A novel approach to fake news classification using LSTM-based deep learning models

A novel approach to fake news classification using LSTM-based deep learning models

08 January 2024 | Halyna Padalko, Vasyl Chomko, Dmytro Chumachenko
This study addresses the challenge of fake news detection by developing and evaluating deep learning models, particularly Bi-LSTM and attention-based Bi-LSTM architectures. The rapid spread of misinformation on digital platforms necessitates sophisticated tools for accurate detection and classification. The proposed models integrate an attention mechanism to focus on significant parts of the input data, enhancing their effectiveness. The models were trained on 80% of the WELFake dataset and tested on the remaining 20%, using metrics such as Recall, Precision, F1-Score, Accuracy, and Loss. The attention-based Bi-LSTM model outperformed other models in terms of accuracy (97.66%) and other key metrics. The study highlights the potential of advanced deep learning techniques, especially attention mechanisms, in combating misinformation. However, limitations such as data dependency, overfitting, and language specificity were acknowledged. The research contributes to the ongoing efforts to combat fake news by providing robust and effective models for real-time detection and classification. Future research should focus on enhancing data diversity, model efficiency, and cross-language applicability.This study addresses the challenge of fake news detection by developing and evaluating deep learning models, particularly Bi-LSTM and attention-based Bi-LSTM architectures. The rapid spread of misinformation on digital platforms necessitates sophisticated tools for accurate detection and classification. The proposed models integrate an attention mechanism to focus on significant parts of the input data, enhancing their effectiveness. The models were trained on 80% of the WELFake dataset and tested on the remaining 20%, using metrics such as Recall, Precision, F1-Score, Accuracy, and Loss. The attention-based Bi-LSTM model outperformed other models in terms of accuracy (97.66%) and other key metrics. The study highlights the potential of advanced deep learning techniques, especially attention mechanisms, in combating misinformation. However, limitations such as data dependency, overfitting, and language specificity were acknowledged. The research contributes to the ongoing efforts to combat fake news by providing robust and effective models for real-time detection and classification. Future research should focus on enhancing data diversity, model efficiency, and cross-language applicability.
Reach us at info@study.space