Table 1.

Deep learning based systems results on the DDI corpus for the DDI classification task (best results in italic)

Systems	Approach	P	R	F1
Sahu and Anand [24]	Combined B-LSTM + AB-LSTM	73.41%	69.66%	71.48%
Liu et al. [17]	Combined CNN + DCNN	78.24%	64.66%	70.81%
Sahu and Anand [24]	B-LSTM	75.97%	65.57%	70.39%
Liu et al. [17]	MCCNN	75.99%	65.25%	70.21%
Liu et al. [17]	DCNN	77.21%	64.35%	70.19%
Liu et al. [15]	CNN with MEDLINE word embedding	75.72%	64.66%	69.75%
Zhao et al. [21]	Two-stage SCNN	72.5%	65.1%	68.6%
Zhao et al. [21]	One-stage SCNN	69.1%	65.1%	67%
Sahu and Anand [24]	AB-LSTM	67.85%	65.98%	66.9%
Suárez-Paniagua et al. [16]	CNN with random word embedding	69.86%	56.1%	62.23%