BBGC: BERT-BiLSTM-GCN Fusion with Collaborative Attention Mechanism for Text Classification Model

Text categorization serves as a fundamental objective within the domain of Natural Language Processing (NLP). Text data often contains rich structural and contextual information, such as correlations between words, logical relationships between sentences, and diverse contexts, which makes it especially critical to effectively capture and process full-text information. In this paper, a model named BBGC is proposed to improve the capture and processing of full-text information. The model extracts contextual and structural information of text by combining BERT-BiLSTM and Graph Convolutional Network (GCN), respectively, and employs a co-attention mechanism and a gate mechanism to fuse features from different perspectives of the GCN and BERT-BiLSTM, which enhances the attention to key features and integrates features from different sources more effectively. In addition, we conducted experimental validation on four datasets, namely R52, R8, Ohsumed, and MR, with accuracies of 96.65%, 98.72%, 73.83%, and 90.41%, respectively. The experimental results demonstrate the effectiveness of the proposed method in improving the accuracy of text categorization and outperform the existing methods.