Table 3 Comparison of number of parameters and inference speed of different models.

From: Adaptive feature interaction enhancement network for text classification

Model name

Number of parameters

THUCNews

Inference speed (ms/it)

F1 (%)

TextCNN

2.13M (1\(\times\))

5.52

91.03

AFIE-TextCNN

2.14M (1.01\(\times\))

9.10

97.55

DPCNN

2.21M (1\(\times\))

5.17

91.15

AFIE-DPCNN

2.25M (1.02\(\times\))

8.65

93.62

LSTM

5.04M (1\(\times\))

2.27

91.00

AFIE-LSTM

5.24M (1.04\(\times\))

6.53

91.56

Transformer

3.48M (1\(\times\))

3.68

89.74

AFIE-Transformer

3.49M (1.01\(\times\))

7.32

90.01

MacBERT

102.9M

42.25

94.56

  1. Models * and AFIE - * form a comparison group. The bold is used to mark the group with better experimental performance in this comparison.
Back to article page