Federated Learning in the Detection of Fake News Using Deep Learning as a Basic Method.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Cybersecurity

Background:

• Federated learning offers a privacy-preserving approach to machine learning by training models on decentralized data.
• Fake news detection is crucial in online spaces, but traditional methods raise privacy concerns due to centralized data aggregation.
• Deep learning models are effective for complex pattern recognition tasks like fake news identification.

Purpose of the Study:

• To explore the efficacy of federated learning combined with deep learning for training fake news detection models.
• To assess the impact of federated learning on model accuracy and data privacy.
• To design and evaluate different deep learning architectures within a federated learning framework.

Main Methods:

• Developed three deep learning architectures: embeddings, Convolutional Neural Networks (CNNs), and Long Short-Term Memory (LSTM) networks.
• Implemented a federated learning approach where models are trained locally on decentralized data and aggregated centrally.
• Compared the performance of models trained with and without federated learning.

Main Results:

• The Long Short-Term Memory (LSTM) architecture achieved the highest performance with an F1 score of 0.92.
• All three deep learning architectures (embeddings, CNNs, LSTM) demonstrated comparable results.
• Federated learning did not significantly reduce model accuracy despite data being decentralized.

Conclusions:

• Federated learning is a viable and secure method for training fake news detection models.
• Deep learning architectures, particularly LSTM, are effective within a federated learning setup for fake news recognition.
• The privacy benefits of federated learning can be realized without substantial compromise on detection accuracy.