Search results improvements

Tags: #<Tag:0x00007f076526a110>


In relation to MB Release

In a search of Releases for Jérusalem:
The new search and the old search both return Releases by Artists named Jerusalem higher in results than Release titles that contain Jérusalem.


Do others expect, in a search of Releases, to have all Release titles that contain Jérusalem to appear before Releases that do not contain Jérusalem in the Release title?

Current and old search both appear to be over-weighting appearance of this search term, or a variant of it, in an irrelevant field.

Current search is returning Releases with Jerusalem in title before Releases with Jérusalem in title.

Search appears to be placing results that match on number of words in Release title higher than results that match spelling (diacritical marks) of Release title.

Is this desired behaviour?


Which is what I eluded to a few posts earlier.
New search is far nicer than the old search which gave hundreds of unrelated results. But there should be a way to loosen how tight the results are.
Instead of 1 exact result with the new search or 100 unrelated results with the old search, maybe 10 results is better - a combination of the 1 exact match and 9 possible matches.


I had some cases where I entered an artist search (from relationship editor) with typos or switched some letters… and get no results at all.
Or for some longer names (with 3 or more parts) I only got a result when nearly the whole name was typed in. Even when it was unique previously.


It would have been easy to limit the number of searches to a more manageable number, it would be rather more difficult to change the way the search worked depending on the slider value. But if you are not getting enough results now there needs to be a significant change to how the new search works so that it can do fuzzier searches.


I’m looking at Picard>Options>Metadata>Preferred Releases - Preferred release types
which has 15 sliders allowing search results to be adjusted to the user’s understanding.

There are times when a very strict search is what is sought.
Exact spelling, exact word number match, exact field.

And other times when a match with a differently spelt alias, with a different number of words, in a non-specified and only loosely related field, would be best.


I am not using picard. I am on the website.
Also, I am not talking about a specific number of results, I just used those numbers as examples.


Separate from the “results could be expanded” issue…

I have noticed that sometimes search switches languages on me.
No, I have not noticed a pattern. It just seems to, at random times, say:
Recherche indexée
Recherche indexée avec la syntaxe de requête approfondie
Recherche directe dans la base de données

Searching again brings it back to english.


Pls excuse my lack of specificity.

Picard>Options>Metadata>Preferred Releases - Preferred release types
which has 15 sliders allowing search results to be adjusted to the user’s understanding
is an example of where a search is tailored by the searcher to get returns closer to what they seek.

The search on Musicbrainz could be set up in a way that somewhat parallels the set-up at Picard>Options>Metadata>Preferred Releases - Preferred release types
This would allow one person to get highly specific returns and the next to get very broad returns.


Searching for “riot”, “Riot” or “RIOT” shows partial matches first (e.g “Riot V”, “Quiet Riot” or “Tha Riot”). I would expect it to display stuff like RIOT, Riot or Riot at the top.


This post was flagged by the community and is temporarily hidden.


I was happy that my new artist were not “available” without doing a direct search, but I may have added duplicate artist because I cannot effectively search as I use to.

A artist with the first name “Richard” could be in the DB as “Rich”, “Rick”, “Ricky”, Richy", “Richard”, other. IMO a search of “R xxxxx” should return all artist with first name starting with “R”. “Robert” is a more interesting first name as it could be “Bob”, Bobby", “Rob”, “Robby”, other.


@ijabz That is fully and entirely incorrect.

@samj1912 has created a ton of tests, got a ton of community feedback and has toiled long hours to deliver this code. If anyone hasn’t done their homework it is you, since you clearly have not followed along with the community nor did you provide timely feedback when he asked for it repeatedly.

Please keep your pointless disparaging remarks to yourself – they are not welcome here.


I did actually look for some test in the code repos but was unable to find any, so if there are some perhaps you post there whereabouts. I never said he hadn’t worked hard on generally done a good job, previously I commended him on his work. I also did give him some feedback on some of the issues found (such as the !!! issue), but there was actually very little time to give feedback between there being a working version on beta and the actual release. My comment was born out of frustration that I had already suggested to him he should write some tests and he failed to reply. I also raised an issue on Solr but he just closed it and told me to move to this thread instead, I cant understand why issues with search don’t get their own SOLR tickets anymore making it much more difficult to manage.


Richard can also be Dick, Robert could be Bert. Other problem names include Alexander, William, Elizabeth, Kathryn (including variant spellings). Name variations seems like a big can of worms.


Agreed. Isn’t that what alias relationships are for? To provide the alternatives for the search engine?


Well, I thought aliases were for names the artist actually uses (including misprints on releases). But if someone is searching for an artist without knowing whether it’s Kathryn or Catherine, etc, that might not help.


Good point. I’m guessing you’re probably right and that is the primary purpose of the alias.


It’s for both! :slight_smile: If I typo an artist name all the time, even if it’s not printed anywhere, I’ll still add it - I assume if I make the mistake others might too.


If we don’t have Rich and Rick and Richard automatically show up in the same results, we end up with things like this edit:

I created John Gunnell after a search yielded no results. But then, over time, I started thinking about how precise Solr is, so I searched using just the last name and found Jon Gunnell already existed. They are the same person, and now they are being merged.
Jon has been added to the alias list to prevent someone from not finding him in the future.


I have been reading the docs and experimenting with searches but cannot figure a way to do a search on first character of first name along with last name “R Clark” and get all Clark’s with with name begining “R”. I suspect there is not the concept of first/last name.

Found 6 results for “R Clark”
100 David R. Clark Clark, David R. Person Male
98 Ian R. Clark Clark, Ian R. Person
67 Rozlyne Clarke Clarke, Rozlyne Person Female
59 Don R. Clarke Clarke, Don R. Person
52 Steve Clarke (Steve R. Clarke) Clarke, Steve Person Male
50 Ray Clarke Clarke, Ray Person

Found 9 results for “Rick Clark”
100 Rick Clark Clark, Rick Person Male
95 Rick Clark (producer/engineer) Clark, Rick Person Male
95 Rick Clark (Brother of Gene Clark) Clark, Rick Person Male
54 Rick Clarke Clarke, Rick Person