A CNN-BiGRU-GNN Model for Sentiment Analysis of Complex English Texts

Current deep-learning approaches often encounter significant challenges in English literary analysis, including low accuracy, information loss, and semantic ambiguity. Therefore, this study introduces a novel model combining convolutional neural networks (CNNs), bidirectional gated recurrent units (BiGRUs), and graph neural networks (GNNs). The CNN component is utilized for efficient feature extraction from the input text, whereas the BiGRU layer captures bidirectional dependencies, enabling a more comprehensive understanding of the textual context. Additionally, GNN is employed to model the global dependencies within the text, which is crucial for tasks like sentiment analysis and topic recognition. To further enhance feature extraction, Bidirectional Encoder Representations from Transformers (BERT) is integrated into the model, leveraging its deep contextual representations. For sequence labeling tasks, conditional random fields are utilized to improve prediction accuracy by capturing interdependencies between labels. To optimize the GNN hyperparameters, particle swarm optimization algorithm is applied, ensuring the model’s fine-tuning for better performance. Experimental evaluation on the Microsoft Academic Research Corpus dataset demonstrates the efficacy of the proposed model. Compared to baseline models such as BERT-BiLSTM, BERT-CNN-BiGRU, and BERT-CNN-BiLSTM-ATT, the CNN-BiGRU-GNN model yields significant improvements in accuracy, with gains of 4.57%, 4.11%, and 3.01%, respectively. These results highlight the ability of the model to effectively address the complexities of English literary analysis.