Serving the Quantitative Finance Community

 
User avatar
MHill
Topic Author
Posts: 21
Joined: February 26th, 2010, 11:32 pm

Website Top Posts

September 20th, 2020, 10:53 pm

I was looking at a webpage that ranks users' posts.  Each post has four pieces of information - the rank (1 is highest), the number of times it's been viewed, the number of "likes", and the time it was posted.

How do I figure out the ranking algorithm?

Newer posts seem to be favoured over older posts if they have the same number of "likes".
Posts with more "likes" are favoured if posts have similar time stamps.
I've no idea if the number of views is relevant.

I tried a simple "Likes" / "time since posted", but that didn't work.  What's a good method to figure out the algo?

Thanks!
 
User avatar
katastrofa
Posts: 7440
Joined: August 16th, 2007, 5:36 am
Location: Alpha Centauri

Re: Website Top Posts

September 21st, 2020, 12:05 am

Since it's about a ranking system and you know the input variables (post time, nb of views and nb of likes), the first idea which comes to mind is fitting the multivariate logistic regression (the standard way of calculating risk scores).
The general "black box" approach is to find any model (a polynomial or a neural network?) which fits the data well and use it hoping for the best (it often works if you have enough data).