Multi-label classification of legal text based on label embedding and capsule network

With the development of deep learning technology and the disclosure of legal texts, the classification of legal texts has attracted the attention of researchers. At present, research on the classification of legal texts is mainly focused on multiclass classification. There are few studies on multi-label classification for legal texts. This paper addresses the use of a label sequence generation model to study the multi-label classification of legal texts at the sentence level. The current general multi-label classification methods are often designed for long texts and ignore the transfer relationships between labels. We propose a method based on label embedding and a capsule neural network for the multi-label classification of legal text. Our proposed method applies the graph convolutional network to learn label embeddings and the correlations between labels, a fusion layer to combine the label information with the contextual semantic information of texts and a capsule neural network to extract the spatial feature information of text. Experimental results on three legal text datasets show that our proposed model outperforms the baseline methods, verifying the effectiveness of our proposed model for legal text with an uncertain number of characters in words and short lengths. In addition, we experimented on two datasets that are usually applied in multi-label classification, and the performance of the model shows that the method we proposed is competitive with state-of-the-art models of multi-label text classification.