So I picked The S1 episode. Then it was another S1 episode. Repeat. Then an S2 episode.
My second option never changed from that first episode from S18 so I never picked it. Perhaps I could have gone through 12-17 seasons of episodes always picking option one until two episodes I've never seen went head to head.
IMO, both options need to randomize each time until first pairings are exhausted.
Yeah, I'm afraid so. At the start, selection two would always be the same as the algorithm doesn't know anything about the standing of any episode; it's trying to decide where in the array to place it (above or below selection two).
It might be worth seeing if I can randomise the episodes displayed, if only so it doesn't feel so repetitive.
Definitely looking into seeing if I can come up with something better though! The problem is making sure that whatever algorithm is picked remains as close to O(n log n) as possible. Randomising options in a way that makes require a lot more comparisons would be far worse.
For the algorithm, but not for the people taking time to do the rankings. Which do you want to prioritize?
Start with uniform similarity, and as preferences are made, adjust them.
Then you have personalization.
The decisions made by other users get a weight assigned to them, which is individual to each logged in user. So every users viewpoint is personalized.
(Apologies for the short explanation, I’m on mobile. If you want to ask more you can email me.)
You're essentially doing A/B comparisons across the entire set.
It looks like you have it so you're basically setting where "B" is before moving on to the next item.
This isn't strictly necessary. You could just generate a novel pair every time and ask the user to choose between them. The thing is that you'd need a way to track a user. So you can make sure that user hasn't seen a certain pair already.
Once you've exhausted all the pairs, you'll know exactly how to sort the array. You'll have an idea before.
You might have an issue with circular lists though. People are fickle. You could have someone who says that A > B > C > A.
In this case, I'd allow for repeat pairings after a certain amount of time. To allow the person to reevaluate essentially.
You could also take the comparisons across all users and compile a general sort of "best of" ranking.
The page you're on already knows what comparisons you've made (otherwise how it would move forward), so this is entirely possible!
I've come up with a way to randomise it (by just picking a random element from an array of comparisons that haven't yet been made), that's the next step. I deployed it earlier, but there was a bug with it so I've had to rollback until I can look into it.
If I say selection one is better, selection one becomes the next episode, S01E02, while selection two stays the same episode. If I say either or selection two is better, selection one is picked at random and selection two stays the same episode. And then if I go back to selection one be being better, selection one moves back to the next episode, ie S01E03 now.
Is that the intended behavior? As I quickly got bored of saying series one was better than this one episode in series three.