Comment by freehunter | Hacker News Reader

Comment by freehunter | Hacker News Reader

Parent Full thread

My business does a lot of posting on Facebook of sometimes controversial things (we're a news site and we sometimes cover local politics). By far the most engaged stuff is the most controversial where people begin to fight and attack each other.

I have ~3k followers on one of my pages. Usually ~200-300 people see any given post if it has no engagements. If it has normal engagement, it might get to 1000-2000 views, still short of my follower count. If it has sparked controversy and there is a fight, I've had the views spike up over 8000-9000 without any shares. Facebook posts to your timeline "your friend commented on this" and others start piling in too. Facebook emails me saying "this post is getting more attention than 95% of the rest of your posts, please pay us money to show it to more people". The more toxic the comments, the more views it gets and the more Facebook begs me to pay them for it.

That's the problem with these algorithms that humans don't watch over. Usually it works great and good content is seen by the people who want to see it. But every now and again it goes out of control and people end up getting hurt and Facebook/Twitter profit from it and even promote it. And as the person who posted it, I have zero ways to stop it from spreading other than deleting the post.

-edit- oh another story... I run a news site for a town, lets call it Townsville. There is another Townsville in another state, but it is not my Townsville. I had a post go super viral, 90,000 views from my 3k followers, because somehow the post made it to the wrong Townsville and 87,000 people were being shown the wrong news article. Again, I had no way to stop this, no tools to correct it. Absolute insanity.

> By far the most engaged stuff is the most controversial where people begin to fight and attack each other.

Well, duh. If people are fighting they are definitely "engaged", but not in a good way. Defining that more engagement equals more success is what made social networks so toxic. The current state that more eyeballs equals more ad money needs to change.

Why do you state that? Some of the largest advances in our society have come from what would be seen as toxic or hostile engagements.

examples?

This is kind of like how the value of information is how surprising it is. Classic information theory, but with a twist, because in these cases the "surprise" element is often not the value of the information itself, but the way it is presented. So these algorithms are quite good at recognizing when things are surprising and thus perhaps deserving of wider sharing, but bad at recognizing what kind of surprise it is - is it actually new information, or is it just froth?

> That's the problem with these algorithms that humans don't watch over. ... and Facebook/Twitter profit from it and even promote it.

Don't be fooled.. Someone is spot checking the training data against the output at a statistically valid sampling interval.. ML in this case is the data equivalent of a limited liability corp..

Why does someone need to spot check the training data against the output? I'd assumed the algorithm is trying to maximize a metric, which is best completely automated without human intervention to slow it down or provide biased opinions. Even fuzzier goals like 'not be racist' would get turned into metrics (eg. proportion of words used that are on this watchlist or similar scoring algorithm)

> Why does someone need to spot check the training data against the output?

because one wants to make sure ones program is operating as intended?

The surprising thing about all of this is that sentiment analysis--machine learning algorithms that can interpret the feeling you're trying to convey with your words--is a fairly solved problem.

These companies can easily put a filter over this engagement maximization algorithm and they are choosing not to.

But why would they do it? They have some pretty solid data to support the argument that more engagement, whatever the quality, equals more money to the poster, and more money to the platform. There is no bad publicity. The question is what can we do about it? It exploits basic human psychology.

Of course, machine can read the commenters' minds, this is a solved problem.

I, for one, welcome our mechanical overlords.

---

https://text2data.com/Demo

This document is: positive (+0.52) Magnitude: 1.39

Sorry, can you provide some citations to "sentiment analysis being a solved problem". Last time I looked, it was decidely non-trivial, though I admit I haven't followed the topic in a few years.

And it is worth noting that if these companies could stop this with limited engagement impact, they totally would as it would get them out of the horrible political hole they are in right now.

Mark Zuckerberg is the 21st century's Pablo Escobar.

Because he runs a vast, violent empire that smuggles illegal drugs?

Probably because the parent dislikes Zuck, and the Hitler comparison has gotten too overused in recent times (and Zuck isnt a politician), so Escobar was chosen as an appropriate one. Fully expecting to see that comparison being used against Jeff Bezos soon.

tl;dr: “everyone i dislike is a nazi or something close” syndrome

Reply on news.ycombinator.com