A considerable effort has been made to extract biological and chemical entities, as well as their relationships, from the scientific literature, either manually through traditional literature curation or by using information extraction and text mining technologies. Medicinal chemistry patents contain a wealth of information, for instance to uncover potential biomarkers that might play a role in cancer treatment and prognosis. However, current biomedical annotation databases do not cover such information, partly due to limitations of publicly available biomedical patent mining software. As part of the BioCreative V CHEMDNER patents track, we present the results of the first named entity recognition (NER) assignment carried out to detect mentions of chemical compounds and genes/proteins in running patent text. More specifically, this task aimed to evaluate the performance of automatic name recognition strategies capable of isolating chemical names and gene and gene product mentions from surrounding text within patent titles and abstracts. A total of 22 unique teams submitted results for at least one of the three CHEMDNER subtasks. The first subtask, called the CEMP (chemical entity mention in patents) task, focused on the detection of chemical named entity mentions in patents, requesting teams to return the start and end indices corresponding to all the chemical entities found in a given record. A total of 21 teams submitted 93 runs, for this subtask. The top performing team reached an f-measure of 0.89 with a precision of 0.87 and a recall of 0.91. The CPD (chemical passage detection) task required the classification of patent titles and abstracts whether they do or do not contain chemical compound mentions. Nine teams returned predictions for this task (40 runs). The top run in terms of Matthew’s correlation coefficient (MCC) had a score of 0.88, the highest sensitivity ? Corresponding author
分享
引用
