Table 2.

Hyper-parameter settings

Type	Hyper-parameter
Training	α=0.03,λ=10−8
Embedding	dim(emb(w i))=200
	$\mathit{\text{dim}}\left(\mathit{\text{emb}}\right(p_i),\mathit{\text{emb}}(d_i\left)\text{or}\mathit{\text{emb}}\right(l_i^e\left)\right)=25$
CNN	dim(emb(c))=25,C=3
	dim(r w)=25
Bi-LSTM-RNN (Entity)	$\mathit{\text{dim}}\left({\vec{h}}_i\right)$ or $\mathit{\text{dim}}\left({\stackrel{⃖}{h}}_i\right)=100$
	$\mathit{\text{dim}}\left(h_i^e\right)=100$
Bi-LSTM-RNN (Relation)	dim(↑ h a,↑ h b,↓ h a or ↓ h b)=100
	dim(h r)=100

dim denotes vector dimensions and emb denotes feature embeddings