The Webis Open Directory Project Corpus 2010 (Webis-ODP-10) is a corpus for the evaluation of cluster labeling algorithms. The corpus contains about 5,000 web pages which are grouped into 4 main categories and 12 subcategories based on the human made Open Directory Project (ODP) classification. The web pages were collected in May 2010.
To download the corpus use the following link:
(117.4 MB, MD5 sum: 430a16404425f6a8ce1e3f2338f4a937).
If you use the dataset in your research, please send us a copy of your publication. We kindly ask you to refer to the corpus via [bib].
In order to create the corpus, 4 different categories of the Open Directory Project were chosen: movies, databases, health and recreation. Each of these categories is again consists of up to six subcategories. In the case of movies, these subcategories correspond to six different titles of two directors. The categories were chosen to allow for different settings of experiments. Example for tasks are to distinguish the two directors or to separate web pages about recreation from those about health.
The categories are listed in the table below.
|Movies/Kubrick||Arts/Movies/Titles/2/2001 - A Space Odyssey|
|Arts/Movies/Titles/E/Eyes Wide Shut|
|Movies/Hitchcock||Arts/Movies/Titles/Psycho - 1960|
|Databases||Computers/Software/Databases/Data Warehousing/Data Integrity and Cleansing Tools|
|Health||Health/Pharmacy/Drugs and Medications/Antibiotics|
For each category, the linked web pages were downloaded if they still existed. Furthermore, web pages which are linked from these web pages were downloaded as well and considered to be in the same category. However, only web pages of a length of 1,000 or more characters (excluding HTML tags and scripts) were kept.