Comparative performance analysis of LSTM, GRU, and bidirectional neural networks for political ideology classification

Abstract

Political ideology classification is crucial for understanding social polarization, monitoring democratic processes, and identifying bias on online platforms. This study compares the performance of long short-term memory (LSTM), gated recurrent unit (GRU), and bidirectional GRU (Bi-GRU) neural network models in classifying liberal and conservative political ideologies from social media text data. The Bi-GRU achieved the best results with 88.75% accuracy and 89.16% F1-score, highlighting its strength in contextual analysis. These findings suggest their applicability in areas such as election monitoring and the analysis of political discourse. This study contributes to the field of political text classification by offering a comparative analysis of deep learning architectures. The dataset utilized covers a wide range of issues, including social, political, economic, religious, and racial topics, demonstrating its comprehensive nature. Visualizations using WordCloud and uniform manifold approximation and projection (UMAP) reveal distinct ideological patterns, validating the dataset’s quality for training models. The findings underscore the importance of utilizing advanced bidirectional architectures for nuanced tasks, such as ideology classification, where contextual understanding is crucial. These insights open avenues for future research, such as the application of Bi-GRU in analyzing multilingual political ideologies or real-time sentiment tracking during election campaigns.