Problem: The current search function is quite poor unless you know exactly what article you are searching for, and what it's title is. If you're searching based on a category (say, "French films of the 1970s"), the search is quite useless.
Who would benefit: Everybody using the search function.
Proposed solution: Use Wikidata to find search results. This would work by comparing terms in the search query against the Wikidata profiles of pages. So searching for "French films of the 1970s" in Wikipedia would find articles which are categorized as a "film", as "French", and as related to the "1970s". Those articles which would have all three would then be in the top search results if no article had a title with a close match.
If this were possible, we would already be doing it at Wikidata Query Service. Mapping English language phrases into Wikidata claims is impossible with some very serious AI. And even if it were possible, it would be much too slow for the regular search interface. Kaldari (talk) 02:47, 9 December 2020 (UTC)[reply]
Do any support voters have a response to this? I see that this request is gaining a lot of traction, but if that's just since we all wish search functioned better rather than since this would actually be a good idea, it might not be the best thing to push to the top. {{u|Sdkb}}talk10:32, 9 December 2020 (UTC)[reply]
only because who's working on this right now couldn't figure out a way to do it doesn't mean it's impossible. if enough people think this should be done, there might be a threshold of collective brain power which is able to find a solution. and, if not, at least we'll have a better case for the impossibility of it today. that being said, i don't know enough of the technical policies, but i also don't believe there's any such search problem that can't be solved with a bigger index and less updated results. it is, after all, how every web search work in the end. (ps: i'm probably unable to support this idea any further. wished to vote for integration with fossil because of offline search among other things, but got in too late. this is my only contribution for the 2021 wishlist, though. cheers! 😘) --cacawee (talk) 07:55, 13 December 2020 (UTC)[reply]
I am running a website (German, non-profit) covering some 60000 pages on basic maths and physics. A search field takes a query and first looks for exact matches based on file names. The query then looks for similiar matches, mixing two algorithms (Levenshtein and the PHP similar-function). Finally, a list of search results with short preview texts is displayed. The search-engine is self-programmed in PHP, but not at a professional level. One (big) shortcoming is lack of speed. I'd gladly share any ideas and experiences. --Rhetos (talk) 12:51, 14 December 2020 (UTC)[reply]
Oppose If data from Wikidata is incorporated into the search engine protocol, it's going to end up skewing the sequence of the search results and preclude people from quickly and easily locating and accessing existing material on the site. Tyrekecorrea (talk) 18:56, 9 December 2020 (UTC)[reply]
Comment This proposal is less clear. Shall Wikidata search results be used as snippets, like other crosswiki search results? Shall Wikidata be amongst all other Wikipedia (or specific project) search results? Shall Wikidata results be JavaScript pop-ups? How else can Wikidata results be well executed? George Ho (talk) 07:49, 10 December 2020 (UTC)[reply]
Support good thing to keep on the table for brainstorming (assuming it is as complex/"impossible" as some folks say); it would be useful (as an OPTION) Philiptdotcom (talk) 13:45, 14 December 2020 (UTC)[reply]
Support Anything that improves our search is a good idea, and this would be very useful (often more so than exact-text matches, but it depends on what you're doing). Should be able to turn it off, both in the search form and its results pages, and as a permanent setting. — SMcCandlish ☺☏¢ >ʌⱷ҅ᴥⱷʌ< 08:18, 15 December 2020 (UTC)[reply]