针对传统多标签文本分类方法忽略标签相关性及文本特征的深层语义信息,导致分类精度和泛化能力不足的问题,提出一种基于融合注意力机制的多标签文本分类方法。首先,构建深度学习模型,通过算法设计与融合注意力机制加强了模型对文本关键信息的提取能力。在模型优化方面,结合标签重要性加权和多任务学习,通过损失函数设计进一步提升了模型性能。为验证所提方法的有效性,在Reuters-21578和RCV1-v2两个数据集上进行实验,将所提模型与现有模型进行比较。实验结果表明,基于融合注意力机制的多标签文本分类方法相较于传统多标签文本分类方法在准确率、召回率、F1值等指标上有显著提升,能够更好地处理文本中的复杂语义和标签之间的依赖关系,展示了更高的分类精度和泛化能力,为多标签文本分类研究提供了新的视角和解决方案。A multi label text classification method based on fusion attention mechanism is proposed to address the problem of insufficient classification accuracy and generalization ability caused by traditional multi label text classification methods ignoring the deep semantic information of label relevance and text features. Firstly, a deep learning model was constructed to enhance its ability to extract key information from text through algorithm design and the integration of attention mechanisms. In terms of model optimization, the combination of label importance weighting and multi task learning has further improved the model performance through loss function design. To verify the effectiveness of the proposed method, experiments were conducted on two datasets, Reuters-21578 and RCV1-v2, to compare the proposed model with existing models. The experimental results show that the multi label text classification method based on fusion attention mechanism has significant improvements in accuracy, recall, F1 value and other indicators compared to traditional multi label text classification methods. It can better handle the complex semantics and dependency relationships between labels in the text, demonstrate higher classification accuracy and generalization ability, and provide new perspectives and solutions for multi label text classification research.
暂无评论