The CHEMDNER corpus of chemicals and drugs and its annotation principles

Table 1 CHEMDNER corpus overview.

	Training set	Development set	Test set	Entire corpus
Abstracts	3,500	3,500	3,000	10,000
Nr. characters	4,883,753	4,864,558	4,199,068	13,947,379
Nr. tokens	770,855	766,331	662,571	2,199,757
Abstracts with SACEM	2,916	2,907	2,478	8,301
Nr. mentions	29,478	29,526	25,351	84,355
Nr. chemicals	8,520	8,677	7,563	19,805
Nr. journals	193	188	188	203
TRIVIAL	8,832	8,970	7,808	25,610
SYSTEMATIC	6,656	6,816	5,666	19,138
ABBREVIATION	4,538	4,521	4059	13,118
FORMULA	4,448	4,137	3,443	12,028
FAMILY	4,090	4,223	3,622	11,935
IDENTIFIER	672	639	513	1,824
MULTIPLE	202	188	199	589
NO CLASS	40	32	41	113

This table provides an overview of the CHEMDNER corpus in terms of the number of manually revised abstracts (Abstracts) with their total sizes as number of characters and tokens, the number of abstracts containing at least one chemical entity mention (Abstracts with CEM), the number of annotated mentions of chemical entities, the number of unique chemicals annotated (the non-redundant list of mentions) and the number of corresponding journals for the annotated abstracts. The number of mentions for each CHEMDNER entity class (see Figure 1) is provided for each set and the entire corpus in the lower half of the table.

ISSN: 1758-2946