|
The short version:
The blog ratings at http://www.teamliquid.net/blogs/ are much more meaningful now. Assuming people vote sanely, the best blogs (Hi Plexa!) will be up top.
Technical stuff:
It now uses IMDB's rating system (Bayesian estimate). There's still room for improvement (it might be better to do this for each blog entry, and take the average of those, I'm not sure).
weighted rank (WR) = (v*R + m*C)/(v+m)
v = number of votes
R = average rating
m = essentially the minimum votes required to be taken seriously in the ranking (currently 10)
C = site-wide average rating
m is a silly little thing, you can read some discussion about it here: http://answers.google.com/answers/threadview?id=507508
If you're curious, C is currently 3.44 (out of 5). IMDB's is 6.7/10, so clearly TL blogs are slightly better than movies.
Update: I made it so you can't vote on your own blogs (sorry :p). C is now 3.29. 475 out of 495 of the self-ratings were 5's.
|
Hmm, this is a pretty nice feature, but a few individuals are ruining the full enjoyment for all (I.e. - the mysterious rater who dishes out 1/5 on every blog entry).
Is it possible to track down these raters? I mean, of course there are bound to be 1 star ratings once in a while, but I think it would be easy tell the intentions of the rater when something like 10 entries in a row receive a 1 within seconds of each other.
|
Ideally at some point we'll take into account each user's voting habits, but I haven't looked into good ways to do that yet.
We can't just use their average because a lot of folks rate all 1s and 5s, and they may balance out. I never really liked the math behind statistics, so I'm open to suggestions.
And I know who the mystery man is, just hoping it's a bug. Otherwise he has some serious issues (I deleted his 800+ 1 ratings yesterday and he's made 30+ more already).
|
United States7488 Posts
On February 02 2008 06:34 SonuvBob wrote:
And I know who the mystery man is, just hoping it's a bug. Otherwise he has some serious issues (I deleted his 800+ 1 ratings yesterday and he's made 30+ more already).
Wow, that'd be a lot of time spent dishing out 1's
|
Very nice updates!
Btw semioldguy, SonuvBob has direct access to the database so he probably was able to do it in a couple minutes 
|
Melbourne5338 Posts
I think he was talking about the mystery man who was giving out 1s, unless it's SonuvBob who has a vendetta against high rated blogs.
|
Belgium8305 Posts
|
On February 02 2008 06:34 SonuvBob wrote:
Ideally at some point we'll take into account each user's voting habits, but I haven't looked into good ways to do that yet.
We can't just use their average because a lot of folks rate all 1s and 5s, and they may balance out. I never really liked the math behind statistics, so I'm open to suggestions.
And I know who the mystery man is, just hoping it's a bug. Otherwise he has some serious issues (I deleted his 800+ 1 ratings yesterday and he's made 30+ more already).
|
is awesome32280 Posts
|
i think animenewsnetwork uses some system to weigh less for people who only rate super high and super low, not sure how it works though...
|
Aotearoa39261 Posts
|
United States5262 Posts
This is freaking awesome, thanks SoB
|
On February 03 2008 00:32 Aepplet wrote:
i think animenewsnetwork uses some system to weigh less for people who only rate super high and super low, not sure how it works though...
Looks like they won't divulge their method (so people can't figure out how to abuse it).
There's basically three voting patterns that skew the results:
1. People who are essentially "too nice" - always voting high (no 1's or 2's). Not a serious problem, but they raise the ratings of all the blogs they vote on, essentially lowering the blogs they don't read or vote on. Same with people who always vote low.
I think at least at some point IMDB took this into account, normalizing each user's votes.
2. People who always vote 5 for things they like and 1 for things they hate, figuring their vote will carry more weight.
This is easily detectable - if the standard deviation is close to 0 (always one particular number) or 2 (all 1's/5's) for a user with more than a handful of votes, we can give their votes less weight.
3. Malicious users. Same as #2, except they tend to vote down all of a user's posts.
These are easy to detect, but I'm not sure what to do with them. I guess we could have a voting ban or something.
|
mrmin123
Korea (South)2971 Posts
Dammit, I was enjoying my #6 spot. ;;
|
|
|
|
|
|
EPT
