Skip to main content ↑ ↓

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: Mood-tagging and why it won't work (Read 9983 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Mood-tagging and why it won't work

The topic pops up again and again - describing the "mood" of music. The "demand" appears to be there. At least software-technologically however, implementations aren't available. I also think i know why they aren't available. Its not mainly because of lack of interest, but instead because implementers as well as users lack the abilities to make it work efficiently. The main problem with that is the user-side.... it just needs one gifted implementer to create the tech - but every single user needs to be able to understand and use it.

To avoid various misconceptions beforehand, some definitions first...

With mood-tagging i mean describing the mood-relevant "properties" and "characteristics" of music. I do NOT mean "listening situations" like i.e. "chill out / relaxing" or "party". Neither do i mean singular classifications like "trippy". What i mean with mood-tagging is identifying the "components" of which moods consist and then rating tracks according to those components. In other words, i'm not talking about the dish but about the ingredients.

To consider such a system to "work efficiently" i demand the following requirements to be fulfilled: A user with thousands of tracks across many different genres, must be able to rate those tracks without too much effort. Afterwards, he must be able to build playlists just by looking at mood and genre tags, and get consistent results (low amount of "unsuited tracks").

The problem now is quite simply, that you need a very introspective, self-understanding and musically experienced user to rate tracks efficiently and consistently regarding their mood. Even something as simple as "energy".... so, if a track is more relaxing or stimulating... would overstrain the ability of most users, because it is not simply a matter of tempo, and it requires quite a bit of understanding if you want to get it right consistent across genres. In the case of "energy", most users would probably be able to consistently rate on a 1-3 scale, but not more. With other more psychological aspects like "uplifting" and "depressive" it gets way more complicated. To get to the point: most users are not able to consistently mood-rate thousands of tracks across many genres - mood-rating music has way more to do with consistently understanding ones own feelings and perception, than it has to do with rating songs.


However, that this most probably wouldn't work for most people, isn't even a problem. Why? Well, because most people never actually wanted mood-tagging. Mood tagging isn't a solution - it is just a means. A means for what? Well, a means to easily find suitable music for certain OCCASIONS (listening situations). This what most people actually want. But unless you are a VERY varied person, you do NOT need mood tagging for easily finding suitable music for a certain occasion. All you need for that is a fieldname like i.e. "occasion" and a fixed amount of values like "dance", "relaxing", "party", "lunch", "work" etc. - then just apply those labels consistently across your music, et voila, you can now easily select matching music for your favorite listening situations - especially if you combine it with genres.

- Lyx
I am arrogant and I can afford it because I deliver.

Mood-tagging and why it won't work

Reply #1 –
Mood-tagging and why it is working for me


Quote
What i mean with mood-tagging is identifying the "components" of which moods consist and then rating tracks according to those components

At the moment I'm tagging a lot of my tracks with attributes of which the "mood" of the songs consists (I've tagged more then 1000 of my ~5000 tracks now, and its not really a lot of work, I'm just doing it while listening, the same as I simply add ratings for the tracks).
I use only three "components" the first is for Happy/Unhappy feeling the second for Fast/Medium/Slow and the last one I call "blurred" (its the different feeling which is created through "clean" electronic  or indie/lofi/punk sound, I can't discribe that easily).

Quote
To consider such a system to "work efficiently" i demand the following requirements to be fulfilled: A user with thousands of tracks across many different genres, must be able to rate those tracks without too much effort. Afterwards, he must be able to build playlists just by looking at mood and genre tags, and get consistent results (low amount of "unsuited tracks").

As said, for me it's not that much effort.
I can create Playlists very easily and fast with facets (I used browser before). I just select "sad" and "slow" and get a very melancholic playlist (which I sort with some big coustom sort string that gets the tracks on the top which I rated high, played a lot and the last time played it is some time ago).

I planed to use more than my 3 attributes, but I noticed that the ones I use are really enough for me, and I get playlist I really want to listen to without the need of skipping more then 1 of ten songs.

btw: I use different colors for displaying the "mood" of the tracks in my playlist. Their was some thread about something like this somewhere in the forum. I really love this, you can see very fast what mood a track has (or the full album/playlist consists of).


I never liked having playlist or tags for listening situations. I don't always listen to the same music if I'm doing the same thing (I could relax to slow-trip-hop or metal, too for example). But I always now what the music should "feel"/sound like when I want to listen and the system I use is mush better for me than using genres, occasions or whatever.
Of course many time I select tracks or albums myself and listening to them, but sometimes I don't now exact what artist I want to listen to, and so it's very nice to get a mix of tracks "automatic".



Another thing:
I'm planning/trying to create some online database for tags like this. I want it to be like this: everybody can submit his values for things like tempo and other attributes. Then you could get the rounded infos of all users for every track in the database and use it for tagging your own very easy (I have a lot of other usings in mind...).
I will something/more about this later in some days/weeks.

Mood-tagging and why it won't work

Reply #2 –
From your descriptions, it seems that you use a 1-3 rating scale for the individual components. I have no doubt that most people can handle a 1-3 scale.... but i doubt that more than a handful of people are able to consistently use more fine-grained scales.
I am arrogant and I can afford it because I deliver.

