castorini/hedwig

Implements multiple neural architectures including DocBERT, Hierarchical Attention Networks, and character-level CNNs, with support for extreme multi-label classification tasks. Models leverage pre-trained word2vec embeddings and NLTK preprocessing, designed to work with benchmark datasets (Reuters, AAPD, IMDB) across document and sentence-level classification. Built on PyTorch 0.4 with modular architecture enabling direct comparison of different deep learning approaches for text classification.

596 stars. No commits in the last 6 months.