Label Oriented Hierarchical Attention Neural Network for Short Text Classification

Text classification task is a very common in natural language processing. The conventional text model often uses word bag or representation model, but the existing usually deals with long task, which not suitable for short classification. Short features are relatively fewer and need more sophisticated feature extraction, role of keywords play great importance roles text. In this paper, we propose label-oriented hierarchical attention mechanism network achieves better results on public data sets Tiao Weibo, compared convolutional neural CNN, GATE control unit GRU, gate fusion GRU-CNN translation Transfomer. It proved that has good performance. Our two significant advantages: (1) it structure consisting levels mechanisms, facilitates interpretation analysis; (2) Compared other architectures, can be used to extend multi-label tasks. addition, long-term analysis many industrial scenarios.