[09:41:11] @SMalyshev ... yes (sorry, chat seemed not update then I left tab open), w2s as the github doesn't work for me, vendor/autoload.php is missing, the package dependencies aren't clear, e.g. PHP ParserGenerator etc ... what are the extact dependencies .. [09:42:39] I'm running blazegraph and try to learn about SPARQL, for now I access wikidata with wdq tool which right now doesn't work with SPARQL (already posted an issue on github about it) - I try to comprehend how wikidata can be queried ... [09:43:22] e.g. I pushed latest.json into MongoDB direct, simple queries work, but the moment I like to travers some 'claims' MongoDB timeouted and tries to load entire set into memory (>40GB) [11:42:08] is there anyone with +2 right on mediawiki/core to take a look at this patch? https://gerrit.wikimedia.org/r/278841 [11:42:19] aude: ^ :D [12:17:38] Where's Lydia? [12:18:35] in a meeting, I see [13:37:21] what's the recommanded way to handle trolling/vandalism such as this one? ^^ https://www.wikidata.org/wiki/Q51793 label: Jar Jar Binks / description: *Sith lord* in Star Wars [13:38:14] those clicking "cancel" on the different incriminated edits do the trick? [13:38:43] zorglub27: Rollback/ undo or restore a "good" revision [13:42:12] hoo: thx [13:54:21] DanielK_WMDE__: is https://phabricator.wikimedia.org/T133381 something that would be ready for the next sprint already? or more something for the storytime/other discussion meeting? [14:02:13] Tobi_WMDE_SW: yes, but this is more urgent: https://phabricator.wikimedia.org/T132592 [14:10:51] DanielK_WMDE__: thx! yes, that ID serialization ticket is there already and it is ranked higher [14:44:20] SMalyshev: The Wikidata Query Service is very fast. Thank you for your hard work. [14:49:07] Hi all! I'm building a web application where users can search for protein/compound/etc. names and view their 3D structure using WebGL. I'm currently directly using the PubChem/RCSB API to provide some autocomplete data, but I found that Wikidata also has many of these names with their DB indices! The most important reason to try to autocomplete compound and protein names via Wikidata is to allow users to search in different languag [14:49:09] es (RCSB and PubChem, a protein and a compound database, generally only provide English names). However, I could not find a suitable API for this. I tried building a SPARQL query but that became waaay to slow. I could not find an option to limit full-text searches to a specific subclass in the search APIs. Can you please help me? [14:57:54] Hi [14:58:39] Am looking for a tool to browse the class hierarchy of wikidata [14:58:45] any ideas ? [15:10:08] melderick: like this? https://tools.wmflabs.org/sqid/#/view?id=Q36534 [15:10:54] hermanbergwerf_: you want to use the wbsearchentities api module. let me find the link... [15:11:13] hermanbergwerf_: https://www.wikidata.org/w/api.php?action=help&modules=wbsearchentities [15:11:25] I looked at that one, but I cannot limit the results to chemical compounds only [15:11:35] So I have to iterate every single one to see if it is a compound [15:13:18] hermanbergwerf_: that is true. we don't have good support for combining semantic queries with soft text matches. it's tricky to implement this in a way that performs well. the underlying technologies (fulltext search vs. query engine) are very different. [15:14:05] Your use case is common enough that that I hope we will have something nice for this in the future. For now, you can try to combine the results from the SPARQL endpoint with the results from the label search on your side. [15:15:10] DanielK_WMDE__ : I was looking for something more graphic, but it's a start :) [15:15:11] hermanbergwerf_: you don't have to iterate over every single one. You can query and cache the set of all compounds. Then you filter the result you get from the search against that cached list (list too big? use a bloom filter). [15:15:49] Yeah, I thought about setting up an Elastic DB for that, not sure how many million names there are in wikidata [15:15:55] melderick: since it's not a hierarchy but a graph, a diagram gets very messy very fast. cool to look at, but close to useless in my experience [15:17:17] DanielK_WMDE__: oh i see [15:17:53] hermanbergwerf_: an elasticc db for the search part? just wait a few months, that's what we are working on. [15:18:15] for page content, wikimedia already uses elastic anyway. we just have to make it work for item labels, too [15:19:31] i recommend to use wbsearchentities live, and cache the result of whatever sparql query you want. my guess would be that you end up with a million or so item IDs. Each can be expressed as a 32bit integer. That's 4MB. Not too terrible to keep in memory. [15:20:32] Not sure, I don't want to put too much load on the client side, but with today's RAM, probably an option [15:21:16] I can prefetch all names for all languages serverside and pass them to the client as JSON (per language of course) [15:22:23] Or use another tricky method, many wikidata chemical compound entities have 'chemical compound' or 'pharmaceutical drug' as description [15:22:28] I could make a whitelist of descriptions [15:22:37] tricky but it could work [15:23:05] (e.g. check out the results for Benzene in Finnish: https://www.wikidata.org/w/api.php?action=wbsearchentities&search=Bentseeni&language=fi) [15:28:17] Were any algorithms/conventions used when importing chemical compounds? https://www.wikidata.org/wiki/Help:Description does not mention any but for compounds there seem to be some pretty straightforward constrains. [15:28:26] At least that would be a very lightweight solution for me. [15:28:40] I can still combine it with search results from other databases. [17:23:22] * aude waves [18:33:53] does anyone know of pre-trained entity embeddings for wikidata? like word embeddings calculated with word2vec or gloVe? [18:34:20] theoritically, it is possible to use transe or similar models to calculate [18:34:41] *TranSE [18:35:19] I aim to calculate them, however it would be very nice if there exists such a resource [19:33:41] How do I get the Wikidata page id into a template when the small Wikipedia I’m working on has no Wikidata module? Is it like {{#property:pageid for this article}} or something? [20:50:53] LL_: there currently is no way I know of, but it's a sensible request to have a parser function for this. Please file it on Phabricator. I think we already have a request to allow this from Lua. [22:07:43] hi [22:08:07] what's the right way to add a reference in https://www.wikidata.org/wiki/Q23921645 ? [22:08:35] for the date of birth [22:09:17] cf. http://data.bnf.fr/12629705/andre_de_paniagua/