Since the latest update, it’s taking me 4-5 times to connect to search. Anyone else having this issue?
Off and on, but when I pressed enter again it seemed to work 90% of the time
Though it has been better again for the last couple of days.
I did not know the data base needed to be updated after installing the latest update. That mostly solved the search problem I had been experiencing. There’s still a minor problem when searching and failing, which seems to be connected to the term remaining in the search box and can be resolved by deleting it
Yes I have come across that a few times and then wonder why it can’t find anything.
I’m not 100% sure there was a database change, but my system took several minutes to become available after restarting and that usually means a database change.
I’ve said it many times before and it still bears repeating:
Search is so inconsistent that it is almost useless. The Roon Team needs to post a few instruction videos which show the users which hand positions, days of week, wind direction, etc. produce useable search results.
I’ll bite. It is terrible. And I can get with your funny characterization.
But… they now have engineers who are 100% focused on search. And they’ve made a massive shift architecturally to enable them to use more modern tools that could not feasibly be run on a core/within a network/on the streaming catalogs (witness the whole “you’re saying I can’t run Roon without internet access” debacle). I sense a major investment, and a substantial change a-coming. Maybe I’ll be wrong; but I’d bet that there’s big search stuff in the pipeline. I have zero inside information - this is pure inference. I’m happy to be patient given my confidence in reading the tea leaves. I’m sure what comes out will be both way better, and will be still somewhat annoying/broken — at least for a while. Such is how software works. I’m wishing Team Roon good luck on what I’m sure is a major change. Modern tools and toolkits are available, and it seems they are getting ready to avail themselves of them. Misspellings and transpositions and stuff like that are no longer that hard to fix; but it needs a focused team treating it like a first class problem, and I think we are beginning to see the outlines of that.
I think you’re being a bit optimistic.
Well… they hired a search specialist, at the end of 2020. Whether he’s an “engineer” remains to be seen. Whether he’s still solely devoted to search remains to be seen.
Brian wrote:
ElasticSearch is a much larger analysis engine with various Web crawler and document parsing pieces, but at the core is Apache Lucene, a system built on top of the idea of inverted indices of the words in the various documents. Here “document” is the text available from the metadata of a track or album or biography. When you input a search term like “revolver”, the search engine will find all the “documents” which include that word. Documents which include it more than once (say, a song with the term “revolver” in the chorus) will typically score higher. Documents with fewer words in them typically score higher (“Ave Maria” beats “I Just Met A Girl Named Maria” for the search term “Maria”). Short words are often discarded, so does “I Just Met A Girl Named Maria” have seven words or five or maybe just four – that would affect the weighting.
There are other issues around weighting. Should words in titles “weigh” more than words in descriptions? Should album titles “weigh” more track titles? How should words in names of performers (John Cougar Mellencamp) compare in weight to words in titles of albums? Some album reviews are shorter than others – should the words in them really have more weight because of that? And so on and so forth.
It’s all very tricky. I’ve built several search engines to support document collections like this on top of Lucene, and there’s a lot of experience you have to go through to get to reasonable results. And in a something like Roon, you’re going to have individual users with different ideas about what constitutes “reasonable”. Ideally, you’d like to characterize each user by what they listen to and what albums they have in their library, and then use that characterization to change the weights, not only on particular pieces of metadata, but on individual words! For instance, you could create a “word cloud” from the words in the user’s metadata, then use that to calculate a cosine similarity score to the words in each of the hits from the inverted index search, to change the rankings of the hits. But would that be the right thing to do?
And of course different people in different parts of the world use different words, even with the same language: “boot” in the UK versus “trunk” in the US, “lift” versus “elevator”, etc. So you really should have region-specific vocabularies so that you know what the words the user uses actually mean.
Brian’s right: search is a forever problem.
100% true, but “a journey of a thousand miles” and all that, and up until now I’d say that they hadn’t taken the first step (without Lucene or elastic or something analogous it’s really hard to do anything meaningful). It’s not purely tech, it’s rules and datasets and how you use them.
As you rightly point out. But without the right underlying tracks you can’t start to build rules or datasets. I’m an optimist, even if it is a forever problem.
Bear in mind that this has been going on for the better part of two years now, according to Brian’s post.
That’s right. Almost none of it is tech, as Lucene provides most of what you need. Using the tech in the right way is the hard part.
For what it’s worth with Revoler original CD, 24 bit, 2008 remaster 2022 remix and Mono versions in my library I get
And
It’s hard to fathom how these results occur, hence if I want revolver by the Beatles I focus on albums in my library.
.sjb
Sonos manages it as a global search fairly easily across multiple services. I just don’t get why Roon finds it so hard enough of excuses. It’s top result under all of them.
Am i the only one seeing this latest post as unread by the forum software? Doesn’t matter how many times i have a looksee here…
No it has been there since last night.
I was going to post and delete it if it lasted much longer, so thanks Mikael
Your post has caused it to become read
Nope, not just you.
It isn’t just this thread. It’s been happening to other posts, seemingly without a pattern. As soon as another post hits beyond the post you’re stuck on, then it clears up.
I’ve noticed this thought it was just my discourse app being funny.
So Search has a hard time coming up with Björk. She’s buried abut 30 artists deep. Even if I favorite her it makes no difference. She moved up to about 22nd if you search with Björk. On the Qobuz app I just type Bj and instantly she appears at top as most popular artist. Shouldn’t room take most popular into consideration? This terrible search is probably my number one reason for wanting to use native streaming apps. Really needs to be fixed.
So now it’s coming up 1st on my end too. Just such strange behavior. I tried like 5 times before all with those same terrible results.
Edit: now she’s no longer 1st.