Grant project

Automatic citation tool in Wikimedia. URL -> citation. fetches metadata from source (author, title, etc.) which is far easier than doing this manually. Problem is that sometimes the metadata has errors or is missing fields. this system (citoid) requests URL and extracts the metadata from the webpage. It uses Zotero (3rd-party) software to extract the metadata. if metadata appropriately embedded via standards, great! generally not the case though so Zotero community writes specific "Translators" for each domain. A lot of work though and webpages change and the translator breaks.

To fix this, a few options:

Manual: doesn't benefit others and can be slow

Convince webmasters to fix site: very slow and doesn't work often

Fix Zotero translator: requires coding experience though so high barrier to entry and slow iteration

Web2cit attempts to solve these issues: complements Citoid. When Citoid works, no web2cit needed. When it doesn't though, web2cit can help fix the gaps with a lower barrier to entry by just editing configuration.

Demo! If web2cit is installed (user script) and you use Citoid, both original Citoid output and web2cit output shown. Can investigate web2cit output and edit output to be what is actually expected. The saved edits are saved as JSON on metawiki.

Once saved, can see how good the overlap is between extracted content and expected content. Could stop here but can also edit the extractor. Can do things like setting hard-coded values (fixed), using a specific Citoid field, HTML selector (xpath), or pull from site's json-ld.

under Web2cit/monitor, can see table (updated approximately hourly) of all the domains and their scores and details. by watching the page, can see if score changes because something changed in the site.

Don't need user script (can adjust the URL you add to Citoid) but much easier to install user-script

As part of project, measured current performance of Citoid. To do so, extracted references from Citoid for featured articles and compared to manually-curated results. Used these for automatically generating a set of initial expected web2cit outputs.

Can we name and shame the publishers who do poorly?

Is it feasible to auto-generate patches for Zotero?

Yes but hard. Probably better to hand-code but the tests themselves can be very useful for anyone with the skills to create translators on Zotero

Will this work for any language edition of Wikipedia?

Yes, we chose a subset of fields that would provide a basic citation. for instance, we don't support DOIs yet but that might be a good one to add. the templates are stored on meta so they are not language-specific. they output fields names that match to what is used by Zotero/Citoid so as long as a community has configured Citoid to work with their citation templates, they can use web2cit. Perhaps some UI etc. issues with certain languages though -- e.g., right-to-left languages.