摘要
To address the difficulty single neural networks encounter in capturing both local features and overall contextual features of text data simultaneously,a dual-channel text classification method utilizing the ERNIE-CNNBiLSTM-MultiAtt model was proposed in this study.It obtained more accurate word representations through ERNIE word embeddings and extracted local features using an improved multi-scale Convolutional Neural Networks(CNN).It also combined Bidirectional Long Short-Term Memory(BiLSTM)model and multi-head attention mechanism to strengthen the word order dependency and feature extraction.The dual-channel fusion strategy enhances the multidimensional representation of features,improving the classification performance.The experimental results showed that the classification effect of this model is better than other comparative models,with better classification accuracy and semantic representation capability.
为了解决单一神经网络难以同时捕捉文本数据的局部特征和整体上下文特征的问题,本研究提出了一种基于ERNIE-CNN-BiLSTM-MultiAtt模型的双通道文本分类方法。该方法通过ERNIE词嵌入获得更准确的词表示,并采用改进的多尺度卷积神经网络(Convolutional Neural Networks,CNN)提取局部特征。同时结合双向长短期记忆网络(Bidirectional Long Short-Term Memory,BiLSTM)和多头注意力机制,以加强词序依赖和特征提取。双通道融合策略增强了特征的多维表示,提高了分类性能。实验结果表明,该模型的分类效果优于其他对比模型,具有更好的分类准确性和语义表示能力。
出处
《印刷与数字媒体技术研究》
北大核心
2025年第6期98-104,121,共8页
Printing and Digital Media Technology Study
