Creating a random set of references
taleevenhuis opened this issue · comments
Feature Request
Following the SAFE-procedure (Boetje and Van de Schoot, 2023: https://osf.io/preprints/psyarxiv/c93gq), is it possible for the Screening Phase to create a random set of, say, the most recent 33 references, the middle 33 references and the oldest 33? In this way the model in ASReview is not only trained on recent terms, but also on older terms. If you use only the most recent references, the model will only be trained on new terminology cq jargon. Example: PTSD is a recent term, but Shell Shock is a term which describes the same phenomenon in WO-I.
Interesting idea! I would have thought this would be taken care of if the screening was truly random, then in theory these different naming conventions could be picked up. It does feel safer using your idea though to make sure there is not a jargon bias. Especially if a researcher is not aware of these terms ahead of time so did not mark those articles as relevant initially to train.
@taleevenhuis has a point. Disorders indeed have their terminology revised across the years, whether to include new concepts or to avoid stigmatizing language. PTSD was once called shell shock, traumatic neurosis, combat fatigue, etc., and has recently encompassed terms like secondary traumatic stress and compassion fatigue. More recently, Hoarding Disorder was considered a disorder itself rather than solely a symptom of OCD. For studies on psychology and psychiatry, taleevenhuis' idea could be helpful
This might be a good idea, especially in fields where terminology undergoes significant changes over time. However, I am curious about the relevance of articles using older terms. One might question the significance of these articles and how applicable they are to current research.
The feature has been introduced in asreview/asreview-datatools#41 is now available in datatools.