Grants talk:IdeaLab/Wikidata Paintbrush

What is your experience with Listeria and its associated Javascript (wikidata:User:Magnus_Manske/wd_edit.js). I recall using the Manske system for annotation of painting, - perhaps this one: Wikidata:User:Fnielsen/Autolists/Værker af L.A. Ring i Den Hirschsprungske Samling. I wonder want another system would bring extra? — Finn Årup Nielsen (fnielsen) (talk) 19:56, 29 June 2017 (UTC)

Sharing projects

Dear all,

It is really great to see this project which targets an important lack of Wikidata.

Actually, I'm working on a similar project (prototype available at http://imagedescription.histoiredelart.fr/webapp) since one month, after discussing this lack with another Wikidata volunteer. My aim is to develop an interface to describe depicts of paintings and maybe to run a collaborative project around to educate people to the value of Wikidata (so, very similar to yours :)).

Currently, the tool collects every Wikidata item which is a painting and which has a picture. When a user launches it, it picks randomly one painting and asks the user to describe it. It currently stores the added depicts in a mysql database, and the idea is to push the data into Wikidata after some point. There's also a ranking function to add gamification into the project.

The tool should be finished in around 2 weeks. I still have to render properly some data (dates) and develop functions to work on specifics collections, periods et creators. Also need to fix issues and to finish to translate it in English. It will be a first version to test and which can be used by people. Then, the idea is to work on a second one with data mining in order to suggest depicts to the user.

Source code is available at https://github.com/orgs/ImageDescriber The API is based on Symfony3 and the webapp on AngularJS

If you would like, maybe we could merge the two projects. It would be sad to work on the exact same thing. Feedback on the prototype are very welcomed, also idea and involvements!

Best --Gordibach (talk) 21:44, 1 July 2017 (UTC) (cc Pharos, Spinster, Susanna Ånäs (Susannaanas))

@Gordibach: This is really awesome! Thank you so much for sharing such a cool project. I am really looking forward to it being more widely available, Astinson (WMF) (talk) 18:57, 22 August 2017 (UTC)

@Astinson (WMF): Hi Astinson, thanks for your message. I assume you didn't succeed to reach the app, it was down since the upgrade of this week-end. It's now back. New URL is https://imagedescription.histoiredelart.fr/#!/ The final version should be available by the end of this week. Feel free to use it :) --Gordibach (talk) 21:29, 22 August 2017 (UTC)

Hi @Gordibach: Thanks for the live demo: I have a few initial concerns, if this is supposed to be going directly into Wikidata: how do we encourage or register confirmation, of what, in many cases, will be very subjective descriptions of the content? For example, I saw a painting with Napoleon on horseback: because I know a fair number of folks won't know its Napoleon, or my confidence in identifying that person might be off (especially, if its a less well-known/distinctive individual). Is there a way to build in confirmation from users, before its get pushed into Wikidata? That would help, especially as folks want to make sure that depicts tags are descriptive enough, and not redundancies of similar but divergent topics.

Also making sure that @SandraF (WMF): sees it. Astinson (WMF) (talk) 20:04, 23 August 2017 (UTC)

@Astinson (WMF): Hi Astinson. Actually, the data isn't store directly into Wikidata. My API stores it, and I can send it to Wikidata via another script (wikibot). So it is possible to look on the published data, logs are available at the following url with possibility to view each depict added: https://imagedescription.histoiredelart.fr/#!/statistics. I plan to add a delete button in this page to remove incorrect depicts. Does it answer your concern? Best! --Gordibach (talk) 20:29, 24 August 2017 (UTC)

@Gordibach: I would think you would want two options: both a delete and adding a "confirm" button - that can trigger that being added by the bot. This could allow for these concepts to be reviewed by at least one more person, or possibly even more if the data quality turns out to be poor, by separate contributors -- that way relatively inexperienced folks or folks with poor interpretations of the media don't rampantly add lower-quality depicts statements that are completely invisible in the recent changes feeds on Wikidata (and eventually Commons, I would assume). If you leave this process of "add your own description", there is likely to be a lot of original research that some folks don't understand is original research/subjective interpretation -- or adding overly general (i.e. military general vs Napoleon in the example above), or overly specific (i.e. classifying as a particular type of bird in an illustration, when that identification is not clear) depicts comments that aren't actually the right kind of statements for the property. Also, if you can store the user names for each of these actions via oauth, it would allow your bot to add them to the edit summary, to provide attribution for the actions. Unlike the kinds of data that you get from Mix'n'match (which is a "Same/Same" match), or the Wikidata game (which is "yes/no/not sure" options) a wide open "add what you think this is" is bound to need some process for creating confidence in the statement -- and is how other organizations doing similar gamification on media, like Zooniverse, ensure that we aren't attributing arbitrary new and/or inappropriate data. Astinson (WMF) (talk) 22:35, 24 August 2017 (UTC)

@Astinson (WMF): Hi Astinson. I agree with your point of view. Unfortunately, I think it will take some time for me to develop such a controller page (especially the OAuth side). I hope having time to do it before the end of September. I'll keep you in touch! --Gordibach (talk) 20:47, 27 August 2017 (UTC)