Hoax News Classification with Back Translation-EDA and SELU-Transformer

Hoax news deceives and manipulates readers, making them believe fabricated stories. Unfortunately, only a small amount of Indonesian hoax news data is available. To overcome this data limitation, augmentation techniques such as back translation-EDA can be used. Back translation involves translating between Indonesian and English, while EDA employs methods like synonym replacement, random insertion, random swap, and random deletion. Text classification serves as a potent method for Indonesian hoax detection, with transformer architecture standing out in deep learning. This study comprised two key stages: firstly, to overcome the limitations of Indonesian hoax news data, and secondly, classification using the SeLU-transformer architecture, replacing the ReLU mathematical activation function with The SeLU mathematical activation function on the transformer hidden layer. The results obtained in this study are accuracy, precision, recall, and f1-score reaching 96%. The limitations of Indonesian hoax news data, and secondly, classification using the SeLU-transformer architecture, replacing the ReLU mathematical activation function with The SeLU mathematical activation function on the transformer hidden layer. The results obtained in this study are accuracy, precision, recall, and f1-score reaching 96%. The results show the SeLU-transformer architecture has excellent performance in text classification and the detection of Indonesian hoax news.