Ratings calculation

MikeEnRegalia wrote: ^ So you think that with two 5 star ratings the album should have a "score" of 5.0 and be listed above long standing albums with 4.x averages?

That's the key point! Leave the algorithm for ratings as it was (it took into account who gave the review and how "long" it was).
But when you need to list the albums anywhere... It is this place where to use an algorithm which takes the number of reviews into account as well.

So the ratings would remain in a state people got used to already (when a collab rate 5, and a noncollab rate 4 the average used to be equal to 4.60 or near - I guess, it was Ok for most visitors), and least reviewed albums won't pop up on tops of rating-lists or something.

What's done now is going to make most popular albums even more popular and the least popular albums would receive even less attention. Visitors will just look at the rating (which is now modified by a number-of-ratings-dependent algorithm, but how a visitor would now THAT??) and see it is 3.89 or so - would they be willing to get it?? It doesn't attract even to read the reviews!

Hope, you got my point of view: the current algorithm scheme is not that evident to an average visitor (it just isn't natural) and surely would make some potentially good albums unattractive.

Edited by XPEHOPE3KA - July 15 2007 at 11:09

MikeEnRegalia wrote: ^ So you think that with two 5 star ratings the album should have a "score" of 5.0 and be listed above long standing albums with 4.x averages?

Yes, and no. It isn't listed above them, at least if there's tow ways to list them and users could choice the way. And the previous system worked well.

And there's always top 100 (or 500) lists. There are lists for genres also. Why not for artists? It is only calculating, not taking too much resources.

It is very odd that you must look into each album's ratings trying to calculate yourself the absolute  average.  And who says reviewers' ratings are better? Or that the album is better when there's more ratings. I think proggers are quite intelligent people and because there's so few ratings in many albums, we should take them seriously and not "banning" non-reviewers or single voices.

And whatabout situation when there's album with three 4-star votes, one with two five-star votes? And Im still not sure if it is a weighting factor whether you are reviewer or not.... obscure. Very obscure things.

Can't go on like this.

Dying.

It makes my life a lot more difficult!!

Though, i love and hate everything at the same time...

http://www.progarchives.com/album.asp?id=13410 there's 3,77 for example, with two 5-stars.

Maybe we should vote about it. Why are rating averages changed??

This is f*cked system, cannot work this way,,, and people fixing some random ratings...

weighted rating averages can remain, but only if we take the absolute average with it.

People can check themselves who many ratings some albums have and whether a Reviewer has rated it.

We can't go with the popular system because so many have only few ratings.

v v


And where's the function who all is being calculated??

I think there could be many was to calculate things (for example three) but they should be shown clearly.
And it would be nice if they were in tables that can be modified,, i mean, that the list could be ordered in many ways, for example by year or rating


Edited by progressive - July 13 2007 at 11:06

A major recalculations was made in the past hours, the Average Rating of a particular album is still the same though. And therefore should have fix the 1 rating issues and others , please get back to me ...

Now, the difference is in the TOP 100 list ...

Before : Log N order by
Now: Weighted Avg, comparing an album to all others album in the site

The WR of an album is calculated like this:

I weight the average number of votes and avg rating of albums against the average rating of a specific album.

Every time a review/rating is posted I recalculate the WR for all the albums.

Like this:

    SELECT @avg_num_votes = AVG(n_ratings) FROM CD
    SELECT @avg_ratings    = AVG(CONVERT(FLOAT,rate)) FROM Rating
   
    UPDATE CD SET weighted_ratings = ((@avg_num_votes * @avg_ratings) + (n_ratings * avg_ratings))/(@avg_num_votes+n_ratings)


and the weighted_ratings is used for the position in the TOP 100.

What do you think ?
Mike ?


Max


Edited by M@X - July 12 2007 at 07:23

avestin wrote: Mandrakeroot wrote: MikeEnRegalia wrote: weighted average does not explain these anomalies ... either there are mistakes in the calculation, or M@x is doing something completely different. Either way, I'd be happy to take a look and maybe help him to find a solution. If one album have zero ratings... The average is zero. Instead they come appraised also those albums.  For me a serious anomaly!!!     Anomaly, yes... or a fine example of Artificial inteligence... Although judging by some of the albums that got a rating, an inteligence with not so good taste in music....

Well... Is this the Artificial intelligence of the "2002 Space Odissey"'s computer, maybe?

MikeEnRegalia wrote: weighted average does not explain these anomalies ... either there are mistakes in the calculation, or M@x is doing something completely different. Either way, I'd be happy to take a look and maybe help him to find a solution.

If one album have zero ratings... The average is zero. Instead they come appraised also those albums.  For me a serious anomaly!!!

weighted average does not explain these anomalies ... either there are mistakes in the calculation, or M@x is doing something completely different. Either way, I'd be happy to take a look and maybe help him to find a solution.

It's more than one album.  1309 albums in the top 4000 all-genres alone by my count last night. As I wrote in that other thread:

I wonder why do all the unrated albums in the top 4000 now have a value of 3.86 (by my fatigued checking)?  I have a theory, but...

I don't find that useful.  Taken from the top 4000 most popular albums (CLICK): Every non-rated album in the top 4000 is given a rating of 3.86 and they all fall between 1224 and 2533 (they are surrounded by reviewed albums).  Between 2882 and 3286 all the albums have 2 ratings and are given 3.83. Between 2881 and 3153, the first albums only rated once are found, and are given 3.84. And at 3379 to 4000 are all the other albums given one star - at 3.83.  All of the albums only rated three times are given a score of 3.85, 3.84, or 3.83.

The album at 4000 with one review/rating of 4 by Sean Trane which gets a list rating of 3.83.  I see looking at the Kevin Ayers page that the album ratings do vary slightly between the albums only rated once... from 3.79 to 3.83....

Rivertree wrote: Easy Livin wrote: There has been a fundamental change to the algorithm used. See the thread in the Prog Music lounge. I can't find any thread in the Prog Music Lounge dealing with this problem