A Gated Graph Neural Network With Attention for Text Classification Based on Coupled P Systems

• Published in: IEEE access Vol. 11; pp. 72448 - 72461
• Main Authors: Zhang, Jiaqi, Liu, Xiyu
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 2023; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Artificial neural networks; Attention mechanism; Biomembranes; Classification; Computational modeling; Feature extraction; graph neural network; Graph neural networks; Graph theory; Natural language processing; Nodes; P systems; Representations; Semantics; Task analysis; Text categorization; text classification
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2023.3295572

Text classification is a classical task in natural language processing. Prior traditional text classification methods rely on manually extracted features to a great extent, which are easily influenced by human subjectivity. Some existing text classification methods based on the artificial neural networks sometimes neglect the contextual semantic relationships of discontinuous word sequences, resulting in poor learning results. To alleviate these problems, we propose an attention-based gated graph neural network in the framework of coupled P systems (CPGANN) to automatically extract feature representations of nodes. The gating unit with attention is introduced to aggregate neighbor information to capture context semantic relations, and effectively alleviate the long-term dependence on discontinuous words. In order to obtain more discriminative nodes for classification, the attention mechanism is employed in CPGANN to extract keyword nodes before readout to aggregate subgraph representations. Extensive experiments on four real-world datasets demonstrate that CPGANN outperforms all other state-of-the-art baseline algorithms.