Update to popularity score algorithm

Changes, updates etc. related to this website will be posted here.
Message
Author
ChipsNHotSauce
Posts: 16
Joined: Mon Jun 06, 2011 9:35 am
Location: Growls Garden

Re: Update to popularity score algorithm

#16 Post by ChipsNHotSauce »

I second on "Recent" and "All time" high scores... that's what is used by most websites and it works.

abacusFanboi
Posts: 18
Joined: Thu Jan 14, 2010 7:56 pm

Re: Update to popularity score algorithm

#17 Post by abacusFanboi »

just appropriately label whichever score.
- page visits (from search engines, links posted elsewhere, and bookmarks).
- survey votes

disclaimer: i don't much notice these types of popularity score things

User avatar
Andrew Lee
Posts: 3048
Joined: Sat Feb 04, 2006 9:19 am
Contact:

Re: Update to popularity score algorithm

#18 Post by Andrew Lee »

Here are the popularity scores as of today (minus old scores):

Typhoon 2001 => 13701
VirtuaWin => 12773
TrueCrypt => 11353
Opera => 11174
XnView => 9510
Converber => 9098
FastStone Image Viewer => 8994
Portable Start Menu => 8368
freeCommander => 8105
KeePass => 7727

Compare that with old score inclusive:

FastStone Capture => 196930
Yod'm 3D => 99423
Foxit Reader Portable => 68417
Undelete Plus => 29393
SilentNight Micro CD Burner => 27443
PixaMSN => 20691
EVEREST Home Edition => 19900
Universal Extractor => 19585
CCleaner => 17341
Mozilla Firefox, Portable Edition => 16610

User avatar
SYSTEM
Posts: 2041
Joined: Sat Jul 31, 2010 1:19 am
Location: Helsinki, Finland

Re: Update to popularity score algorithm

#19 Post by SYSTEM »

Andrew Lee wrote:Here are the popularity scores as of today (minus old scores):

Typhoon 2001 => 13701
VirtuaWin => 12773
TrueCrypt => 11353
Opera => 11174
XnView => 9510
Converber => 9098
FastStone Image Viewer => 8994
Portable Start Menu => 8368
freeCommander => 8105
KeePass => 7727
Quite surprising. :o Many of these applications aren't well known.

And where are Foxit Reader Portable and Undelete Plus :?: They have got a lot of points relatively recently.

The change will be more radical than I thought. (Not sure if it's a good or bad thing. :? ) Maybe 30 days is a too short time to measure popularity.
My YouTube channel | Release date of my 13th playlist: August 24, 2020

Hydaral
Posts: 194
Joined: Tue Mar 09, 2010 7:36 pm

Re: Update to popularity score algorithm

#20 Post by Hydaral »

SYSTEM wrote:
Andrew Lee wrote:Here are the popularity scores as of today (minus old scores):

Typhoon 2001 => 13701
VirtuaWin => 12773
TrueCrypt => 11353
Opera => 11174
Quite surprising. :o Many of these applications aren't well known.
What surprises me is that Opera is above Firefox. Does this mean people are switching from Firefox to Opera?

User avatar
m^(2)
Posts: 890
Joined: Sat Mar 31, 2007 2:38 am
Location: Kce,PL
Contact:

Re: Update to popularity score algorithm

#21 Post by m^(2) »

Hydaral wrote:What surprises me is that Opera is above Firefox. Does this mean people are switching from Firefox to Opera?
Hardly. While it is possible that it's true, the data doesn't support such conclusion. It says that there are more people trying Opera than Firefox. It can mean many things, but it's likely that on Windows Opera grows faster than Firefox.
ADDED: No, I think jumped to conclusions too early, it doesn't even say anything about overall growth. Portable browsers is a specific niche and 2 of the biggest players in regular web browser area don't compete here (IE and Safari). Users of these browsers have to either give up portable use of switch to something else, but only for portable usage. Many if those people won't use the portable browser of their choice on their PCs, so dynamics that include them don't reflect overall situation well.

User avatar
Andrew Lee
Posts: 3048
Joined: Sat Feb 04, 2006 9:19 am
Contact:

Re: Update to popularity score algorithm

#22 Post by Andrew Lee »

The change will be more radical than I thought. (Not sure if it's a good or bad thing. :? ) Maybe 30 days is a too short time to measure popularity.
But isn't this the purpose of the change, to make the list more dynamic? The reason why this change was proposed was because some of you mentioned that the most popular list is almost static due to the accumulated scores.

All things being equal, the size of the window should be directly correlated with how dynamic the list is. The small the window size, the quicker the list should change, in theory.

