Skip to main content

Table 4 The effectiveness of the integrated attention mechanisms according to mean F1 scores for 30 different random seeds

From: Relation extraction between bacteria and biotopes from biomedical texts with attention mechanisms and domain-specific contextual representations

Model

PE

F1 score

Fulls

SDPs

 

Mean

SD

Min

Max

BLSTM

CNN

45.96

2.87

42.09

52.19

BLSTM

CNN

48.49

4.76

38.75

55.40

BLSTM-Attn

CNN

49.02

3.62

42.03

56.51

BLSTM-EAttn

CNN

50.24

3.72

43.14

55.72

BLSTM-EAttn

MAttn

53.42

2.51

46.67

56.70

  1. All of the highest scores are highlighted in bold except for the SD. The first-row results derive from the best results of previous experiments (i.e., the last row in Table 3). Note: “PE” denotes positional encoding, “Attn” denotes the use of only Additive attention, “EAttn” denotes the use of both Additive and Entity-Oriented attentions, and “MAttn” denotes the use of Multi-Head attention