<![CDATA[Hate speech can have significant social impacts, necessitating automatic classification using deep learning. One of the most widely used models is the Transformer. The Transformer is an effective model for capturing global context in text, but it has limitations in processing time-series data. LSTM is a model capable of processing time-series data using a gate mechanism. This study proposes the LTrans model, a combination of the LSTM and Transformer models. The LSTM model is placed at the beginning to preserve the temporal order of the data, while the Transformer is placed at the end to process the data globally. Thus, it is expected that the data sequence remains intact, the meaning of the sentences does not change, and no information is lost. The research methods include text preprocessing and data augmentation. Text preprocessing is used to clean the data of irrelevant words, normalize it, and reduce noise so that the model can learn more effectively. Data augmentation is performed using Back Translation to translate text into other languages, and BERT Augmentation to enrich data variations without altering the meaning of the sentences. This study aims to classify hate speech using 9 labels with the LTrans model. Evaluation of the LTrans model’s performance yielded an accuracy of 95.89%, a precision of 97.8%, a recall of 95.89%, and an F1 score of 97.8%, indicating balanced performance. Overall, this study demonstrates that LTrans is capable of improving classification quality, accurately detecting hate speech, and effectively handling various targets.]]>
Copyrights © 2026