User avatar
SYSTEM
Posts: 2041
Joined: Sat Jul 31, 2010 1:19 am
Location: Helsinki, Finland

Re: Update to popularity score algorithm

#23 Post by SYSTEM »

Andrew Lee wrote:
The change will be more radical than I thought. (Not sure if it's a good or bad thing. :? ) Maybe 30 days is a too short time to measure popularity.
But isn't this the purpose of the change, to make the list more dynamic? The reason why this change was proposed was because some of you mentioned that the most popular list is almost static due to the accumulated scores.
Exactly, but the list based on new scores might be too dynamic, placing applications in quite random order.
My YouTube channel | Release date of my 13th playlist: August 24, 2020

User avatar
Andrew Lee
Posts: 3048
Joined: Sat Feb 04, 2006 9:19 am
Contact:

Re: Update to popularity score algorithm

#24 Post by Andrew Lee »

Exactly, but the list based on new scores might be too dynamic, placing applications in quite random order.
Random??!! Hardly. Here are the scores for today.

Typhoon 2001 => 13705
VirtuaWin => 12774
TrueCrypt => 11367
Opera => 11178
XnView => 9511
Converber => 9099
FastStone Image Viewer => 8997
Portable Start Menu => 8370
freeCommander => 8106
KeePass => 7728

When in operation, the list should still be mostly static until an app (old or new) suddenly becomes more active for whatever reason. The new system will simply give this app a chance to appear in the top list. Under the previous system, the newly active app will not stand a chance at all.

User avatar
SYSTEM
Posts: 2041
Joined: Sat Jul 31, 2010 1:19 am
Location: Helsinki, Finland

Re: Update to popularity score algorithm

#25 Post by SYSTEM »

Andrew Lee wrote:
Exactly, but the list based on new scores might be too dynamic, placing applications in quite random order.
Random??!! Hardly. Here are the scores for today.

Typhoon 2001 => 13705
VirtuaWin => 12774
TrueCrypt => 11367
Opera => 11178
XnView => 9511
Converber => 9099
FastStone Image Viewer => 8997
Portable Start Menu => 8370
freeCommander => 8106
KeePass => 7728

When in operation, the list should still be mostly static until an app (old or new) suddenly becomes more active for whatever reason. The new system will simply give this app a chance to appear in the top list. Under the previous system, the newly active app will not stand a chance at all.
Well, I refuse to believe that e.g. Typhoon 2001 could stay the most popular application even for few weeks. I see that it has become more popular within the latest 30 days, but... why? It was not updated or even mentioned in the forum in the meantime.

Do you get what I mean? A reason independent of TPFC can raise an application to the first place! :shock:

Finally, let me repeat that IMO this rapidly changing scores might actually be good. It depends on what the popularity scores are designed to measure in the first place.
My YouTube channel | Release date of my 13th playlist: August 24, 2020

User avatar
Lupo73
Posts: 1012
Joined: Mon Mar 19, 2007 8:55 am
Location: Italy
Contact:

Re: Update to popularity score algorithm

#26 Post by Lupo73 »

I think current solution cannot work fine, because remove popularity history is not a good solution at all (in my opinion).. I searched for a better solution and I found a couple of interesting articles:
http://www.evanmiller.org/how-not-to-so ... ating.html
http://blog.linkibol.com/2010/05/07/how ... -proud-of/

First of all, they explain why easy solutions like the following don't work fine:
score = positive votes - negative votes
or
score = positive votes / total votes
The two best solutions seem to be:
1. the bayesian average formula, eventually adapted to the needs
2. the Lower bound of Wilson score confidence interval for a Bernoulli parameter

I think the second one could be used for PFC database.. and I think the two different ratings (non-registered and registered) may be unified in a unique one (giving for example a *10 to registered user votes)..

Another solution could be to create a modified bayesian average formula, including: non-registered votes, registered votes, download counter, etc.

User avatar
Andrew Lee
Posts: 3048
Joined: Sat Feb 04, 2006 9:19 am
Contact:

Re: Update to popularity score algorithm

#27 Post by Andrew Lee »

Thanks for the links! I read the two pages over the weekend but have not totally digested them.

I will keep working on them and update the rather simple formula used here once I can totally understand the formula.

User avatar
Lupo73
Posts: 1012
Joined: Mon Mar 19, 2007 8:55 am
Location: Italy
Contact:

Re: Update to popularity score algorithm

#28 Post by Lupo73 »

I'm studying them too and trying to write a bayesian adapted formula.. we could discuss the implementation if you want.. I'll post the formula soon..

User avatar
Lupo73
Posts: 1012
Joined: Mon Mar 19, 2007 8:55 am
Location: Italy
Contact:

Re: Update to popularity score algorithm

#29 Post by Lupo73 »

I wrote a script to better understand the formula and to test it:
http://www.lupopensuite.com/files/score.au3

Obviously in AutoIt :mrgreen:
At the beginning you can see also a description of each term of the formula..
I'm still not fully sure that it is correct, but the example result seems good..
It returns the list of software ratings as: + votes, - votes, % popularity

Test results:
Image

An external revision is welcome :wink: ..I'm going to integrate also support to registered-users rating (I think a unified score could be very good), we need only to decide the weight of registered-user votes..

User avatar
m^(2)
Posts: 890
Joined: Sat Mar 31, 2007 2:38 am
Location: Kce,PL
Contact:

Re: Update to popularity score algorithm

#30 Post by m^(2) »

Why 100+ 100- has lower popularity than 10- 10-?
IMO it should work the other way...

Post Reply