Label-representative graph convolutional network for multi-label text classification

Multi-label text classification (MLTC) is the task that assigns each document to the most relevant subset of class labels. Previous works usually ignored the correlation and semantics of labels resulting in information loss. To deal with this problem, we propose a new model that explores label dependencies and semantics by using graph convolutional networks (GCN). Particularly, we introduce an efficient correlation matrix to model label correlation based on occurrence and co-occurrence probabilities. To enrich the semantic information of labels, we design a method to use external information from Wikipedia for label embeddings. Correlated label information learned from GCN is combined with fine-grained document representation generated from another sub-net for classification. Experimental results on three benchmark datasets show that our model outweighs prior state-of-the-art methods. Ablation studies also show several aspects of the proposed model. Our code is available at https://github.com/chiennv2000/LR-GCN.