Assess (meta)data quality

[PGijsbers]

For a given dataset, assess its (meta)data quality.
This may include things like:

• Does it have a clear title?
• Is a clear description of the data provided (note that this is not specific to the description field)?
  • Do the features have clear names and are they described?
  • Does the description mention where the data is sourced from?
  • Does the description mention the intended use-case for the dataset?
  • Is there author information and a way to cite it?
• Is the data used by others?
  • Does it have runs?
  • Is it included in benchmarks?
• If multiple versions of the dataset exist, is it clear what the difference is to other versions?

A good source of inspiration can be the "datasheets for datasets" paper.

We can use this for multiple purposes, e.g.:

• Identify what metadata is frequently missing, and better prompt the users to provide this information on upload.
• Add prompts to the dataset page that make users aware important information is missing, or allowing them to add it.
• Influence dataset search results based on dataset quality.
• Set up projects to improve dataset quality w.r.t. specific aspects which may be easier to crowd-source/automate.
• Deactivate datasets of exceptionally poor quality.

[PGijsbers]

Unassigning Taniya as she is currently also busy with the filter extraction (and AIoD). Taniya, you are free to join back on this issue later. I am sure this general topic is not something easily solved. That said, in the mean time I want to make clear that it is also open for contributions from other people.

[Taniya-Das]

Sounds good, I will attach the code repo so others can contribute.