HYBRID DEEP LEARNING FOR TEXT CLASSIFICATION: INTEGRATING BIDIRECTIONAL GATED RECURRENT UNITS WITH CONVOLUTIONAL NEURAL NETWORKS

Abstract

Text classification remains a foundational task in natural language processing with wide-ranging applications, including sentiment analysis, topic categorization, spam detection, and information retrieval. While convolutional neural networks (CNNs) are adept at capturing local n-gram features, and recurrent neural networks (RNNs) excel at modeling sequential dependencies, standalone architectures often struggle to fully leverage both aspects simultaneously. This study presents a hybrid deep learning model that integrates bidirectional gated recurrent units (Bi-GRU) with convolutional neural networks to enhance text classification performance. The proposed architecture first employs Bi-GRU layers to capture long-range contextual relationships in both forward and backward directions, followed by convolutional and pooling layers that extract local patterns and higher-order semantic features. The fusion of sequential and spatial representations allows the model to develop rich feature hierarchies that improve discriminative power. Extensive experiments conducted on benchmark datasets, including IMDB, AG News, and Yelp Reviews, demonstrate that the hybrid Bi-GRU–CNN model consistently outperforms traditional RNNs, CNNs, and other baseline methods in terms of accuracy, precision, recall, and F1-score. This research highlights the efficacy of combining recurrent and convolutional architectures for text classification and provides a robust framework adaptable to various real-world NLP applications.