Etudiant : HAIKI HANAE

Filière : Master Systèmes Intelligents et Décisionnels (MSID)

Encadrant : Pr. NFAOUI EL HABIB

Annèe : 2021

Résumé : The present document constitutes the synthesis of the work, realized within the project of end of studies for the objective of obtaining the diploma of Master system Intelligence and decisional in the Faculty of Sciences Dhar El Mehraz of Fez. The proliferation of social media has provided individuals to spread information without cost, with little investigation and fewer filters than before. Therefore, fake news has become a major concern nowadays due to the negative impact it brings on the communities. For instance, the pandemic caused by COVID-19 led to the distribution of excessive pseudoscientific information and fake news that has confused the general population. In addition, during the COVID-19 vaccination, fake news has been spread to show the ineffectiveness of the AstraZeneca COVID-19 vaccine, which imposed Europe stopping it for a timestep. Thus, the identification of fake news has become the main research field in natural language processing (NLP). The key challenge is to determine whether the news is real or fake. Besides, the recent achievements of deep learning techniques in complex natural language processing tasks, make them a promising solution for fake news detection too. This work proposes several experiments based on deep learning techniques in order to introduce an automatic fake news detection system. For this purpose, we have evaluated the performance of using two deep learning models namely Long short-term memory (LSTM) and Convolutional neural network (CNN) by means of using Word2vec as word representation. Furthermore, we have fine-tuned the Bidirectional Encoder Representations from Transformers (BERT) model in order to detect fake news. The model was successfully validated on a public fake news dataset collected from the Kaggle website1 . The experimental results demonstrate that LSTM deep based model outperforms CNN deep based model and the fine-tuned BERT for the fake news specif task, achieving an accuracy of 97% and an F1-score of 95%. Keywords: fake news, deep learning, LSTM, CNN, BERT, word2vec.