Fig. 2
From: Automatic literature screening using the PAJO deep-learning model for clinical practice guidelines
Dataset construction process. We first applied our keyword and phrase retrieval matching rule to all PubMed articles. We then performed deduplication and removed records for which the associated journal information was unavailable. Finally, we classified Set A as the complete collection of 27,406 samples. Set B was classified with 1,005 articles that were cited in the CGPs and systematic reviews (positive samples). Finally, Set C contained the 26,401 negative samples