tESA: a distributional measure for calculating semantic relatedness

Table 1 Presentation of the general characteristics of the corpora used in the experiments

	MEDLINE	PMC OA	Wikipedia
Size	14073912	1024890	3807314
Type	Scientific	Scientific	Encyclopedic
Documents	Abstacts and titles	Mostly fulltext + abstracts + titles	Fulltext + titles
Snapshot date	Autumn 2015	September 2015	December 2015
Token count [M]	2531,14; 264,84	3684,89; 15,8	2434,55; 11,13
Unique token count [M]	3,85; 1,24	35,57; 0,48	12,53; 0,98

Token counts and unique token counts are expressed in millions. These statistics are collected for raw texts (before preprocessing) and raw corpora (e.g. there might be an uneven number of titles and abstracts in Medline). For each corpus and count type we provide two metrics - of the documents’ textual contents (abstract or full articles) and titles. The statistics are included to highlight the compositional differences between the corpora

ISSN: 2041-1480