Comment by basch
9 years ago
>"hurt user engagement"
if Facebook is personalizing everyones feed based on what they click, isnt the clicker to blame for the result of personalization, not the technology provider. it sounds like your friends didnt engage with your posts, they did engage with other posts, and the facebook feedback loop built them a filter bubble of things they liked more than your posts.
I blame Facebook. I've ranted on HN about what I consider to be an abuse of 'curation' in the past, but I think 'curation' is killing the internet.
I agree with the grandparent about Facebook killing engagement, but the place I noticed it first was Last.fm. I used to spend tons of time on Last.fm looking up music. I made sure every track I played was scrobbled, and I listened to their recommendations all of the time.
As time passed, the recommendations got worse. Mind you, the recommended songs were closer to what I was listening to, but I appreciated when Last.fm used to recommend music that was somewhat off the wall. If it was a draft, I guess you could call that kind of recommendation a flier.
I personally blame machine learning. The recommendations I get in Last.fm now sound a lot like music I'm already listening to. Not only is it boring, but it's stupid, in the sense that finding music that sounds almost exactly like music you already like is easy.
I want content to be curated the way a friend would curate it for me. For instance, I had a friend recommend the band Sleigh Bells to me, because he knew I listened to punk all of the time, but that I really loved pop (and noise). Last.fm has never recommended me anything remotely as great as that.
I'm almost loathe to like anything on social media anymore, since it basically means I'm going to get similar posts shoved down my throat until I completely give up on the site.
Seems like you dislike bad curation specifically, rather than curation itself. After all, Last.fm was already curating the songs they recommended (it wasn't like they literally threw a random sample of all their catalog).
That said, I agree with you that this kind of "finding the closest match" model is quite silly for recommendation systems. And you can see the same problems with ad matching too.
>Seems like you dislike bad curation specifically, rather than curation itself.
Yeah, that's what I was trying to get at by putting 'curation' in single quotes. It's not really curation if you aren't specially choosing things. For instance, if I used Google to search for "bands that sound like [x]", nobody would claim the results are 'curated'.
These systems are basically search functions that return the most similar results that are distinct from the ones you already know about.
I think the fundamental issue is that Markov Chains are actually really terrible at curating content in the long term, but in short-term they are fantastic.
http://www.bbc.co.uk/blogs/adamcurtis/entries/78691781-c9b7-...
personally i somewhat agree with you. watercoolers like drudge report, techmeme, aldaily, redef show everyone the same page, not some regurgitation of their past. you experience new things by being shown something other than timehop.
If their algorithm causes such a positive feedback loop that only the most vapid of our content is seen I argue that is still Facebook's problem.
People already complain young people are superficial and self interested, algorithms feeding into/fueling that narrative sure doesn't help.
No. Which people and pages's posts I get to see because I follow them is up to me no matter how many times I click on their content. I am old enough to unsubscribe or ignore them by myself.
Facebook filter should be an opt-in.
In the simple case where they look at what your likes are to determine the bubble they create around you, sure. But I don't think that's all they look at. Content could be being buried simply because a friend of a friend doesn't like that content, but likes a bunch of other content you like. It's not as simple as a reaction to your action. There's a lot of "secret sauce" that goes into it that makes it not quite so transparent.
It could also be that they fail to understand the metrics. A topic rarely showing up gets fewer chances for the user to express interest.