Post Reputation model - stats and excel
New here? Learn about Bountify and follow @bountify to get notified of new bounties! x

I have a social site that i need to rank posts in terms of popularity. We have a hand full of metrics to pull on for each post to help with our definition of popular. Our current approach doesn't seem to consider time as a really negative variable, we also don't have non linear math helping here which I think is needed. Let me explain the scenario

when someone makes a post. we measure the following things

-Clicks on that post ( eyeballs on it)
-Comments (like a blog post)
-A series of rating interations (here is our buttons so you have context http://screencast.com/t/OmliqCcEL )
-Time post was created (we tend to think older content needs to be less popular unless overwhelmingly active)

here is our current model
https://dl.dropboxusercontent.com/u/2982102/Scoring_model.xlsx

We wish for the following
Visits in our early days need to be a big deal since we might not get a lot of ratings or comments
Comments are a massive deal and should be a heavy indicator of popular
Feedback ratings are important but aren't a big deal until we get a lot of them, meaning 1-2 is probably a more casual rating than 20.
Time is slippery slope, if old we need it to allow new things to show up higher, but the difference in 1 day old vs 1 week old is a bigger deal than 1 month vs 2 months. so i assume we need a non linear approach here.

Deliverable here should be a functioning excel sheet we can use to run scenarios against.

This is a very interesting problem. I think i can solve it, but could you please explain some of the variables? What does the const mean? Also what does the 'min date' and 'last updated' represent?
Sebastiandk 6 years ago
promise not to laugh, we used constant as a way to get the sum of the click interactions to be more meaningful. its 100% arbitrary and where i think our linear model has problems. this was a band-aide for it. last updated is the creation date for the most part, min date is today. it was our way to find the age of the post so we could determine age - its an epoch timestamp... thanks!
Qdev 6 years ago
awarded to Sebastiandk

Crowdsource coding tasks.

1 Solution

Winning solution

Ok. I'm new to dropbox but here's my solution. Tell me if you want any changes/improvements to it, or if you need anything explained. :)

https://www.dropbox.com/s/9ufiwwwisut0628/ScoringSystem.xlsx?dl=1

This looks cool. I'll play with scenarios but this looks to be what I needed.
Qdev 6 years ago
Can you give me a high level write-up on why you took some of these approaches. I can shoot you a bonus for the extra work here on documentation. But having this will help us explain and defend the approach we took here. just shoot it to john at quotient dot net
Qdev 6 years ago