What happened to TOP 100???

Sofagrisen wrote: MikeEnRegalia wrote: Do you keep daily snapshots of the chart? I don't and I don't think many other people do, so I guess you're alone on this one. What saddens me though is that you continue to say things that aren't true, or at least aren't based on facts - but you make it look like they are. No, sorry, I don't keep snapshots. I have followed the development in the ratings of Fear of a Blank Planet fairly closely though, which makes it possible for me to tell you the development in its score, only from memory though. :( (I have been thinking about studying ratings development more closely though.) The thing is you can see it on all albums. As they get more votes, their score become lower. The more votes they get, the less significant the effect becomes though. When Riverside releases their new album we will see an excellent example of it. If it is a very good album, as I hope, one might see a development in average score like this: 1 vote: 5 10 votes: 4,85 50 votes: 4,55 100 votes: 4,40 200 votes: 4,30 Basically you can see the same development in all albums. To me it’s not about quality (ratings) vs. popularity. It is about predicting quality, rather. If it was true ratings were stable and that when an album received more votes its ratings would basically be the same, then yes, the number of ratings should be taken out of the equation. But because basically all albums get lower ratings with more votes, the main concern of an algorithm should be take this fairly into account. This is by the way the kind of things one should have studied before one makes an algorithm, the way ratings usually behave. I am sorry if you people haven’t seen this effect, but it is very much real, and you will see it soon enough in the next new high scoring album with many votes.

Long post, short answer: Of course the problem you describe is not related to the ranking algorithm at all. It's simply because the higher a prog metal album (or modern prog rock album) climbs in the top 100, the more negative "hateboy" reviews/ratings it receives. That's because initially those who don't like prog metal or modern prog aren't aware of them album, but the more prominently it is featured in charts, the more attention is drawn towards it, and the more the prog "hardliners" try to push these albums down the chart.

^ I just finished reading all of it...let's see how buried this post is when I get home

^ true. What irritated me about your previous posts was that you seemed to blame the algorithm ... it's obvious that the biggest problem for the accuracy of charts is not the algorithm, but abusive ratings.

This top 100 is impressively laughable! certainely due to "abusive ratings"

Edited by philippe - July 13 2007 at 09:10

Just keep working them angels Rich, you'll get there.

How much more weighted are the collab reviews? I agree there should be a difference between reviews and just ratings but if there must be a gap between collab and non, it should be very small.

andu wrote: Sofagrisen wrote: Albums like The Sky Moves Sideways, Blackwater Park, Remedy Lane vs albums like The Silent Corner And The Empty Stage, Elegant Gypsy, Cantofabule (Cantafabule), Be Live, Essere O Non Essere?... I happen to like the latter albums a lot more, and to find them more significant to prog-rock, despite being a lot less known, thus the new algorithm making some justice (btw I like the new algorithm). If there are others who think like this - and I have reasons to believe so, then I think you should reconsider your stand.

It's not about what one personally like more, it is about numbers.

Let's say album A has 50 votes and a rating of 4,5 and that album B have 150 votes and a rating of 4,4. Which should be higher listed? The way I think when approaching this, is that I ask: If they had equally many votes, which album would have the better rating? In this case, if album A had 150 votes, its rating would probably be lower than the rating of album B, therefore album B should be higher listed. I must base this conclusion upon earlier albums in the same range. What is the average 150 votes rating of an album which had a rating of 4,5 at 50 votes? My guess is something like 4,3, which is based upon casual observation.

Edited by Sofagrisen - July 13 2007 at 09:46

laplace wrote: I agree with Sofagrisen; I would also like the top 100 chart to represent possible rating outcomes in parallel dimensions.

You show such incredible understanding of my argument, that was exactly what I meant. Also we should take into account what possible alien civilisations will probably like. This might be done by observing which music is being preferred by aliens in science-fiction movies. Star Wars or Star Trek would be good places to start collecting empirical evidence.

Edited by Sofagrisen - July 13 2007 at 10:30

andu wrote: I agree with you but I find this a matter of nuance, so to respond to your question, I find nothing wrong in the algorithm which favors the A album (for the reasons I mentiooned in my previous post). The point is that by bringing the A album to attention, it will get more  ratings, and unless everybody rates it *****, it will decrease in general rating (just like you explain it) and will sink to the place it (supposedly) deserves. By this we all gain - both you who think the B album should get a top spot and I who want to promote the A album as worthy. Of course this does not work if we're all biased&frustrated over top positions. I think this is the point of the new algorithm and that you should not be worried, as it has included it's own correction mechanism.

I am sorry, but that correction mechanism will only work to a very moderate degree.

We have an album A and B, again. :P Both albums have 500 votes. You know two facts about the score of them. The rating of album A is 4,5. In the case of album B you only know its rating was 4,5 when it had 50 votes. What do you guess the rating of album B is now? My guess would be about 4,15.

The point is, this show how important the amount of votes is for the rating of an album, and that it must be taken very much into consideration. There is simply a vast difference between a 4,5 album and a 4,15 album, and there is absolutely no doubt which album should be ranked higher…

I don’t think you would be in doubt either. But the argument has nothing to do with popularity. It has all to do with the average behaviour of ratings. If I had everything my way I would like to see a unique regression analysis on every album, actually, but I realize that is a bit too much to ask. :P

Edited by Sofagrisen - July 13 2007 at 11:11

greenback wrote: there are many possible algorithms that would give good results.   coming from my imagination:   taking in consideration:   partial_album_score1 = f(number_of_ratings) where f could be a cubic root curve or a log function   partial_album_score2 = g(avg_rating_of_album) where g could be the famous weighted average discussed in this thread.   Thus, the total score of an album could be expressed as:   total_score = log(f)*g^3  =  log(number_of_ratings+1)*weighted_avg_rating^3   the last thing would simply consist in making a top 100 only according to this "total_score" value.   i'm curious how the output (top 100) would look like!

It looks like a very good algorithm to me, far better than the one we have now. I have tested it some. Was this the old one? :P

Edited by Sofagrisen - July 13 2007 at 12:21

Sofagrisen wrote: greenback wrote: there are many possible algorithms that would give good results.   coming from my imagination:   taking in consideration:   partial_album_score1 = f(number_of_ratings) where f could be a cubic root curve or a log function   partial_album_score2 = g(avg_rating_of_album) where g could be the famous weighted average discussed in this thread.   Thus, the total score of an album could be expressed as:   total_score = log(f)*g^3  =  log(number_of_ratings+1)*weighted_avg_rating^3   the last thing would simply consist in making a top 100 only according to this "total_score" value.   i'm curious how the output (top 100) would look like!   It looks like a very good algorithm to me, far better than the one we have now. I have tested it some. Was this the old one? :P

Edited by greenback - July 13 2007 at 13:15