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

Table 2 Results of ablation experiments when evaluated on Precision, Recall, Specificity, Accuracy, F1 score, and area under the response operating characteristic curve (AUC)

Model	Precision	Recall	Specificity	F1-score	Accuracy	AUC
PAJO-T	69.19	72.64	86.14	70.87	82.09	86.71
PAJO-A	72.22	71.14	88.27	71.68	83.13	89.72
PAJO-TJ	58.87	77.61	76.76	66.95	77.01	86.98
PAJO-TA	70.40	78.11	85.93	74.06	83.58	89.15
PAJO-AJ	71.95	79.10	86.78	75.36	84.48	89.98
PAJO-FULL	71.55	82.59	85.92	76.67	84.93	91.84

PAJO-FULL performed best in all metrics apart from Precision. Definitions: — PAJO Paper title, Abstract, and JOurnal model, PAJO-A Base model only intakes abstract features, PAJO-AJ Base model only intakes abstract and journal features, PAJO-FULL Complete model, PAJO-T Base model only intakes title features, PAJO-TJ Base model only intakes title and journal features, PAJO-TA Base model only intakes abstract and title features

ISSN: 1472-6947