimizing the spread of disinformation on social media.