Fig. 1

Performance comparison of CLAMP, cTAKES, and MetaMap. Shown here are the performance statistics (precision, recall, and F1 score) of the three tools in extracting ASD terms from (a) 544 PubMed full-text articles and (b) 20,408 PubMed abstracts. Using a rule-based matching approach, a benchmark set of ASD terms was used to label what was considered to be the true entities in the texts. A true entity counts as a true positive if a predicted entity (from CLAMP, cTAKES, or MetaMap) overlaps with the true entity. The precision is the number of true positives divided by the total number of predicted entities (by one of the three tools). The recall is the number of true positives divided by the total number of true entities. The F1 score is calculated as (2 × precision × recall)/(precision + recall). The solid bars represent the results when using the unprocessed predictions from the three tools, and the hatched bars represent the results when first filtering the predicted entities according to the process described in “Methods”