Browsing by Author "Mochten Ayat Elrrahmane"

Filter results by typing the first few letters
Now showing 1 - 1 of 1
  • No Thumbnail Available
    Item
    AI Powered Fake News Classification through Advanced NLP Techniques
    (university of ghardaia, 2026) Bouharkat Lina Djihane; Mochten Ayat Elrrahmane
    Fake news has emerged as a major challenge in the digital era due to the rapid dissemination of misinformation across online platforms and social media, with signifi- cant implications for public opinion, decision-making, and societal trust in information sources. This study investigates the effectiveness of transformer-based deep learning models for fake news detection under the hypothesis that pre-trained language models can capture rich contextual and semantic representations of textual data, enabling accurate binary classification of news articles as real or fake. The main objective is to design and evaluate a robust and comparative framework based on state-of-the-art transformer architectures, including BERT, RoBERTa, DeBERTa-v3, and a hybrid en- semble model, and to assess their performance across two benchmark datasets, ISOT Fake News and WELFake. The adopted methodology consists of fine-tuning pre- trained models on labeled datasets following a unified preprocessing pipeline involving data cleaning, tokenization, sequence standardization, and label encoding, followed by evaluation using standard metrics such as accuracy, F1-score, ROC-AUC, and con- fusion matrix. The experimental results demonstrate that transformer-based models achieve strong and competitive performance in fake news detection, with RoBERTa achieving the best results on the WELFake dataset and BERT showing the most consis- tent performance on the ISOT dataset, achieving the best result with 95.67% F1-score, while DeBERTa-v3 and the hybrid ensemble model exhibit comparatively lower perfor- mance, highlighting the sensitivity of model effectiveness to dataset characteristics and training configurations. Overall, the findings confirm the effectiveness of transformer- based approaches for misinformation detection and underline the importance of model selection depending on dataset properties, with potential applications in automated fact-checking and social media monitoring systems.