基于混合粒度全局图的多标签文本分类方法OA

Multi-label text classification is designed to assign multiple labels to each instance of text.Traditional multi-label text classification methods usually rely on coarse-grained feature representations,ignoring the multi-level and multi-scale semantic in-formation in the text.In order to solve this problem,this paper proposes a multi-label text classification method based on mixed granularity global graph,which extracts fine-grained text features through MHA to capture the interaction information between words and labels,and uses Bi-LSTM to extract coarse-grained text features.Subsequently,the two features are fused through the gated fusion mechanism to obtain mixed granular features with multi-level semantics.The fused mixed granular word representa-tions,texts,and labels are used together to construct a global graph,and the global graph is processed through a graph convolu-tional network for classification.Experiments are carried out on two datasets,AAPD and RCV1-V2,and the experimental results show that the proposed method can effectively improve the performance of the model.