A Multi-Channel Neural Network Architecture for Text Classification

This paper presents a Multi-Channel Neural Network architecture integrating Transformer-based encoders, Convolutional Neural Networks (CNNs), and Recurrent Neural Networks (RNNs) with an attention mechanism. Our approach leverages the contextual understanding of Transformers, the pattern recognition of CNNs, and the sequential dependency handling of RNNs. The attention mechanism dynamically weighs features from these channels, enabling the model to focus on relevant text aspects. Our model demonstrates significant accuracy improvements over traditional models and outperforms Google’s Gemma 2B and DoBERTa3 in F1 scores. This research offers a comprehensive text classification solution, with applications in various NLP domains, showcasing the benefits of combining multiple machine learning techniques for superior performance.