Poliloom - skip and flag statement conflict

@Monneyboi As I’ve been working through Poliloom for US politicians I’ve noticed there are two annoying behaviors.

  1. When a statement is vague or hard to verify the accuracy skipping the response results in continuously having the unanswered statement presented over and over in the queue to the same user. It would be preferable to count the number of skips for a statement then not show that suggestion to the same user as they probably didn’t miss answering, but are unsure if it should be accepted or rejected. For instance Q111530119 is suggesting an island as the location matching “Long Island” for place of birth, but geographic features often have multiples of common names in an area so this is a hard question to answer without some GIS work to check if other “Long Island” locations exist or a more specific citation. Its not necessarily a reject, but more a hard to verify to accept.

  2. For birth year extractions infrequently there have been inconsistencies within Wikipedia articles where the infobox has one date and the article body has a conflicting birth date and the citation is not clear on either. Importing both to Wikidata would be one approach, but maybe better would be to have a reject and add a template for citation needed or self contradiction for the statement in question. If not auto inserted this could be put in a queue to review the statement and fix the source or have the relevant template set to be added manually.

If either are on the roadmap, great. If not, these could be useful ways to improve the quality.

Another example of conflict between infobox and body content on start date and missing the content from the section further down which clarifies the probable reason for the infobox difference as its the elected date for the start rather than assumed the position as start time. :

Political career

Ziogas was elected to the Connecticut House of Representatives in 2016 and took office in January 2017, succeeding Frank Nicastro.[6]

Hi @wolfgang8741

Thank you for your continued efforts, they’re much appreciated!

skipping the response results in continuously having the unanswered statement presented over and over

That’s a real bug, and something that we should fix. I’m now tracking it, and will pick it up in the next sprint.

and add a template for citation needed or self contradiction for the statement in question

I see where you’re coming from, but don’t think we can afford to support this kind of source-specific logic for something this infrequent.

PoliLoom’s extraction logic isn’t Wikipedia-specific, it’s built to pull from arbitrary web sources, and Wikipedia just happens to be the lion’s share of what we ingest today. We’re currently working on trying to automatically identify and ingest sources from the rest of the internet, so I’d rather keep the pipeline source-agnostic.

For the core problem, a source that self-conflicts, I think the pragmatic approach is probably just inserting both dates into Wikidata rather than trying to arbitrate which one is correct.

Thanks again!