基于标签推理和注意力融合的多标签文本分类方法

目前许多多标签文本分类方法主要关注文档表示，而丢失了大量标签相关的语义信息，导致分类效果不理想。针对以上问题，提出一种基于标签推理和注意力融合的分类方法，挖掘文档中与标签相关的特征以及相似标签之间的相关性，学习标签信息进行标签推理，同时采用注意力机制自学习地融合文档表示和标签表示，最终完成多标签分类任务。在AAPD和RCV1-V2数据集上进行实例验证，该方法的F1值分别达到了0.732和0.887，与其他最新方法相比其准确度均有提升，实验结果证明了标签推理和注意力融合策略的有效性。

Multi-label text classification method based on label reasoning and attention fusion

Recently, many multi-label text classification methods mainly focuse on document representation, but lost a lot of label-related semantic information, resulting in unsatisfactory classification effect. In view of this, this paper proposed a multi-label text classification method based on label reasoning and attention fusion. Label reasoning detected text-label related features and similar label-label related features. Attentional mechanism self-learned the fusion of document features and label representation. A lot of cases studied on AAPD and RCV1-V2 datasets show that the F1 values of this method are up to 0.732 and 0.887, respectively, which is more accurate than other latest methods. These experimental results prove the effectiveness of label reasoning and attention fusion.