REUTERS

Sinopsis

REUTERS newswire corpuses

Description

There are many versions of the REUTERS corpus in use. See Yiming's paper on text categorization for more details.
On AFS we have some different versions:
Reuters_21450:

/XEROX parc filtered

and apte

Reuters-21578:

The original corpus contains the file in SGML. You can use tdt-reuters to extract the smart files. Then you can use eugeneng_clean_cat.c to extract the 90categories sets, described bellow.
A subset of ApteMod that contains only documents whose categories belong to one of the 90 categories for which there exists at least one training document and one testing document, was created by e Sing Eugene Ng . A version of it (they dont have the titles) was called ApteMod

Well documented by David D. Lewis (AT&T Labs - Research, lewis@research.att.com)