Mood-tagging and why it won't work

Reply #3 –
I think the question is, if you really need a bigger rating scale. I don't think that you're suddenly thinking: "hey lets listen to music now that is 20-30% happy, 70-90% energetic and a minimum of 56,5 km/h  fast)
No, you will just be thinking "I need some hard music now, and I'm unhappy with everything", or something like that.

So unless you want to do something like getting the most similar tracks on the hole world for some special track (and last.fm, pandora etc are working great for that), I don't think that you will need a bigger range than 1-5 and not a lot of "components".

Mood-tagging and why it won't work

Reply #4 –
Mood-tagging have limited appeal to me, but I think tagging UI (not just mood) can be very useful in fb2k.

Example: When i'm relaxing, I might choose to listen by mood, but when I'm writing papers, I'd like to listen to track that're all instrumental and energetic and from japanese rpg soundtrack.

The solution now is obviously manually add track to the appropriate playlist, but it's hardly a constant. The process of add/remove tracks from playlists gets harder and harder overtime as your amount of playlists / tracks grows. And god bless you if you rename or make directory structure changes outside fb2k.

I know ID3v2.4 support multi-value fields, so even the genre tag can be used to tag whatever I like (well okay maybe not such a good idea, but still), But some sort of UI tailor toward tagging would be useful, by mood or not.

Quote from: Lyx on
To get to the point: most users are not able to consistently mood-rate thousands of tracks across many genres - mood-rating music has way more to do with consistently understanding ones own feelings and perception, than it has to do with rating songs.

I don't think users need to "get it right", if there's an easy mean to make changes (on UI), users can correct them incrementally over time; Besides I don't think it's relevant how/what a user tags. As you said, it's just a mean, so long he/she tags it and the tags generate the playlist he/she wants, the only barrier is the ease to do it.

Quote from: Birk on
So unless you want to do something like getting the most similar tracks on the hole world for some special track (and last.fm, pandora etc are working great for that), I don't think that you will need a bigger range than 1-5 and not a lot of "components".

I don't think you need a set scale; I don't need a scale personally, but if I really want one, no one will stop me from tagging my tracks happy-01 and happy-50. Preference for tagging varies person to person, I don't think you're going to find a scale that can make everyone happy.

Mood-tagging and why it won't work

Reply #5 –
Wouldn't a solution be something similar to MoodLogic (just open and free) with guides on how to tag properly? If a user is presented a well-explained guide, they can easily tag music.

This should be collected in a online database, and multiple submissions could be evaluated to define the accurateness. Also you could present the user for several levels of analysing (say, a base level for defining the mood and extended levels for describing instruments, key, etc.)

I definately think this is possible, and already exists in some forms (Pandora, MoodLogic and the thing distributed along Winamp, which name I forgot - They are just propetriary).
Can't wait for a HD-AAC encoder :P

Mood-tagging and why it won't work

Reply #6 –
Quote from: odyssey on
This should be collected in a online database, and multiple submissions could be evaluated to define the accurateness. Also you could present the user for several levels of analysing (say, a base level for defining the mood and extended levels for describing instruments, key, etc.)


Maybe I'm misunderstanding the nature and scope of mood-tagging, but wouldn't something like this be far too subjective for an online database? (The mood component, at least, not stuff like the key and tempo and such.)

Mood-tagging and why it won't work

Reply #7 –
Quote from: DocBeard on
Quote from: odyssey on

This should be collected in a online database, and multiple submissions could be evaluated to define the accurateness. Also you could present the user for several levels of analysing (say, a base level for defining the mood and extended levels for describing instruments, key, etc.)


Maybe I'm misunderstanding the nature and scope of mood-tagging, but wouldn't something like this be far too subjective for an online database? (The mood component, at least, not stuff like the key and tempo and such.)

Why? The moods of a song is just like wine. People may have different views on the taste, but the overall taste would still consist of the same elements. That's why I say, that different input could be evaluated to describe an overall mood for each song.

Addendum: The only thing I see that won't really fit into this is rating, since people would never agree on this.
Can't wait for a HD-AAC encoder :P

Mood-tagging and why it won't work

Reply #8 –
Except wine can't be ironic. Or fatalistic. or any other complex human emotion.

e.g.
How are 90% of Montreal songs going to get rated?  Half the people are going to rate it "very happy" and the other half are going to actually read lyrics and rate it extremely depressing. The end result will be a compromised hunk of worthlessness.
MoodLogic was abandon for a reason.



Now, that said, I wouldn't mind a free repository of key, time signature and BPM info. That could be helpful and interesting.
elevatorladylevitateme
 

Mood-tagging and why it won't work

Reply #9 –
i'm curious.. what about http://en.wikipedia.org/wiki/Moodbar
(plugin for amarok)?
i don't run linux, so i can't really test it beyond looking at the screenshot and conclude it does something, but it seems at least moderately interesting (or at least the bars look shiny?)

Mood-tagging and why it won't work

Reply #10 –
Quote from: boombaard on
i'm curious.. what about http://en.wikipedia.org/wiki/Moodbar
(plugin for amarok)?
i don't run linux, so i can't really test it beyond looking at the screenshot and conclude it does something, but it seems at least moderately interesting (or at least the bars look shiny?)

It looks awesome. I have had this thought myself. I know you can't compute EVERYTHING, but you get a long way with something like this.
Can't wait for a HD-AAC encoder :P