The point of this post is to understand whether averaging track ratings gives a valid indication to how good an album was preceived to be.

I find album ratings are quite useful when having to quickly pick "good" albums, possibly of varied genres, from a large library.

I realise this is a very subjective topic, but the idea of ratings is to differentiate tracks by degrees of user preference. So if the track ratings are reliable, it might extend right up to the album level too. Sure there will be exceptions but they tend to be in the minority in my experience.


Having used averaged ratings for some months now, i'm beginning to have doubts about it.

example
take 2 albums, A & B of similar genres.
A has 10 tracks, of which 2 tracks are rated at 4 stars and the rest at 3 stars.
B has 15 tracks with the same ratings as A.

The avg album scores for A & B are 3.2 & 3.1 (rounded)

Can we say that A is better than B given it has a higher score ?

Seems so ... but i think they are equal in rank.

Averaging gives albums with fewer tracks, a higher score than larger albums with just as many good tracks.

i tried to experiment with another method instead of averaging.

Give each album a base score of 3.0, i usually give all tracks a base score of 3 stars and increase/decrease as i listen.

4 star tracks, add 0.1 ,
5 star tracks,add 0.2
2 stars take away 0.1
1 star take away 0.2

from the album base score of 3.0

Using the above it possible to get album ratings based on how many good or bad tracks they have regardless of how many 3 star tracks they might have.

So in the example above both albums would have the same album rating of 3.2.

I'm thinking this produces a more realistic score as listening to an album is cummulative ie (nice / ok / not so nice) as opposed to averaging.

It also gives you ranks in discrete intervals of 0.1, so given a score were 3.4, one could "roughly" tell it has 2 (5 star tracks), 4 (4 star tracks) or a combination of both. An averaged album rating is less obvious.

if you got a 20 track CD and 11 of them were 5 stars, the album rating would be 5 stars period. Or any other combination that resulted in an album score > 5. The same would apply in the other direction towards 0.

I want to clarify here that these ratings would work on a CD level, so in the case of multi-CD albums, each CD would have its own rating as opposed to averaging the ratings of all the CDs in the album.

What do people think about this ?

Yeah I've thought the same thing.

So I just added an albumrating to quicktag, and voila, works excactly the same as with ordinary rating, though you do have to give all the tracks the same tag.

If there was an option to add the albumrating to the database, you'd only have to update the albumrating one time, instead of for each track, even though that's pretty easy in itself, I could imagine people would want to keep it in the database.

If all the tracks in the album have the same rating then fine...but i find lots of albums where this is not the case.

You misunderstand.

I added %albumrating% to all songs on an album with the same rating, but each individual song has it's own normal %rating% rating

I don't understand what you have against using average ratings. You said that the problem occurs when a 5-track album is rated all 5s, and a 20-track album with 5 5-star tracks is compared...the 5-track album comes out with a higher rating. Well, shouldn't it? It's more consistently good...this seems to me to be a textbook example of when averages should be applied.

Also, how do you plan on applying this? It might be useful if it could be automated, but to do this you'll definitely be doing some math in your head, and with a calculator if you're doing averages; and without any automation whatsoever, as long as you have to do this manually and work out the numbers yourself, wouldn't you rather use that time to rate an album based on your "gut" feeling about that album? (a gut feeling you should have if you've already rated all the songs)

Seems to me that if it were automated, you might be surprised at the results. Say an album is full of four-star songs, but you don't really like the album as a whole, and are surprised to see what a high rating it got. This is a result I'd be willing to acccept if they were due to the fact that they're automated and require no intervention from me. In that case, ease wins out over accuracy. So, again, as long as human intervention is required, why not just rate the album how you feel it should be rated?

Also, how can this possibly be implemented once ratings are applied? You can use the extended playlist generator to create a playlist full of songs whose %ALBUMRATING% tag is higher than 4, but...this will include the 1-star songs on that album as well.

Plus, what if your favorite album on all of planet earth only contains 3 tracks? 1 track?? These albums will have albumratings of 3.6 and 3.2...not very high. So, again, the averages issue...what gives?

I like the idea of a dynamic album rating.

In fact, I already did this with the playlist tree plugin. I used the average track rating method. The plugin allows the use of tagz script in the display of each node (or album). The only problem I saw was it could not sort based on the dynamic album ratings because the plugin processed the tagz script after the list had been output.

Without a way to resort after the tagz script was rendered, the entire idea seemed less appealing.

That's a good point

You are saying that a 5 track album of 5 stars would be more enjoyable than the 20 track album of only 5 (5 star) tracks. An average would give a consistant 5, whereas a cummulative method as i described would give a 4 (5x 0.2 + 3.0).

I think that pretty much sums it up.




Certainly, I find a lot of albums that get lower average scores because they had many tracks are all of a sudden getting higher scores. The same can be said of albums that had many 2 star tracks got pushed below.

A Cummulative album rating method appears to increase the difference whereas an averaging method reduces it.

But i think you made your point, given an album, averaging is immune to the number of tracks it has.

I used to average (rounded) to 2 decimal places, but i think it can be done to just one rounded. If the second digit is more than 5. ie 2.96 then the avg score is 3.