alt Hacker News> Amazon (or whoever) showing you the same stuff you just bought is kind of a separate issue, because it obviously isn't useful to anyone (including Amazon) and surely is so easy to avoid I find it bizarre that it still happens in this day and age.
I've asked people at Amazon on the recommendation algorithm side about this, since it comes up so often in discussions.
Their answer is that, surprisingly, it is a good heuristic. If you just bought something, you've shown intent, and there's a small probability that you're going to be unhappy with your initial purchase and back on the market soon, for whatever reason.
Is the posterior probability that you'll buy a mattress again shortly after buying one, a small chance? Yes, undoubtedly. But is it smaller than the prior probability of a random person who hasn't bought any mattress at all recently, suddenly being on the market for a mattress? Apparently not.
Weird things can happen with joint probabilities when both sides of the condition are very unlikely. Sort of like how your chance of having some rare cancer after testing positive on a ~99% accurate cancer screening test is still quite tiny.
Replies
People always have said this from when I was in ad tech over a decade ago and to be honest, I think it's much more mundane: a delay in the ad tech data pipeline. If you have delayed intent signal, or you have delayed suppression, or even if some conversion gets dropped in the pipeline, you're going to get this effect. People talked about this a lot but almost all the focus on recent purchaser lists was on suppression. So I'm just going to go with the boring explanation: data pipeline error.
I find Amazon recommendations are quite bad. Maybe only 10% are relevant to me.
> If you just bought something, you've shown intent, and there's a small probability that you're going to be unhappy with your initial purchase and back on the market soon, for whatever reason.
It probably helps when your marketplace is filled with knockoffs and misrepresented trash.
> Is the posterior probability that you'll buy a mattress again shortly after buying one, a small chance? Yes, undoubtedly. But is it smaller than the prior probability of a random person who hasn't bought any mattress at all recently, suddenly being on the market for a mattress? Apparently not.
This makes no sense. If I just bought a mattress, the prior probability of a random person buying it is irrelevant. What's relevant is the posterior distribution of what people buy after buying a mattress. Yes, you might buy another one, but what about other bedroom accessories for example?
A mattress might not be the best example though, because what really matters here is the expected value from the posterior probability, which includes the price. So a more expensive product with a slightly lower probability of purchase might yield a higher expected value than cheaper products with higher probability.
Anyway, that's orthogonal to the discussion: my point is that it's the posterior distribution conditioned on the mattress purchase that should be modeled, not the comparison with random people that may or may not just have bought a mattress.