Table 1.

Quality metrics for the embeddings

Data	Dim.	Acc. (%)	RMSE	Recall (%)
PubMedBERT	768	69.7	8.4	–
TF-IDF	4,679,130	65.2	8.8	–
t-SNE(BERT)	2	62.6	10.2	6.2
t-SNE(TF-IDF)	2	50.6	11.2	0.7
Chance	–	4.3	12.4	0.0

Acc., kNN accuracy ($k=10$) of label prediction; RMSE, root mean-squared error of kNN prediction of publication year; Recall, overlap between k nearest neighbors in the 2D embedding and in the high-dimensional space. See experimental procedures for details.