Hey Jimmy, Thanks for the report. This problem is one that we've been aware of in Discovery for quite some time. It actually serves as a good example of a typical problem that we face in improving search: we know there's an issue with a small subset of searches, and could fix this problem easily with a hack, but that hack would make as many searches worse as it makes better. Meanwhile, better solutions take much more time.
But, I have good news! This quarter one of Discovery's goals [1] is to work on a proper solution to this very problem. We previously studied the problem in detail [2]. Now, following on from our upgrade of Elasticsearch last quarter, we're hoping that switching us over to BM25 [3] will fix many of these relevance issues, and we're investigating that more right now. Stay tuned! Thanks, Dan [1]: https://www.mediawiki.org/wiki/Wikimedia_Engineering/2016-17_Q1_Goals#Discovery [2]: https://phabricator.wikimedia.org/T125083 [2]: https://en.wikipedia.org/wiki/Okapi_BM25 On 28 Jul 2016 5:09 a.m., "Jimmy Wales" <[email protected]> wrote: > First, some context: > > I was in Philadelphia for the Democratic National Convention earlier > this week, where I had been invited to speak (in a small side event) > about connectivity and global development. I spoke about our work in > the languages of the developing world, and made a point to say that bad > laws in the developed world which might hurt our work can be damaging > for the development of the Internet in the rest of the world and urged > lawmakers to not just think of various Internet legal questions as being > "Silicon Valley versus Hollywood" but to understand that they impact how > our volunteer community and many other ordinary people online. > > Second, the story: > > The main conference was held in the [[Wells Fargo Center > (Philadelphia)]], an indoor arena where basketball and hockey teams play > normally. > > A journalist friend said to me that he "finally found something that > Wikipedia doesn't have" and he was surprised. What was that, I said? > "The history of Wells Fargo". What?!! Really?!! That seemed impossible > to me. He said we have an article about Wells Fargo that seems to be > mostly about the contemporary bank, and when you search for Wells Fargo > history there's also an article about the Wells Fargo History Museum. > > I popped on my phone and used my own personal preferred method of > finding things in Wikipedia: Google. I typed in "Wells Fargo history" > and sure enough, the first two links are history pages from their > official websites and the third link is Wikipedia - a normal state of > affairs. He started to apologize for raising a false alarm > > I asked him for more details on exactly how he searched, and explained > that I regard it to be very sad if some volunteers spend hundreds of > hours working on an article, painstakingly going over tons of details in > an effort to get it right, and then someone couldn't find it. > > Here's what he did - and I replicated the steps and all was clear. > > Go to http://www.wikipedia.org/ > > Make sure the dropdown in the search box is set to 'EN' - which it would > have been for him. > > Start typing 'Wells Fargo history' and watch as the dropdown selections > narrow. You'll have the experience that he had - you'll see the bank > article prominently featured and then various buildings (they have a > habit of sponsoring sports arenas in various US cities) and finally as > you start typing history it focuses in on the History Museum. > > If you don't choose any of those, then hit enter, you'll get to the > search results page. This is the one with a huge box of options at the > top (which will be confusing and frightening to people who aren't > already wikipedians) and then by my count the desired article is 13th on > the page: [[History of Wells Fargo]]. > > Now, I strongly suspect this could be fixed by making a redirect from > [[Wells Fargo history]] to [[History of Wells Fargo]]. > > Or a more serious fix could be had if the search engine understood that > very very often in English [[X of Y]] can be written [[Y X]]. ([[List > of French monarchs]] becomes [[French monarchs list]], see: > https://en.wikipedia.org/wiki/Special:Search?search=french+monarchs+list > where the desired article is in 10th place. > > But my point is not to argue for any specific fix. My point is to > illustrate that there is a real problem with search, that it is > impacting users, and that we should invest in fixing it. > > --Jimbo > > > > _______________________________________________ > Wikimedia-l mailing list, guidelines at: > https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines > New messages to: [email protected] > Unsubscribe: https://lists.wikimedia.org/mailman/listinfo/wikimedia-l, > <mailto:[email protected]?subject=unsubscribe> _______________________________________________ Wikimedia-l mailing list, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines New messages to: [email protected] Unsubscribe: https://lists.wikimedia.org/mailman/listinfo/wikimedia-l, <mailto:[email protected]?subject=unsubscribe>
