*** Folders ***

* lrd-wat - contains surface forms for DBpedia entities extracted from DBpedia labels, redirects and disambiguations, and from anchor texts of internal Wikipedia links.
Contains not-filtered dataset and two datasets filtered at TF-IDF thresholds equal to 1.8 and 2.6.

* lrd-cc - contains surface forms for DBpedia entities extracted from anchor texts of external Wikipedia links from the Common Crawl corpus
http://blog.commoncrawl.org/2015/01/december-2014-crawl-archive-available/
http://data.dws.informatik.uni-mannheim.de/structureddata/2014-12/wikianchor/
Contains not-filtered dataset and three datasets filtered at TF-IDF thresholds equal to 1.8, 2.6 and 3.8.

* gold - contains two annotated subsets of lrd-wat not-filtered dataset


*** File format ***

* Wikipedia/DBpedia page
* surface form string
* TFIDFscore; empty if surface form does not come from anchor texts
* whether surface form comes from Wikipedia labels (L0), redirects (L1), disambiguations (L2) or intersection of the last two (L3); empty the surface string does not come from any of these sources 
* page-surface from pair count; empty if surface form does not come from anchor texts
* page count; empty if surface form does not come from anchor texts
* surface form count; empty if surface form does not come from anchor texts


*** Gold standard ***

"popular" set - manually selected 34 popular DBpedia entities 
"random" set - randomly selected 81 entity each having at least 5 surface forms

Annotation codes:

* correct ("ok"), with the meaning that a given surface form can indeed be used as an alternative name for a corresponding entity ("the eternal city" for Rome or "red planet" for Mars),
* related, contained ("oi"), when the surface form is part of the entity ("Sao Paulo, Brazil" for Brazil or "Google Japan" for Google), 
* related, contains ("og"), when the surface form contains the entity ("Turkey" for Istanbul),
* related, type of ("g"), when the surface form generalizes the entity ("the city" for Rome	or "book" for The Da Vinci Code),
* related, partial ("p"), when the surface form is an ambiguous partial reference to the entity,
* related ("r"), for the numerous case of the related entities ("Google Blog" for Google, "Martian surface temperatures" for Mars); 
* wrong ("w"), for surface forms that do not refer to the entity ("during World War I" for United States); 
* wrong, formatting ("f"), e.g. surface forms with residual tags.