Use PRISMA techniques and and comprehensive meta analysis software (CMA)

There are a variety of "hedges" that have been developed for different databases that provide a focus on different research types. You could adapt a typical pubmed qualitative hedge to the different databases you are using. This is an example:

AND (“interviews as topic”[Mesh:noexp] OR “focus groups”[Mesh:noexp] OR narration[Mesh:noexp] OR "qualitative research"[Mesh:noexp] OR “semi-structured”[TIAB] OR semistructured[TIAB] OR unstructured[TIAB] OR structured[TIAB] OR informal[TIAB] OR “in-depth”[TIAB] OR indepth[TIAB] OR “face-to-face”[TIAB] OR guide[TIAB] OR guides[TIAB] AND interview*[TIAB] OR discussion*[TIAB] OR questionnaire*[TIAB] OR “focus group”[TIAB] OR “focus groups”[TIAB] OR qualitative[TIAB] OR ethnograph*[TIAB] OR fieldwork[TIAB] OR “field work”[TIAB] OR “key informant”[TIAB])

There are other suggestions here: https://guides.nyu.edu/locating-qualitative-research/medline

Based on the protocol you have you can do a decision tree algorithm where you describe the different steps to decide if the article should be retained or not. I don't know the field of research you're interested in but there may be several tutorial or seminal article tailoring the method to your area.

Si vous pouvez avoir l'aide d'un bibliothécaire cela peut aussi vous aider beaucoup. La lecture des articles peut être relativement rapide pour savoir lesquels inclure ou pas. Ensuite le plus difficile est d'extraire les données des articles que vous devez lire au complet.

There are some fairly well validated study type filters that can help, for example for RCTs. You can add them into your search strategy. There are also some reasonable AI tools for study type classification.

The normal process for a systematic review/meta analysis would be to download the titles and abstracts of your thousand papers into a software such as EPPI reviewer or covidence or one of the free options and do a 'sift on title and abstract to remove the ones that don't match your population/intervention/control/outcomes that you set out in your protocol. That should reduce the numbers enough that you can order the remainder at full text to check in detail for whether they are includes or not.