We describe NatCat, a large-scale resource for text classification constructed from three data sources: Wikipedia, Stack Exchange, and Reddit. NatCat consists of document-category pairs derived from manual curation that occurs naturally within online communities. To demonstrate its usefulness, we build general purpose text classifiers by training on NatCat and evaluate them on a suite of 11 text classification tasks (CatEval), reporting large improvements compared to prior work. We benchmark different modeling choices and resource combinations and show how tasks benefit from particular NatCat data sources.
@article{arxiv.2009.14335,
title = {NatCat: Weakly Supervised Text Classification with Naturally Annotated Resources},
author = {Zewei Chu and Karl Stratos and Kevin Gimpel},
journal= {arXiv preprint arXiv:2009.14335},
year = {2021}
}