[09:20:14] dcausse: I've sent you a meeting invite for Semantic Search, I'm pretty sure I'll need your help! Later this afternoon. [09:20:44] gehel: sure, I should be around [09:20:57] thx [10:42:51] errand+lunch [14:17:43] \o [14:45:30] o/ [15:01:12] ebernhardson, dcausse : I've forwarded an invite to a semantic search meeting later today. I'm alone with the kids, I can't be there. If one of you could join, that would be great! [15:02:59] if Erik can't make it, I can try to attend [15:04:08] \o. I can make it [15:08:27] thanks! [15:36:25] .o/ [16:55:33] sigh...i let claude workout getting semantic highlighting into the browser suite overnight. it claims to have worked, but it's 700 lines of code (and somehow 200 lines of git commit messages). it might work, but it's going to take a moment to ingest :S [16:57:41] Trey314159: quick demo at https://people.wikimedia.org/~dcausse/english_frames.html, I have the impression that something's not right because with the scoring & fallback logic, some queries (e.g. "what is a cat" or "tom age") are completely different (dwim does not necessarily complete on top of the other but rather replace some) [17:29:16] heading out [17:30:04] latersville! [17:35:35] dcausse: thanks for the demo. I will take a look! (I was in the staff meeting) [19:35:09] dcausse: I think I've found the problem. Using cirrusDumpResult & looking at the scores, the query "who is tom" is a mediocre match to "who is america", while the extracted query "tom" is a great match for "tom cruise"... the difference is so great that the discount of 0.5 is not enough to overcome it. (In my wacky local wiki document collection, even 0.1 was not low enough.) [19:35:11] If there's no problem with lower weights, changing the weight to 0.05 or even 0.01 seems like it should solve the problem. I guess it could also potentially be a problem for the other profiles, too, if the original match can be very weak and the transformed match very strong. Any objections to a 0.01 weight? [19:58:05] sigh CI build fails because...the snapshot.debian.org cache is split :S CI sees a 404, i see a 302. same url, but dns resolves to a different ip in eqiad [20:07:09] that's unfortunate. Are those runners in WMF somewhere or Digital Ocean? [20:08:55] inflatador: it says DO, i guess it's not eqiad then :P but same idea, the apt log prints the IP it resolved and that one gets a 404 still [20:12:25] No worries, I was thinking if it was in WMF somewhere I could try jiggling the DNS cache [20:21:19] oh interesting, that ip is giving a timeout now. retrying the CI works again.