Skip to main content
. 2022 Apr 8;3(5):100483. doi: 10.1016/j.patter.2022.100483

Figure 2.

Figure 2

Model predictive performance

(A and B) Shown are cross-validated receiver operating curves (ROCs) for the Full-NLP-CNN model trained on metadata and title and abstract embeddings (orange for validation, purple for held-out test) and the Metadata-MLP model (red) and the Citation-LogisticRegression model trained on citation count per year (blue) for guideline or policy inclusions (A) and patent inclusions (B). The confidence intervals are ±2 SD on cross-validation.