Exploring Multi-label Classification Using Text Graph Convolutional Networks on the NTCIR-13 MedWeb Dataset

The NTCIR-13 Medical Natural Language Processing for Web Document (MedWeb) task requires participant systems to perform multi-label text classification, where labels representing eight different diseases or symptoms are assigned to each pseudo-tweet. While a recent study showed that a method based on Text Graph Convolutional Networks outperforms previous approaches in text classification, its performance on multi-label text classification remains unclear. Hence, in this study, we construct a model based on Text Graph Convolutional Networks (Text GCN) and evaluate its performance on multi-label classification. Our experimental results show that Text GCN could not outperform the baseline system on the NTCIR-13 MedWeb task.