Skip to main content

Table 3 Area under learning curve (ALC) for different methods aggregated over extremely rare (\(prevalence \le 0.02\))

From: When BERT meets Bilbo: a learning curve analysis of pretrained language model on disease classification

Method

HaoDaiFu (89 rare diseases)

ChinaRe (5 rare diseases)

BOW

0.3044

0.8454

BOW_EXP

0.3056a

0.9058

BOW_EXP_KG

0.3115a

0.9034

CBOW

0.1215

0.1945

CBOW_KG

0.1153

0.2136

LSTM

0

0

LSTM_KG

0

0

BERT

0. 3795ab

0.9028

  1. Figure 3 plots the learning curves
  2. aResult significantly higher than BOW
  3. bResult significantly higher than BOW_EXP_KG. (Fisher's randomization test, significance level \(\alpha = 0.05\))