BHGAN: A Hierarchical Graph Network Based on Feature Enhancement

Recently, with the development of deep learning techniques, text classification methods have achieved remarkable results. However, the existing work usually models the entire text, less consideration is given to the hierarchical relationship between the texts, and the internal relationship between the preceding and following sentences is ignored. To address this issue, in this paper, we propose a hierarchical graph network model BHGAN based on BERT and utilizing graph attention network to model the hierarchical relationship of texts. When modeling, the semantic features are enhanced by the output of the intermediate layer of BERT, and the relationship between the whole sentence and the sub-sentence is used to construct a hierarchical graph network corresponding to each layer of BERT. The multi-level hierarchical relationship diagram we established not only considers the semantics of each layer, but also the hierarchical relationship of the text. The experimental results show that the BHGAN model exhibits competitive results on the dataset.