Automatic literature screening using the PAJO deep-learning model for clinical practice guidelines

Table 1 Experimental results of different models when evaluated by Precision, Recall, Specificity, Accuracy, F1 Score, FLOPs, and area under the receiver operating characteristic curve (AUC)

Model	Precision	Recall	Specificity	Accuracy	F1-score	AUC	FLOPs(T)
Random Forest	59.11	79.10	76.54	77.31	67.66	85.28
L1LR	75.00	52.24	92.54	80.45	61.58	85.64
BiLSTM	64.86	59.70	86.14	78.21	62.18	82.21	0.072
BiLSTM + Attention	66.67	68.66	85.29	80.30	67.65	83.63	0.080
TextCNN	66.67	61.69	86.78	79.25	64.08	85.22	0.0001
TextRCNN	61.54	79.60	78.68	78.96	69.41	86.15	0.077
PubMedBERT	71.69	78.11	86.78	84.18	74.76	89.59	3.047
PAJO	71.55	82.59	85.92	84.93	76.67	91.84	15.236

Definitions — BiLSTM Bidirectional long short-term memory, L1LR Logistic regression with L1 penalty, PAJO Paper title, Abstract, and Journal model, TextCNN Text-based convolutional neural network, TextRCNN Text-based recurrent convolutional neural network, PubMedBERT the fine-tuned PubMedBERT model

ISSN: 1472-6947