multi-label-classification.pdf下载

Multi-label text classification is a complex problem. Labels may have relationship or hierarchical structure.We investigate several neural networks for multi-label classification. Our baseline is fastText 1, a very simple model with n-gram features. We also explore two modelswith complex structure: one is TextCNN 2withmultiple filters based on CNN.Another is Hierarchical Attention Network 3with hierarchical structure and attention mechanism.These models are used for multi-label classification on Zhihu question & topic dataset. we show that deep models are moresuperior than shallow models