<![CDATA[The spread of hoaxes has become an increasingly alarming global issue, in line with the growing use of social media. This study aims to develop a mobile application system for automatic hoax detection using a Natural Language Processing (NLP) approach by leveraging the XLM-RoBERTa model. The system development process follows the CRISP-DM methodology and the Agile approach. The dataset used consists of 1,928 verified news articles, divided into training and testing data. The model architecture is designed by adding three hierarchical classification layers, consisting of linear layers, batch normalization, ReLU activation, and a dropout rate of 0.4, with the [CLS] token as the main representation for classification. Optimization is performed using the Adam algorithm with a learning rate of 0.001. To enhance the interpretability of the classification results, the system is equipped with an explainability feature based on TF-IDF and Cosine Similarity. The model is integrated into a Flutter-based mobile application with a REST API backend developed using the Flask framework, enabling real-time news analysis. The application provides outputs in the form of prediction labels (hoax/fact), confidence scores, and a summary of the news content. Evaluation results show that the model achieves an accuracy of 94.51%, indicating excellent performance in identifying hoax news automatically and efficiently. By integrating this intelligent model into a mobile application, the system is expected to help users quickly and reliably assess the truthfulness of information, thereby contributing to efforts to mitigate the spread of hoaxes in the digital era.]]>
