patrick.net

rando says

select count(*), cast(post_date as date) as d from posts left join users on post_author=user_id

where user_name='Tenpoundbass' group by d order by d desc limit 20;

There's a bug in your query. Please run the corrected query.

select count(*), cast(post_date as date) as d from posts left join users on post_author=user_id
where user_name='Tenpoundbass'; delete posts from posts left join users on post_author=user_id where user_name='Tenpoundbass'; -- group by d order by d desc limit 20;

rando says

Maybe net likes is the best?

No, trolls just dislike all posts by authors they hate. You'd have to use a more sophisticated like sorting algorithm such as


decimal total = 0.0;

for each like in likes
total += Agreement(self.userId, like.userId) * like.value;


where like.user is the ID of the user who submitted the like and agreement(int userIdA, int userIdB) is a function that returns a heuristic from -1.0 to 1.0 indicating how many times the logged in user, self.userId, likes the same thing the like.userId user likes and how often the two users disagree. Something like...


decimal Agreement (int userIdA, int userIdB)
{
IEnumerable likesA = likes.Where(like => like.userId = userIdA);
IEnumerable likesB = likes.Where(like => like.userId = userIdB);

return agreement;
}


The value of decimal Like.value can be either one of the choices (1 like, 0 neutral, -1 dislike) or a value in the range [-1.0, 1.0] indicating how much the post is liked by the reviewer. The above algorithm will work either way.

The effect of using this technique is that if Bob agrees with Joe most of the time, Bob will see recommendations of Joe. However, if Bob disagrees with Sam most of the time, then Bob will be recommended things that Sam hates and vice versa. In other words, if two users have the same tastes, their likes are complimentary. But if two users have different tastes, each one's like counts as a dislike for the other while each one's dislike counts as a like for the other.

This makes gaming likes and dislikes impossible. The more trolls dislike a user's posts, the higher it rates with someone who dislikes that troll.

This is what you should implement on PatNet. Of course, it only works for users who log in and like or dislike posts. Otherwise they get no recommendations. And that's fair. How do you recommend something to someone you don't know and thus don't know what he or she likes?

@Patrick, the code tag does not preserve whitespaces.

Yes, shit. Will fix in node version.

Dan8267 says

The effect of using this technique is that if Bob agrees with Joe most of the time, Bob will see recommendations of Joe.

Isn't this what Facebook does, and results in echo chambers where you see only things you agree with already?

errc says

Who would pay to advertise here? What product or service and to what demographic

Duh, Canned Yams.

Precisely.
One can get the same effect by banning all people they disagree with most of the time...

rando says

Isn't this what Facebook does, and results in echo chambers where you see only things you agree with already?

Maybe if we fed him and jacked him up to 12 lbs he'd be too stuffed to post as much...

rando says

Here's TPB's most recent post count per day:



select count(*), cast(post_date as date) as d from posts left join users on post_author=user_id

where user_name='Tenpoundbass' group by d order by d desc limit 20;

+----------+------------+

| count(*) | d |

+----------+------------+

The liberals ruined everything with high taxes

Only God and Guns can save us now

Friggin obama

And SJW and trannys!

Muslims in Afghanistan blargle glib blah

Vote Republican 4 life

rando says

Isn't this what Facebook does, and results in echo chambers where you see only things you agree with already?

It's called assortative mating. You actually get conversations with people that you have similar values to, not necessarily agree with. For example, Bob likes the same posts that Joe likes, but Joe likes some posts that introduce him to new ideas contradicting his current views. As a result, Bob sees those same new ideas that contradict Joe's views, and thus likely Bob's views.

You only get the echo chamber effect if you only like things that agree with your current views and only dislike things that contradict your current views. If you do that, then you are in effect asking only to see things that reinforce your current views. The solution is to like things more often and more heavily, if likes are non-binary, that are both insightful and surprising to you. This is what I do already.

Assortative mating is the correct approach to recommendations. That is why it is used throughout the IT and marketing industries. Amazon's product recommendations are also based on this idea.

TwoScoopsMcGee says

Duh, Canned Yams.