KLASIFIKASI HOAX VS NON-HOAX PADA BERITA BENCANA ALAM BERBAHASA INDONESIA MENGGUNAKAN WORD EMBEDDING

Hoaxs or fake news related to natural disasters can cause panic and widespread misinformation among the public. Therefore, an automated method is needed to effectively classify hoax and non-hoax news. This study implements Word Embedding methods and the Long Short-Term Memory (LSTM) algorithm to classify hoax news on Indonesian-language natural disaster reports. Three Word Embedding models were used: Word2Vec, FastText, and GloVe. The research process includes data preprocessing, dataset splitting, LSTM model implementation, and performance evaluation using accuracy, precision, recall, and F1-score metrics. The results show that the FastText model with LSTM achieved the highest accuracy of 99%, followed by Word2Vec-LSTM and GloVe-LSTM. FastText effectively captures information from rarely occurring words, improving its ability to detect hoax news. Additionally, the data augmentation technique using the Random Synonym Replacement method has proven to enhance dataset diversity and balance, positively impacting model performance. With these findings, this study is expected to serve as a reference for future researchers in developing more accurate and efficient hoax detection systems, particularly in the context of natural disaster news.