[11:13:42] Forwarded from dpriskorn: @harej and I talked yesterday about the lack of trust in statements in Wikidata. [11:13:43] In Wikipedia we have references and a host of eyes keeping the text and references aligned. [11:13:45] In Wikidata most statements only ever get touched by machines. [11:13:46] We need a good solution that scales to billions of statements. [11:13:48] I just read a paper sent to me by Magnus Sälgö aka salgo60 and in it was described how Facebookor some other big tech handles this challenge. [11:13:49] Inspired by that I suggest the following: [11:13:51] A community based trust model that is enabling AI mediated trust-scoring of all our statements. [11:13:52] How does it work? [11:13:54] The AI gets fed with users trust indications. [11:13:55] When I read a statement with a reference in the UI I can indicate my trust level on each of the references. [11:13:57] The trust level must be on a scale 1-3 where 3 is the highest. [11:13:58] The trust level must be able to be revoked/removed. [11:14:00] As a user I want to be able to search all my trust indications. [11:14:01] If my trust in a source has changed eg after a hostile takeover pf a newspaper or government I want to be able to revoke in bulk. [11:14:03] All statements without references defaults to untrustworthy and the editors should get a warning in the UI. [11:14:04] WDYT? [13:57:11] Why does it have to be on a scale from 1 to 3? That seems awfully arbitrary. (re @dpriskorn: @harej and I talked yesterday about the lack of trust in statements in Wikidata. [13:57:12] In Wikipedia we have references and a host of e...) [14:01:21] When you say "All statements without references defaults to untrustworthy and the editors should get a warning in the UI." Does that mean that you think that a query using `wdt:` should not yield any result or would you still consider it truthy? (re @dpriskorn: @harej and I talked yesterday about the lack of trust in statements in Wikidata. [14:01:21] In Wikipedia we have references and a host of e...) [14:54:28] putting warnings on all statements without references in wikidata would not improve the quality of the data. it would most likely cause people to start adding more "imported from wikipedia" references (whether it was or not) to make the warnings go away [14:54:57] the only way we're going to get more people to add good references is by making it *easier* to add good references [14:55:18] or have the same effect as the California cancer warnings – it's everywhere, so people just ignore all warnings, including the ones they really shouldn't ignore [15:21:36] It was just an arbitrary scale I made up. Any scale that that works for the editors and the ML is fine I guess. (re @Jan_ainali: Why does it have to be on a scale from 1 to 3? That seems awfully arbitrary.) [15:23:41] I have not thought about that. I didn’t know it was a possibility even. [15:23:42] Now that you mentioned it it might be nice to have a wdtt=wikidata truthy trusted prefix which relies os the rating I suggested. (re @Jan_ainali: When you say "All statements without references defaults to untrustworthy and the editors should get a warning in the UI." Does ...) [15:25:13] Or perhaps use nudging wisely which we actually do already with the constraint warnings. They affect my behavior a lot FWIW 😀 (re @Nikki: the only way we're going to get more people to add good references is by making it *easier* to add good references) [15:27:38] This book in Swedish introduced me to nudging in software design [15:27:39] https://www.adlibris.com/se/bok/beteendedesign-psykologin-som-forandrar-tankar-kanslor-och-handlingar-9789127819191 [15:28:50] yeah, the solution is a Citoid > Wikidata translation layer, which would allow for easier citations [15:29:52] the reason that it doesn't exist is that it's difficult for computers to identify which item "Doe, Jane" corresponds to, since it's usually unstructured [15:34:17] we normally use https://www.wikidata.org/wiki/Property:P2093 to store unresolved names, and test.wikidata.org has had a feature to help with adding references for a long time, but for some reason it's never made it onto the main site [15:36:57] here's a random test item with a statement where you can see try adding a reference to see what I mean https://test.wikidata.org/wiki/Q78666 [20:41:54] That seems like a really bad reason for not having a user-friendly citation tool. The value differential between having a structured reference and having an unstructured reference is much much smaller than between having an unstructured reference and not having a reference at all. (re @wmtelegram_bot: the reason that it doesn't exist is that it's difficult for com [20:41:55] puters to identify which item "Doe, Jane" corresp...) [20:51:57] https://phabricator.wikimedia.org/T285498 [20:52:19] https://meta.wikimedia.org/wiki/Community_Wishlist_Survey/Updates#October_19,_2021:_Declining_the_Bibliographic_Bot_Wish [20:54:08] https://phabricator.wikimedia.org/T199197 [21:22:02] CiteTool was made into a gadget (courtesy of Nikki ) just the other day. https://www.wikidata.org/wiki/Wikidata:Tools/CiteTool Next step could be to enable it by default (re @gtisza: That seems like a really bad reason for not having a user-friendly citation tool. The value differential between having a struct...) [21:23:49] it doesn't do as much as the thing on test.wikidata.org though and I still don't know why that never made it to the main site :/ [21:35:59] looks like a question for Samwalton9, since they're the person who last commented on it, in April 2020 [21:36:21] I also don't know what the legal concerns mentioned in https://phabricator.wikimedia.org/T199197 are