Table 1. Comparison of proposed approach with EMU approach on benchmark datasets.

The parantheses values correspond to (true positives, false positives) for precision and (true positive, false negatives) for recall.

Corpus	EMU: Without sequence filter	Our baseline approach: Co-occurrence only	Our approach: Without sequence filter	EMU: With sequence filter	Our full approach: With sequence filter
PCA Precision	0.39 (151, 237)	0.37 (154, 263)	0.75 (132, 42)	0.59 (127, 89)	0.82 (144, 32)
Recall	0.80 (151, 37)	0.82 (154, 34)	0.70 (132, 56)	0.66 (127, 61)	0.77 (144, 44)
F-measure	0.52	0.51	0.724	0.62	0.794
BCA Precision	0.34 (242, 470)	0.33 (252, 504)	0.738 (206, 73)	0.61 (193, 121)	0.742 (207, 72)
Recall	0.85 (242, 42)	0.89 (252, 32)	0.725 (206, 78)	0.68 (193, 91)	0.73 (207, 77)
F-measure	0.49	0.49	0.73	0.64	0.74