The List Central Ranking System

When I first released List Central into private beta back in November it used a ranking system that was completely of my own invention. I decided to re-invent to wheel, and ended up with something that didn’t make any sense.It took being able to see my ranking algorithm in action with real lists before I could see how truly out of whack it was.

The original ranking algorithm took too much into account, and didn’t handle ageing very well. Rather than fix what I had, I hit the streets of the wide web to research what others were doing for their social ranking algorithm. I thought this might be a difficult task, as I thought the ranking algorithms of the world would be tightly guarded trade secrets, but there is actually a wealth of information on the topic.

I easily found the Reddit algorithm, and Paul Graham’s algorithm for Hacker News in HN threads. The post  How to build a Popularity Algorithm you can be proud of at the linkiblog was super helpful as well.

I ended up implementing both the Reddit and the Hacker News algorithms to test out which worked best for List Central. Both were functions of points (X) and time. I had to decide how to get X for lists. Both Reddit and HN use a thumbs up/thumbs down system for point calculation. List Central uses a 5 start ranking system. I came up with a mapping of star ratings to points, with 5 stars being the most, and 1 star being the least. Both the 1 a 2 star ratings have negative values when mapped to points, to enable users’ expressions of distaste to have the intended effect on the list ranking.

With my X in hand, and two algorithms implemented, I let them run to see how they preformed. I was pleased with both the algorithms, and ended up keeping a modified version of the Reddit algorithm for List Central.

The new popularity algorithm took away the need for the time frames List Central used to have on the main page, similar to how Digg is implemented. I only required the one tab “popular”. I added to new tabs to give more points of access to the lists: “new” – ordered by date published and “active” – ordered by an activity measuring algorithm that factors in list hits and comments.

All in all I am pleased with how the new ranking algorithm worked out, and am grateful for the lessons on wheel re-invention the process taught me.