CSR-dataset 8,600 SR, 540,000 primary studies

[mmantyla]

Integrate this dataset to the runner.

https://pmc.ncbi.nlm.nih.gov/articles/PMC12527522/

https://data.mendeley.com/datasets/7sgmg89zb6/1

[mmantyla]

Dataset 1 vs 2. Summary dataset 2 has more junk (easy excludes)

Dataset 1.

" Each Cochrane review’s abstract has a description of the background, objectives and selection criteria of individual studies (e.g., randomised controlled trials, cohort studies, etc), with a list of included studies, excluded studies and additional studies that are related to the review topic but were not included in the evidence synthesis.
...
The list of included, excluded and additional studies can be used to approximate how publications are screened during the screening process of a systematic review.
........
However, the disadvantage of using the Cochrane dataset is that it inflates the false-positive rate (false-negative rate was unaffected) during model evaluation.
" https://pmc.ncbi.nlm.nih.gov/articles/PMC12527522/#sec6

Dataset 2.
"To evaluate the screening performance in actual systematic review, we compiled an additional test set by simulating the searching and screening process of 22 systematic reviews published in the Cochrane Library between 15 December 2023 and 15 May 2024. To do this, we followed the search strategies of these reviews to construct the set of initial publications for each review. Given the diverse search strategies used in different reviews, we only replicated the search in Medline, Embase and Cochrane Central."
https://pmc.ncbi.nlm.nih.gov/articles/PMC12527522/#sec6