I am testing fuzzy search in 1.8 and here are a few observations. I don't know exactly what should work and what is too much to ask:

Wonderful:
"japanese Buildings construction" correctly finds "Building construction in Japan"

Room for improvement:
"Buiding construction" does not find "Building construction"
"Building construction in japa" does not find "Building construction in Japan"
"Building construction of Japan" does not find "Building construction in Japan". It finds "Construction in Japan" and "Buildings in Japan" though. Removing "of" leads to the correct result.

Thanks for the feedback @Nicolas_Raoul !

Yeah Method A does have some drawbacks. When I tested it on our samples it did not find some of the phrases that we guessed also, it just performed better than Method B.

I have looked around a bit previously and haven't found any other viable MediaWiki-based APIs for this purpose. I think I could look into more generic options for approximate string matching (basic algorithms for ignoring words like in/of etc), but I don't know what can or can't be done before it's time to wrap up. Maybe I could just give it a shot and we'll see how it goes? But should I prioritize getting the 'repeat categories' bug in Phase 1 fixed? And how about the Optional task?

What do you think? (cc: @Niedzielski)

Fixing 'repeat categories' is higher priority I would say.

Then manual testing will tell us what cases could be enhanced with tricks like ignoring in/of (capitalization and pluralization seem unnecessary now).

Sorry I forgot what is the Optional task haha

@Nicolas_Raoul okay, I'll work on fixing the repeat categories bug first then.

The optional task is T119288. :)