As I said, you can use https://test.musicbrainz.org to check old results.
2 Likes
Yeah I just noticed that, thanks a lot!
I have criticised the old search. (And separately, by implication, the new as well.)
I am sure that with the resources available you did a good job.
And that you deserve no criticism but instead thanks for your volunteer work.
1 Like
This quote I find the most worrying on this whole thread.
Google search very frequently does a terrible job of returning the sorts of results I seek.
The old MB search was far superior to Google search for my purposes.
It rarely denied the existence of things in the database.
It usually returned somewhat consistent results.
Google search fails at even these basics.
I’d really like MB’s search to continue to be much better than Google’s.
1 Like
Riot V has a Riot alias, and most of their releases are as Riot (name changed a few years ago). This seems reasonable to me: they’re probably one of the biggest bands (formerly) named Riot, so most users are likely to be looking for it.
I agree exact matches shouldn’t be buried, but the popularity filter also allows, say, a search for “Tchaikovsky” to find the actual artist people are looking for instead of all the others. Now that search is not as good as it was at the first release of SOLR. If the two options happen to be too conflicting, maybe we can add an extra option for “search exactly for these terms”? (although I guess really that’s what direct search is…)
5 Likes
The thing is that Tchaikovsky (the composer) is an entity consisting of multiple words, while my only concern so far is one word entities exactly matching the search string not being on top.
Not as I remember.
The aliases and artist credits or something like that did already make fuller names show first like in your example.
1 Like
Then your concern seems to be in good hands:
I haven’t really felt regressions in my one word artist searches compared with old search.
Maybe show with test URL (old search) when you find regression next time.
But new search should not mimic old search because old search was not really as good as new search IMO.
3 Likes
So I retried https://musicbrainz.org/search?query=fred&type=artist&method=indexed and that looks good to me since has all the 'Fred’s first, with the exception of ‘Fred Astaire’ who is very well known so should be at near the top. So my question is why does ‘Fred Astaire’ get to top of list but a search for Tchaikovsky does not. Is it because Tchaikovsky is an alias ?
How does the popularity boost work, seeing the huge number of releases attributed to Tchaikovsky (or Bach) I would have guessed that should allow them to top the list without boosting lesser artists in above exact matches. The thing we did in old search was artificially boost particular artists (such as the great classical composers), although this wasn’t very flexible it was good for fixing issues with particular artists.
This is the old search for Riot using test.musicbrainz.org:
Also it’s interesting to see 4688 results in the old search vs 625 results in the current search. The user would probably not look into more than a couple of pages anyway but it would be interesting to know why so many results are missing.
https://test.musicbrainz.org/search?query=Riot&type=artist&method=indexed
It is one of the improvements of the new search.
Old search showed too many results, to my taste at least, so I was always using the direct search instead.
I have to agree with that, it feels a little weird though, because sometimes you get so much *fewer* results than in before, so the first instinct is “this is wrong” - but I think making things a tiny *bit* more fuzzy like Sambav already has will fix any actual issues.
I also wanted to say, that I also have tested this for a while now - and that there are so many, many improvements, some examples:
https://beta.musicbrainz.org/search?query=London&type=area&method=indexed
https://beta.musicbrainz.org/search?query=atlanta&type=area&method=indexed
https://beta.musicbrainz.org/search?query=paris&type=area&method=indexed
Search results - MusicBrainz (showing everything relevant before “violin family”)
Search results - MusicBrainz
(though the Search results - MusicBrainz now has a tiny issue 
)
So in general I think this very much fixes those issues we had before with non-relevant places showing up, (which was a huge personal peeve for me tbh) and non-relevant instruments (family) not being put in the first results.
3 Likes
Yeah, I like the new search a lot and mad props to Sambhav for all the work he has done on it. This is the first issue that I found so far. Additionally I imagine a lot of editors (new or inexperienced) who won’t even know that they are supposed to use quotes for single word artists to get more relevant results and would just add duplicate artists.
Fixed the area results 
5 Likes
Is there a way to exclude “sort name” or only use “sort name”?
1 Like
I have recently added an improvement to allow searching by popularity and relevance. Similar to amazon searches
This should allow everyone to customize search as per their needs.
3 Likes
Yeah, you can do something like
sortname:querystring
It only works in advanced mode.
Cant you adjust boost so only very popular artists get significant boost so that well known artists to get to the top but in general relevance matches has much greater effect than boost, can you point me to where boosting is configured in the code ?
This is the artist boost config
1 Like
Maybe it would work better if there was only a boost when ref_count was greater than a particular figure ?
Not sure exactly what ref_counts are but lets just assume they relate to number of releases (not release groups) whey they have release artist credit , I wouldn’t want to see a boost for an artist with 5 release over one with only one, but I would like to see a boost for an artist with at least 50