The last update to the popularity scoring algorithm solicited a lot of comments. After working with Lupo73, here's another attempt at the algorithm (version 3).
To recall, V1 of the algorithm accumulates the activity and votes over the entire lifespan of each entry. Some of you complained that it was too static.
V2 of the algorithm accumulates the activity and votes of each entry over the past 30 days. Some of you complained that it was too dynamic and random
V3 of the algorithm computes the popularity score based on two components: activity over the past 30 days, and votes over the entire lifespan of each entry.
Activity is defined as clicks to the "website" and "download" links. "download" clicks are weighted 3 times more than "website" links. Recent clicks are weighted more than aged clicks using an exponential window.
Votes are weighted based on the ranks of the user. The final score is computed using the Wilson Popularity Formula.
The two components are combined together using a 25% (activity) / 75% (votes) ratio to form a final figure that is between 0 to 100. The idea is that votes are more important that activity for the final score, but to compare between apps with few votes, the activity will give an indication of its popularity.
The TPFC website has been updated with the new scoring algorithm. The five-star image on the right of each entry gives a graphical overview of its popularity. Mousing over the image will give you the actual score (rounded to 1-decimal place).
We hope this new formula will bring a better balance for the popularity score. I am sure we can work together to refine it even further. Meanwhile, have fun!
PS: Many thanks to Lupo73 for his fantastic work on this update.
Yet another update to the popularity score algorithm
- Andrew Lee
- Posts: 3064
- Joined: Sat Feb 04, 2006 9:19 am
- Contact:
Re: Yet another update to the popularity score algorithm
Looks really good!
Re: Yet another update to the popularity score algorithm
On a side note the overall database interface is more 'clean' or is it just my eyes and Opera?
-
- Posts: 1212
- Joined: Wed Jul 18, 2007 5:45 pm
Re: Yet another update to the popularity score algorithm
Cool!
A minor aesthetic thing, but I'd like to see a better star graphic than the puke-green colored star that's currently in use.
A minor aesthetic thing, but I'd like to see a better star graphic than the puke-green colored star that's currently in use.
is it stealth?
Re: Yet another update to the popularity score algorithm
puke-green? I'm guessing it's already changed because it's a plain yellow on my side...
Like the update though, should work great!
Like the update though, should work great!
Re: Yet another update to the popularity score algorithm
It depends on your monitor. Sometimes grey-yellow colors can look greenish. Assuming there are color options within whatever code your'e using for this, I would go with a subdued grey. Additionally, I think its important we somehow let people know this isn't like Softpedia's rating system: e.g. 1/2 star doesn't reflect on the program's quality. One possible way to demonstrate this is just put "popularity" next to the stars.dany wrote:puke-green? I'm guessing it's already changed because it's a plain yellow on my side...
Like the update though, should work great!
Here's how those two elements might look together (made using PhotoFiltre):
Re: Yet another update to the popularity score algorithm
I just realised that's a 'tick box image' next to the number of members who have voted for the application. It wasn't very clear in the browser (Opera) so for me it looked like an HTML parsing error of some description, but in the image it's more clear.
- Andrew Lee
- Posts: 3064
- Joined: Sat Feb 04, 2006 9:19 am
- Contact:
Re: Yet another update to the popularity score algorithm
I have rearranged the various data fields so that there is less tendency to wraparound on smaller screens. Could that be it?On a side note the overall database interface is more 'clean' or is it just my eyes and Opera?
I have changed the tooltip to read (Popularity: XXX/100). Also, putting the stars close to the rating links should make it clear they are correlated with user ratings.Additionally, I think its important we somehow let people know this isn't like Softpedia's rating system: e.g. 1/2 star doesn't reflect on the program's quality. One possible way to demonstrate this is just put "popularity" next to the stars.
Yup, that's actually a Unicode wingding (☑). But I just changed that to a "heart" (♥) 'cos I thought that more closely relates how we feel about the app. If you mouse over the link, it should say "xxx member points".I just realised that's a 'tick box image' next to the number of members who have voted for the application.
Re: Yet another update to the popularity score algorithm
OK, thanks Andrew for the changes.
Re: Yet another update to the popularity score algorithm
Good update ..I'm considering to write an article about the new algorithm used, to allow other websites to test and use it.. anyway I'll wait few weeks to test it here and verify its quality..