Need Help in Designing Ranking Algorithm | Hacker News Reader

Need Help in Designing Ranking Algorithm

Hey Guys, I have this site called http://rivals4ever.com. In one line it is "Community based on Rivalries". You can create your own rivalries on this site and invite your friends to join and have kinda virtual fight with each other.

Now we need help in divising proper algorithm for the homepage listing of Rivalries. We tried 2-3 so far but none is giving satisfactory result.

We want something like this site where news on homepage constantly gets changed everytime you refresh or something like reddit or digg. But in our case there are many more parameters which can be considered while designing the algorithm.

1) creator rating 2) number of videos/images/weapons/weakness 3) quantity and quality of content etc etc etc.

Can anyone please guide me on this ? I want the best rivalries on homepage same as this site where best news are on homepage.

Any kind of help is greatly, sincerely appreciated :)

Abhishek

View on HN

I'm not completely sure I understand the problem-space, so excuse me if this if way off-base:

It appears to me that what you are looking for is not really an algorithm per se, but a formula. If you have a series of parameters, and you want to weight them to create an overall ranking, the challenge is really one of choosing the appropriate weights.

So, to use your example:

RANK = (X * CreatorRanking) + (Y * NumberOfVideosEtc) + (Z * QuantityOfContent)

where X, Y, and Z are chosen to emphasize whatever factors you think are most important to determining the "best" rivalry.

Is that what you had in mind, or am I missing something?

Yes you are right, I am talking about Formula only. Sorry for using the wrong word here. Can we have some formula which gives most interesting rivalries everytime on homepage ?

Well if the formula you need is as given in this thread then the only thing you need to define is the parameters X,Y, and Z...

The method you use to find these parameters would then be the problem

the only problem with this kind of formula is that it does not get rid of older rivalries from homepage because they will be having maximum fans, contents etc. for e.g. pepsi vs coke. I dont want this rivalry to be on the homepage everytime someone gives a visit.

In that case, you make "time since creation" one of your parameters, and assign that an appropriate weight.

That formula is already applied in the site but I think time variable should be less now

Well, there you go. You're really the only one who can decide what the appropriate weights are, based on your own individual preferences.

It would be interesting to hear you enumerate all signals you're considering, as one part of the problem will be to consider what other signals should be used.

Also, do you need the formula to be resistant to spam?

-- Dave

Yes Dave it should be spam resistant. I am looking at to utilize quality and quantity of contents to rank a rivalry. Contents are in form of weapons, weaknesses, images, videos, passion talk and news. Other sub contents can be comments on each of these contents. We can have number of fans on each side into consideration. Plus ranking and loyalties of those fans. So there are many things we can take into account but the problem is how we can make sure that homepage has good amount of variations in terms of rivalries it displays.

I hope I am clear here.

do you want the organic-style "cream of the crop rises to the top"-style ranking that the digg and reddit front pages use?

what have you tried so far?

yes cream of the crop rises to the top style ranking will be better but that does not mean we dont give proper exposure to new rivalries which are being added. We have tried this kind of ranking before but in that case old rivalries remain on top because they have the maximum contents, fans etc. We dont want that. We want dynamism on the homepage.

have you tried adding a simple decay (exponential might be too sharp - i start with logarithmic and tweak from there) based on the amount of time that the rivalry has existed?

hotness could be as simple as starting with #votes/time alive and playing around with the variables from there.

Time is a factor right now but it is linear factor which i guess is not working right now. Need to make it logarithmic.

also: http://www.thebroth.com/blog/118/bayesian-rating

this book may help: http://www.oreilly.com/catalog/9780596529321/

Thanks :) looks cool

good book

Reply on news.ycombinator.com