Table 3.

Performance (F1 score) of the classifiers for identifying the discourse connectives

	BioDRB ∩ PDTB			BioDRB ∉ PDTB			BioDRB ∅ PDTB
	% Of conns	Performance as DCONN	Performance as non-DCONN	% Of conns	Performance as DCONN	Performance as Non DCONN	% Of conns	Performance as DCONN	Performance as non-DCONN
Cross-domain	96.7%	0.62	0.92	3.3%	0.03	0.97	0%	0	0.86
UnWeighted	84.3%	0.70	0.93	10.5%	0.21	0.98	5.2%	0.55	0.91
In-domain	74%	0.78	0.94	19.8%	0.65	0.98	6.2%	0.63	0.91
Weighted	75.7%	0.75	0.94	17%	0.51	0.98	7.3%	0.7	0.92
Pruning	93.4%	0.67	0.93	3.3%	0.08	0.97	3.3%	0.14	0.87
FeatAugment	72.8%	0.70	0.93	21%	0.58	0.98	6.2%	0.5	0.9
Weighted-Pruning	75.3%	0.77	0.94	18.5%	0.60	0.98	6.2%	0.63	0.91
Weighted-FeatAugment	72.6%	0.77	0.94	20.2%	0.67	0.98	7.2%	0.7	0.92
Hybrid	74.4%	0.78	0.94	19.5%	0.66	0.98	6.1%	0.67	0.92
Weighted-Hybrid	73.8%	0.78	0.94	20.2%	0.65	0.98	6%	0.67	0.92

BioDRB, Biomedical Discourse Relation Bank; PDTB, Penn Discourse Treebank.