3 ms·

There's a fundamental problems with your approach. To reliably determine if someone has actually read a comment, this means the comment has to be hidden. You c

by geuis 14y ago

There's a fundamental problems with your approach. To reliably determine if someone has actually read a comment, this means the comment has to be hidden. You cannot assume that simply because something is viewable that the user is viewing it. Since you have to hide a comment, this requires a user interaction event to change its state. This is a broken model since comments are primarily a passive, not interactive, experience.

Mahn 14y ago
This is what I thought aswell. Unless you cripple the UI, there's no way you are gonna be able to reliably determine whether a comment has been read programmatically. You could try to guess through a number of things, but it'll never be accurate enough not to break the order by false positives.
Draco6slayer 14y ago
But just because you, in one particular instance didn't see it, doesn't mean anything in the long term. Obviously with any statistical estimate, individual data points can fall far from the curve. But the idea is that, after enough time, enough views, and enough upvotes, the system slides into place.
- KajMagnus 14y ago
  I agree with this. The algorithm doesn't need to be exactly correct, always. It only needs to work well on the whole, given data on many visitors.
KajMagnus 14y ago
Actually, no, there is no fundamental problem with the approach. Search engines have a similar problem: They estimate how useful a search result link is, by counting the number of people that click that link. To do this, they need to take into account that people tend to click the topmost search results only. But they are handling it just fine, without hiding any search results or something like that. Instead they simply count clicks on the search result link, and apply some mathematics related to the probability that you click that link, when it's so and so far away from the top. And this is similar to the approach I suggested in example 1, which relies on clicks on the vote up/down button (instead of search result link). (If people don't interact with the page (don't upvote anything at all), I think an algorithm could simply disregard those people.)

by geuis 14y ago

Mahn 14y ago

This is what I thought aswell. Unless you cripple the UI, there's no way you are gonna be able to reliably determine whether a comment has been read programmatically. You could try to guess through a number of things, but it'll never be accurate enough not to break the order by false positives.

Draco6slayer 14y ago

But just because you, in one particular instance didn't see it, doesn't mean anything in the long term. Obviously with any statistical estimate, individual data points can fall far from the curve. But the idea is that, after enough time, enough views, and enough upvotes, the system slides into place.

KajMagnus 14y ago
I agree with this. The algorithm doesn't need to be exactly correct, always. It only needs to work well on the whole, given data on many visitors.

KajMagnus 14y ago

Actually, no, there is no fundamental problem with the approach. Search engines have a similar problem: They estimate how useful a search result link is, by counting the number of people that click that link. To do this, they need to take into account that people tend to click the topmost search results only. But they are handling it just fine, without hiding any search results or something like that. Instead they simply count clicks on the search result link, and apply some mathematics related to the probability that you click that link, when it's so and so far away from the top. And this is similar to the approach I suggested in example 1, which relies on clicks on the vote up/down button (instead of search result link). (If people don't interact with the page (don't upvote anything at all), I think an algorithm could simply disregard those people.)