Japanese Short Text Classification Based on CNN-BiLSTM-Attention

• Published in: Procedia computer science Vol. 262; pp. 320 - 329
• Main Authors: Chen, Tianyang, Xie, Zexian
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 2025
• Subjects: CNN-BiLSTM-Attention; Global Context Modeling; Japanese Short Text Classification; Local Feature Extraction; Self-Attention Mechanism; CNN-BiLSTM-Attention; Self-Attention Mechanism; Japanese Short Text Classification; Local Feature Extraction; Global Context Modeling
• DOI: 10.1016/j.procs.2025.05.059
• ISSN: 1877-0509

Due to the limited context information of the text, the traditional statistical feature-based method is difficult to effectively model the semantic relationship in the Japanese short text classification task, resulting in limited classification effect. To this end, this paper introduces the CNN-BiLSTM-Attention fusion model, which aims to fully extract the local and global features in the short text and improve the classification accuracy. First, Convolutional Neural Networks (CNNs) are used to extract local n-gram features and identify phrase patterns. Then, the global context information of the text is modeled by Bidirectional Long Short-Term Memory (BiLSTM) to capture the influence of special structures such as auxiliary words and honorifics. Finally, the self-attention mechanism (Self-Attention) assigns weights to different words, so that the model focuses on the key information of classification and reduces the interference of grammatical vocabulary. In addition, Dropout regularization and Softmax classification layer are introduced to enhance the model’s robustness and capacity for adaptation. Experimental results show that the CNN-BiLSTM-Attention model achieves the best performance in all structures, and the overall WOSS (Word Order Sensitivity Score) score is higher than other models. In the SVO (Subject-Verb-Object) structure, the model reaches 0.94, which is 20.5% higher than CNN’s 0.78, indicating that it has a more accurate understanding of standard word order sentences.