Advancing fake news detection: a comparative study of RNN, LSTM, and Bidirectional LSTM Architectures

Gregorius Airlangga (Atma Jaya Catholic University of Indonesia)

Jurnal Teknik Informatika C.I.T. Medicom
Vol 16 No 1 (2024): March: Intelligent Decision Support System (IDSS)

In the era of information overload, the exponential growth of digital content has coincided with the proliferation of 'fake news,' posing a critical challenge to online information credibility. This study addresses the pressing need for robust fake news detection systems by conducting a comparative analysis of three neural network architectures: Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM), and Bidirectional LSTM (BiLSTM). Our primary objective is to assess their effectiveness in identifying fake news in a binary classification setting. To achieve this goal, we employed advanced neural network models and a dataset of news titles. Our applied research method included data preprocessing and the utilization of RNN, LSTM, and BiLSTM models, each tailored to handle sequential data and capture temporal dependencies. we rigorously assessed the performance of RNN, LSTM, and BiLSTM models using a range of metrics, including accuracy, precision, recall, and F1-score. To achieve a comprehensive evaluation, we divided our dataset into training and testing subsets. Specifically, we allocated 67% of the data for training purposes and the remaining 33% for testing. Our research findings reveal that all three models consistently achieved high accuracy levels, approximately 91%, with slight variations in precision and recall. Notably, the LSTM model exhibited a marginal improvement in recall, which is crucial when the consequences of missing deceptive content outweigh false alarms. Conversely, the RNN model demonstrated slightly better precision, making it suitable for applications where minimizing false positives is paramount. Surprisingly, the BiLSTM model did not significantly outperform the unidirectional models, suggesting that, for our dataset, processing information bidirectionally may not be essential. In conclusion, our study contributes valuable insights to the field of fake news detection. It underscores the significance of model selection based on specific task requirements and dataset characteristics.