Fuzzy anchoring to compute document similarity?

[Daniel-Mietchen]

We have the https://github.com/wpoa/JATS-to-Mediawiki converter, but in some cases, it does not recognize an unusual way of tagging things, so that sometimes a reference or an infobox or something in a table might be missing or otherwise deviating from the original.

It seems to me that fuzzy anchoring (as per https://hypothes.is/blog/fuzzy-anchoring/ ) requires this problem to be solved, so I imagine that we could use it - perhaps with some tweaks - to score the perfectness of the fit of the Wikisource copy to the PMC original. Is that assumption correct?

[nickstenning]

Daniel asked me to provide a bit more context on how the anchoring code in Hypothesis works. The blog post linked provides a passable overview but much of the detail has changed and the code itself is a fair bit easier to understand than when that post was written.

The basic approach is this: for each annotation target we store several different "selectors" -- that is, serialisable data that describes the original location of the annotation in the document. We can then use these selectors either individually or in combination when reanchoring annotations to the page.

We do this through the use of a series of DOM anchoring libraries:

https://github.com/hypothesis/dom-anchor-text-quote
https://github.com/hypothesis/dom-anchor-text-position
https://github.com/hypothesis/dom-anchor-fragment

You can see how these are used to reanchor annotations in this part of our code:

https://github.com/hypothesis/h/blob/a71b6a8/h/static/scripts/annotator/anchoring/html.coffee

The anchoring code is different for different document types. See, for example, here for the PDF anchoring code.