Simple tricks for improving pattern-based information extraction from the biomedical literature

Table 6 Results for the extraction of event types other than gene expression (phosphorylation, transcription, and protein catabolism) on development and test data sets.

	Precision	Recall	F1
Protein catabolism - dev data
Baseline	3.6	42.8	6.6
Trigger Word Filter	3.2	38. 1	5.9
Pattern Performance Filter	61.5	38.1	47.0
Protein catabolism - test data
Baseline	1.8	50.0	3.4
Pattern Performance Filter	50.0	14.2	22.2
Best in shared task	66.6	42.8	52.1
Transcription - dev data
Baseline	3.8	26.8	6.7
Trigger Word Filter	4.4	21.9	7.4
Pattern Performance Filter	18.1	24.3	20.8
Trigger Word + Pattern Perf	35.3	20.2	25.5
Transcription - test data
Baseline	4.1	29.2	7.2
Trigger Word + Pattern Perf	23.8	11.2	15.3
Best in shared task	69.2	39.4	50.2
Phosphorylation - dev data
Baseline	2.4	35.6	4.6
Pattern Performance Filter	40.0	34.0	36.7
Phosphorylation - test data
Baseline	3.0	49.2	5.7
Pattern Performance Filter	72.7	47.4	57.4
Best in shared task	91.2	76.3	83.0

The table also reports on winner's performance in the shared task on the test data [21]. Best results for phosphorylation are omitted as these are not directly comparable (see text for expla ation).

ISSN: 2041-1480