Starting things:

For small samples, when determining which variations of the pipeline to use on a larger corpus or articles, and also to review output of any corpus:

• info about the sample itself: # of titles, publication date span, # of articles, average OCR quality [replace with something actually useful]
• how many toponyms total and of each type (location, building, street, etc.) + statistics about the confidence score for each
• which toponyms could not be resolved, or have a confidence score below a certain threshold?
• average # of toponyms per article per title (or other variables)
• for all resolved toponyms, map (possible to interact with based on confidence score/only view one or more types of place)
• spatial "center" of resolved toponyms (e.g. where do they cluster)
• [how to also to bring in OCR quality to sense check/viz results]

• Potential for this notebook to be used to show people what the impact of different choices in using pipeline as well as for internal use

Resources re: OCR quality

• https://www.newseye.eu/fileadmin/user_upload/NewsEye_CONF_Presentation_17032021.pdf
• https://web.archive.org/web/20211113214529id_/https://dl.acm.org/doi/pdf/10.1145/3486187.3490206

First draft available here, based on sample data from @npedrazzini. (only one article—I think?—and so on).. We can easily scale this up once we have more data.

Awesome @kallewesterling ! Will explore tomorrow first thing!!

One Q about the below viz - I'm never sure whether the label I see when I hover is for the individual grey circle or the parent circle. Is there a way to make that clearer?

One Q about the below viz - I'm never sure whether the label I see when I hover is for the individual grey circle or the parent circle. Is there a way to make that clearer?

Yes, I just need to create some kind of "tooltip" for the hovering. the current solution is just a temporary one.

Click for full-res.

Moving future work on this to #179