Does Head Label Help for Long-Tailed Multi-Label Text Classification

Multi-label text classification (MLTC) aims to annotate documents with the most relevant labels from a number of candidate labels. In real applications, distribution label frequency often exhibits long tail, i.e., few are associated large (a.k.a. head labels), while fraction small tail labels). To address challenge insufficient training data on classification, we propose Head-to-Tail Network (HTTN) transfer meta-knowledge data-rich data-poor The is mapping few-shot network parameters many-shot parameters, which promote generalizability classifiers. Extensive experimental results three benchmark datasets demonstrate that HTTN consistently outperforms state-of-the-art methods. code and hyper-parameter settings released for reproducibility.