The “latest” tab works via the hacker news algorithm. Ruby has a footnote about it here. I think we set the “starting age” to 2 hours, and the power for the decay rate to 1.15.
Very helpful, thanks! So I assume the parameter b is what you call starting age?
I ask because I am a bit confused about the following:
If you apply this formula, it seems to me that all posts with karma = 0 should have the same score, that score should be higher than the score of all negative-karma posts and negative-karma posts should get a higher score if they are older.
All karma>0 posts should appear before all karma=0 posts and those should appear before all negative-karma posts.
However, when I expand my list a lot until it íncludes four posts with negative karma (one of them is 1 month old), I still do not see any post with zero karma. (With “enriched” sorting, I found two recent ones with 0 karma.)
Moreover, this kind of sorting seems to give really a lot of power to the first one or two people who vote on a post if their votes can basically let a post disappear?
Could you send me a screenshot of your post list and tag filter list? What you are describing sounds really very weird to me and something must be going wrong.
The list is very long, so it is hard to make a screenshot. Now with some hours of distance, I reloaded the homepage, tried again, and one 0 karma post appeared. (Last time, it did definitely not, I searched very rigorously.)
However, according to the mathematical formula, it still seems to me that all 0 karma post should appear at the same position, and negative karma posts below them?
We have a few kinds of potential bonus a post could get, but yeah, something seems very off about your sort order, and I would really like to dig into it. A screenshot would still be quite valuable.
I will see whether I can make a useful one later on. Still, my main point is about the sorting score as stated in that referenced footnote: if indeed a post karma is divided by whatever, then I expect all 0 karma post to appear at the same position, and I expect the first person who votes to have a strong influence leading to herding, in particularif the personvotes the post to zero or lower. Right?
(c) posts appearing in recent discussion in order that they’re posted (though I do wonder if we filter out negative karma posts from recent discussion)
I often play around with different karma / sorting mechanisms, and I do think it would be nice to have a more Bayesian approach that started with a stronger prior. My guess is the effect you’re talking about isn’t a big issue in practice, though probably worth a bit of my time to sample some negative karma posts.
Is there an explanation somewhere how the recommendations algorithm on the homepage works, i.e. how recency and karma or whatever are combined?
The “latest” tab works via the hacker news algorithm. Ruby has a footnote about it here. I think we set the “starting age” to 2 hours, and the power for the decay rate to 1.15.
Very helpful, thanks! So I assume the parameter b is what you call starting age?
I ask because I am a bit confused about the following:
If you apply this formula, it seems to me that all posts with karma = 0 should have the same score, that score should be higher than the score of all negative-karma posts and negative-karma posts should get a higher score if they are older.
All karma>0 posts should appear before all karma=0 posts and those should appear before all negative-karma posts.
However, when I expand my list a lot until it íncludes four posts with negative karma (one of them is 1 month old), I still do not see any post with zero karma. (With “enriched” sorting, I found two recent ones with 0 karma.)
Moreover, this kind of sorting seems to give really a lot of power to the first one or two people who vote on a post if their votes can basically let a post disappear?
A quick question re: your list: do you have any tag filters set?
I don’t think so. But where could I check that?
Click on the gear icon next to the feed selector
No, all tags are on default weight.
I had a quick look in the database, and you do have some tag filters set, which could cause the behaviour you describe
Thanks. I dud not see any, but I will check again. Maybe I also accidentally set them when i tried to check whether I had set any...
Could you send me a screenshot of your post list and tag filter list? What you are describing sounds really very weird to me and something must be going wrong.
The list is very long, so it is hard to make a screenshot. Now with some hours of distance, I reloaded the homepage, tried again, and one 0 karma post appeared. (Last time, it did definitely not, I searched very rigorously.)
However, according to the mathematical formula, it still seems to me that all 0 karma post should appear at the same position, and negative karma posts below them?
We have a few kinds of potential bonus a post could get, but yeah, something seems very off about your sort order, and I would really like to dig into it. A screenshot would still be quite valuable.
I will see whether I can make a useful one later on. Still, my main point is about the sorting score as stated in that referenced footnote: if indeed a post karma is divided by whatever, then I expect all 0 karma post to appear at the same position, and I expect the first person who votes to have a strong influence leading to herding, in particularif the personvotes the post to zero or lower. Right?
Yep, if the first vote takes the score to ≤ 0, then the post will be dropped off the latest list. This is somewhat ameliorated by:
(a) a fair number of people browsing https://lesswrong.com/allPosts
(b) https://greaterwrong.com having chronological sort by default
(c) posts appearing in recent discussion in order that they’re posted (though I do wonder if we filter out negative karma posts from recent discussion)
I often play around with different karma / sorting mechanisms, and I do think it would be nice to have a more Bayesian approach that started with a stronger prior. My guess is the effect you’re talking about isn’t a big issue in practice, though probably worth a bit of my time to sample some negative karma posts.
Maybe the numerator of the score should remain at the initial karma until at least 4 people have voted, for example.