Reinforcement Learning for Extreme Multi-label Text Classification

Extreme multi-label text classification (XMC) is an important yet challenging problem in the NLP community, which refers to the problem of assigning to each document its most relevant subset of class labels from an extremely large label collection. For example, the input text could be a story document on chinastory.cn and the labels could be story categories that implies the potential meaning. However, naively applying normal neural network models to the XMC problem leads to sub-optimal performance due to the large output space and the label sparsity issue. In this paper, we presents the first attempt at applying reinforcement learning to XMC. Experimental results on public and our own engineering datasets demonstrate that our approach achieves expecting performance compared with the evaluation of the state-of-the-art methods.