Hierarchical Text Classification as Sub-hierarchy Sequence Generation

Hierarchical text classification (HTC) is essential for various real applications. However, HTC models are challenging to develop because they often require processing a large volume of documents and labels with hierarchical taxonomy. Recent based on deep learning have attempted incorporate hierarchy information into model structure. Consequently, these implement when the parameters increase large-scale structure depends size. To solve this problem, we formulate as sub-hierarchy sequence generation target label instead Subsequently, propose Hierarchy DECoder (HiDEC), which decodes using recursive decoding, classifying all parents at same level children once. In addition, HiDEC trained use path from root each leaf in composed document via an attention mechanism hierarchy-aware masking. achieved state-of-the-art performance significantly fewer than existing benchmark datasets, such RCV1-v2, NYT, EURLEX57K.