A Shortest Dependency Path Based Convolutional Neural Network for Protein-Protein Relation Extraction

Table 5

The results of the four proposed models on PPI task. Combined model: using both randomly initialized and pretrained word embedding as inputs and concatenating the outputs of max-pooling layer as features for MLP. Random (update) sdpCNN model: initializing word embedding randomly and updating word embedding during training.

Method	BioInfer	Aimed
	Combined model	75.3	66.6
Pretrained sdpCNN model	75.2	66.0
Random sdpCNN model	73.4	62.7
Random (update) sdpCNN model	74.1	63.3