[00:04:05] bd808 cool thx [11:04:09] What are the video tools used during the showcase? [11:38:51] jeblad: you can follow the stream on youtube https://www.youtube.com/watch?v=fiD9QTHNVVM [11:39:44] more info on how to participate here https://www.mediawiki.org/wiki/Wikimedia_Research/Showcase [16:18:48] hi all. we are doing the last checks and tests for the Research Showcase. [16:19:00] We expect to go live in 12 min. :) [16:19:33] and as you may have noticed, J-Mo is your host today in IRC. :) [16:21:56] thanks Leila :) o/ everyone! [16:29:28] I might be here for 15-20 mins [16:30:39] full meeting day ... i'll have to watch this research showcase video later. [16:31:12] the showcase has begun! [16:31:23] subbu: ok. :) [16:40:37] AV is good from here (Birmingham, England) [16:40:52] Pigsonthewing-te: thanks for letting us know. :) [16:40:55] Awesome! [16:50:49] mgerlach: II know I can follow on Youtube, I asked what tools they use for creating the videos. [16:51:36] jeblad did you have a question? [16:52:48] [12:04] What are the video tools used during the showcase? [16:53:30] usually we use Google Meet. But today we're using Zoom [16:53:33] Bingo square: y axis does not go to zero [16:55:24] Q: What would word embeddings add to this strategy? Would capturing the semantic relatedness of the anchor words give users more flexibility/power? [16:55:39] ack halfak [16:55:48] Thanks J-Mo [17:13:28] Emerauld: can you mute dsaez ? [17:14:31] it's done. thanks. [17:19:04] Please catn someone post a link to the meta page with hat mega-table of links? [17:19:17] "can" [17:19:39] "that" - sorry for my bad typing [17:22:47] Pigsonthewing-te: I don't have the link handy. will ask the speaker (not mentioning id to avoid pings while presentation) after he's done. [17:22:55] (or you should feel free to do it:) [17:24:26] I was hoping that someone was quicker than me at typing the URL that was on the slide ;-) [17:24:41] not me, at least. ;) [17:28:21] Pigsonthewing-te is this the page you want? https://meta.wikimedia.org/wiki/User:Diego_(WMF)/COVID-19_Pages [17:29:32] latest link on the slides is https://tools.wmflabs.org/wiki-topic/ [17:31:11] Thank you, J-Mo; yes. [17:32:07] Pigsonthewing-te: I believe dsaez plans to set up automated daily updates to that meta page, so feel free to share it [17:34:10] Q: How to adjust the results as the Wikidata item often describe the tecnical part or of items, while the article describe the qualities of the entities? [17:34:45] I'm writing a piece on the Wikimedia community's response to COVID-19; I'll include it there. [17:34:53] nice, thanks [17:34:59] We don't really have a problem developing native embeddings on various languages. [17:35:10] But I guess that could help us get past the requirement of a sitelink [17:35:17] Pigsonthewing-te shameless plug, there's also this report! https://en.wikipedia.org/wiki/User:HostBot/COVID-19_article_report [17:35:27] (already linked from WP COVID-19 [17:39:27] sorry for interrupting, Zoom is tricky [17:39:30] huh? [17:39:49] yeah, no prob Diego. I've done worse! [17:40:25] hi jeblad, sorry I did not see your question! [17:40:50] thanks for your participation, everyone and to those of you who made it happen. our speakers as well as djellel, J-Mo, janna_WMF, Emerauld. (I hope I'm not missing anyone:). [17:40:50] No proble, it is kind of hard to answer. [17:41:11] if it is for isaacj or dsaez, they are both still here... [17:41:12] Thanks all, bye. [17:41:26] thank you jmo [18:04:19] jeblad: just reading your question. yeah i think that gets to the heart of the challenge. the wikidata items describe a different aspect of a concept than what we are usually interested in when modeling topics. the approach we're currently considering that bridges the simplicity of wikidata with the richness of wikipedia articles is using the links in a given article to predict topic. so rather than using the words in an article, [18:04:19] take all the links in an article and represent them based on their associated wikidata items. then build a model that predicts topics for that. that way, to build a model for topics in say Haitian Creole (an example given in the presentation), you don't need a model that can parse Haitian Creole and understand each word, you can just use the same Wikidata representations you use for any other language [18:42:42] Use the wikidata rdf embedding for the links from articles, that could work. [20:24:56] dsaez: if you are planning to be around for another couple of hours, let's have a quick touch-base re covid-19 (15-min?). If not, we can do it tomorrow. [20:45:02] leila, do you want to do it now? [20:45:25] dsaez: I'll be available in 40-min. if that's too late for you, let's do it tomorrow? [20:45:57] it's ok, I'm being timezone-free these lasts days. [20:48:22] dsaez: ok. I'll put something on our calendars then. (I'm trying to detacch from screen and eat lunch before I talk with you;) [20:52:24] okk [23:45:49] if someone is interested, we have 36 humans in Wikipedia reported as cause of death COVID-19 https://bit.ly/2U13mWL