A collection of nine multi-label text classification datasets

This is a compressed package containing nine multi-label text classification data sets, including AAPD, CitySearch, Heritage, Laptop, Ohsumed, RCV1, Restaurant, Reuters, and Sentihood.

The datasets of CitySearch, Heritage, Laptop, Restaurant and Sentihood are from the paper of “Bert-flow-vae: A weakly- supervised model for multi-label text classification” (url: https://aclanthology.org/2022.coling-1.104/). The original datasets of Reuters and Ohsumed are from http://disi.unitn.it/moschitti/corpora.htm. The original dataset of AAPD is from https://github.com/lancopku/SGM. The original dataset of RCV1 is from http://www.ai.mit.edu/projects/jmlr/papers/volume5/lewis04a/lyrl2004_rcv1v2_README.htm. For all of these datasets, we adopt the raw text format. For Reuters, we retain the 10 largest classes. In terms of Reuters and Ohsumed, their category words are directly obtained from the descriptive words and seed words defined in [1] and [2]. In

terms of the other datasets, we generate their category words with the protocol described in our proposed category word selection method CWS-SRC.

[1] X. Chen, Y. Xia, P. Jin, and J. Carroll, “Dataless text classification with descriptive lda,” in AAAI, 2015, pp. 2224–2231.

[2] D. Zha and C. Li, “Multi-label dataless text classification with topic modeling,” KAIS, vol. 61, no. 1, pp. 137–160, 2019