IARNN-Based Semantic-Containing Double-Level Embedding Bi-LSTM for Question-and-Answer Matching

Table 2

Advantages of pretraining method on Bi-LSTM.

Pretraining method	Train acc (%)	Train loss	Test acc (%)
Double-level	84.23	0.95	79.15
Character-level	80.75	1.47	77.23
Word-level	83.49	1.35	76.85