3 ms·

Eh, 1 is sort of why I see implicit negative feedback as more useful here. Namely, tracking the duration a user is probably giving their attention to a given it

Eh, 1 is sort of why I see implicit negative feedback as more useful here. Namely, tracking the duration a user is probably giving their attention to a given item, weighted in accordance with how long you'd expect someone to give their attention to an item based on how long it is. For example, I might see a some specific word or pattern of words in a tweet and quickly skip to the next one. That's very low friction but also a powerful signal. There are drawbacks (e.g. the long boring video that looks like something is about to happen but never does), but that's mitigated by combining this with other signals. With 2, it's an explore/exploit trade-off. You try to explore as far and wide as you can while trying to avoid the things the user may dislike, all while sprinkling in just enough of the stuff that you know they'll like.

phailhaus 4y ago
You're talking like an ML engineer. :) Yes, there are signals in human behavior that you can feed to your model. But no, it is never going to learn "on Mondays he works on his comics, so he'd prefer to see webcomic-related content" Don't sprinkle in your explore/exploit experiments. I know what I want, just let me decide based on what I'm in the mood for! TikTok-style feeds are the absolute worst offender here, where they couldn't care less about what you think. They will serve you content, and you will either say "yes" or "no". So the only option for you as the user is to just wander through their content hyperspace. There's no structured way to jump between topics because everything lives in this formless content soup. The other problem is that many social media platforms have you subscribe to the content streams of individuals directly. Individuals are high variance. How can you teach these engines that "I only care about this person's posts about pianos, not their terrarium hobby."
wizzwizz4 4y ago
> Namely, tracking the duration a user is probably giving their attention to a given item, I pay a lot of attention to classes of things I don't want to see at all.
- throwaway29812 4y ago
  Then, according to the algorithm, you do want to see them.
  - phailhaus 3y ago
    That's exactly the problem! I don't want to see them, but the algorithm refuses to ask me. If they think that the length of time I spend on a post is 100% correlated with the amount that I want to see it, they really don't understand people at all. It's an embarrassingly shallow model of how how humans behave.
    - throwaway29812 3y ago
      > It's an embarrassingly shallow model of how how humans behave. It's a trade off. Viewing rates give back feedback 100% of the time. Asking users for a thumbs up or down gives feedback almost none of the time, and still might not be accurate.

phailhaus 4y ago

You're talking like an ML engineer. :) Yes, there are signals in human behavior that you can feed to your model. But no, it is never going to learn "on Mondays he works on his comics, so he'd prefer to see webcomic-related content" Don't sprinkle in your explore/exploit experiments. I know what I want, just let me decide based on what I'm in the mood for! TikTok-style feeds are the absolute worst offender here, where they couldn't care less about what you think. They will serve you content, and you will either say "yes" or "no". So the only option for you as the user is to just wander through their content hyperspace. There's no structured way to jump between topics because everything lives in this formless content soup. The other problem is that many social media platforms have you subscribe to the content streams of individuals directly. Individuals are high variance. How can you teach these engines that "I only care about this person's posts about pianos, not their terrarium hobby."

wizzwizz4 4y ago

> Namely, tracking the duration a user is probably giving their attention to a given item, I pay a lot of attention to classes of things I don't want to see at all.

throwaway29812 4y ago
Then, according to the algorithm, you do want to see them.
- phailhaus 3y ago
  That's exactly the problem! I don't want to see them, but the algorithm refuses to ask me. If they think that the length of time I spend on a post is 100% correlated with the amount that I want to see it, they really don't understand people at all. It's an embarrassingly shallow model of how how humans behave.
  - throwaway29812 3y ago
    > It's an embarrassingly shallow model of how how humans behave. It's a trade off. Viewing rates give back feedback 100% of the time. Asking users for a thumbs up or down gives feedback almost none of the time, and still might not be accurate.