View source for Reuters 21578

== Citation == 
{{MyCiteconference | coauthors =  et al | date = 1987 | first = D.| last = Lewis | title = Reuters-21578 | url = http://www.daviddlewis.com/resources/testcollections/reuters21578/}}

The Reuters 21578 [[Category::Dataset|dataset]] is used for text categorization classification, and consist of documents that appeared on the Reuters Newswire in 1987. 

The dataset consists of 22 files: The first 21 files contain 1000 documents each, and the 22nd contains 578 documents. The formatting of the data is in SGML format.  

The categories in this dataset come from five classes:

* Exchanges: financial exchanges, e.g., "nasdaq"
* Organizations: named entities of organizations, e.g., "GE"
* People: named entities of people, e.g. "Paul Volcker"
* Places: named entities of places, e.g., "Australia"
* Topics: economic subject categories, e.g., "coconut", "gold", "money supply"

Some more information on the distribution:

                     Number of    Number of Categories   Number of Categories 
Category Set  Categories     w/ 1+ Occurrences      w/ 20+ Occurrences  
************  **********   ********************   ******************** 
EXCHANGES   39                32                       7
ORGS             56                32                       9
PEOPLE          267               114                      15
PLACES          175               147                      60
TOPICS          135               120                      57

It is recommended to use the pre-split training-test splits, i.e., either "ModLewis" split or "ModApte" split.