from Flax_vert@feddit.uk to technology@lemmy.ml on 23 Jun 2024 23:39
https://feddit.uk/post/13743460
I think recommendation algorithms and advertising are separate things, however think with defaults and when it comes to what specific data is collected, where do you draw the line? Absolutely no recommendations at all based on an algorithm? Would you say using your ‘like’ history to recommend you more videos is okay? What about watch history, or save history?
Same question can also be asked about where you draw the line on advertising. Just say Youtube showed ads purely based on your video like history, would that be creepy?
I think we can all draw the line at location history, how long you linger on a post, etc. I’d like to know your thoughts on where you’d draw the line for both advertising and content recommendations. (This is two questions)
Sorry that this post is horribly formatted. I’m tired, acoustic and had a shower thought 😝
threaded - newest
If recommendations are being provided to me as a service and the algorithm that goes into it is relatively transparent, I have no issues.
If advertising is based on the value an advertiser sees in the product being advertised, I have no problem.
If I’m the product being sold or an ad distribution network is involved, I’ve got a problem.
I’m on the extreme end. Any kind of “recommendations” are an immediate and complete turn off for me. Not just the obvious stuff like “sponsored” posts or “algorithmic social media” feeds. I abhor and avoid even things like Spotify recommendations, which most people consider useful.
Whether they intend it or not, these engines are built to funnel you back into the lowest common denominator, most broadly appealing stuff, because that’s what the algorithm sees gets the most clicks from the average person. Sure, everyone likes oatmeal, but that’s because its bland and inoffensive.
I want to find my own shit through my own idiosyncratic process.
That’s not my general experience. Spotify for example is good at recommending me songs with less than 10k plays which I vibe on. I’ve discovered many smaller artists thanks to Spotify recommendations.
Same. If anything, it would be in Spotify’s best interest to steer you away from broadly appealing stuff because they’re the mega artists that probably negotiated a better deal. (I’m guessing here)
I’m going to be the odd one out on this.
I prefer ultra customized recommendations, I wish they were even smarter. Especially if I’ve already bought something, I want them to know so they stop advertising that product to me.
I’d rather see ads for products that I may actually buy rather than for shit I don’t have the slightest interest in.
I rarely buy products without significant research, so ads aren’t likely to trick me into buying something of poor quality. I just need to have awareness of things I don’t even know exist.
I agree. Given that I use Gmail, Google ought to know basically everything about me, so why do I keep getting ads for diamonds, instead of GPUs?
Why are they spying on me if they aren’t going to use that information?
Because the diamond guy wants to advertise to people like you, Google just gives them a check list of who they want to target and Google listens, if big diamond wants nerds to see diamond ads Google will take their money
Yeah for real. I have some very niche interests but get very obscure ads
One thing I don’t get is that e.g. I have Revolut installed, I use it regularly, Google knows this yet still half or a quarter of my ads for months now has still been Revolut. Why??
“Hey, it’s me, Amazon. I noticed you recently bought a fridge! Here’s five more fridges for you to look at, on case you need to complete your fridge selection!”
Like. Why?!
I don’t want either. If I miss out on something, life goes on. Half the fun is finding things out for myself.
Baseline requirement is that the algorithms should be open-source, user-customizable (filters), and subject to legal regulations (eg making sure they don’t cause increased depression and suicides in preteen girls)
Basically what Lemmy has (new, hot, controversial, etc)
I strongly support an opt-in model. You are given a list of ‘tags’ that denote content and that are most popular, and you can add them to your home feed. The unselected tags simply show the most popular posts of the most popular unselected tags.
That’s the system I think of at a glance, which I can already think of ways to game with bots, but I think there are likely much smarter folks out there who can work on solving those issues.
I prefer more granular control. I appreciate “controversial” and “sort by date” and “top post for past week/month/year”
For something I’m paying for, I want no ads, recommended or otherwise.
For something I get for free, if it’s easily skippable/ignorable, I don’t really care, I’ll skip it or mute the tab or whatever. If I can’t, I’d rather have a like sniper level targeted ad (use all the data!), really try to show me something I’ll care about (there was some like 10 minutes ad about the science behing glass by one of the guys from MythBusters, I watched the whole ad, it was great). The demographic level targeted ads are my 2nd least favorite, mostly because it feels like I usually need to suffer through what is a targeted ad but if they bothered to exclude some of the audience based on some data points (looking at you luxury car ads, it’s just never going to happen), they’d know I’m a bad target, I’d rather some generic add over those. My least favorite ad though, when I get an ad in a language I don’t even understand, like at least match my primary language, wasting everyone’s time…
Recommendation is part of the service. If they know I like something, then it’s reasonable they recommend me something that’s similar. It’s like going to a restaurant and asking for recommendations.
Advertising is when things are promoted outside the service. It’s like going to a restaurant and they tell me about Raid Shadow Legends. I don’t want that.
I think recommendation should be linked to usage data like watch history on that particular service. Location and other external information shouldn’t be used. I don’t want my recommendations depend on which friends I have or recent activity on a different service.
But the recommendation can make better inference if they have more data wouldn’t they? What should have happened is opt-in for a better recommendation.
<img alt="" src="https://lemmy.ml/pictrs/image/30febad7-ee7a-43d7-b83e-9b3eb2a6a876.jpeg">
comic by Nyazche
I draw the line right behind them. Where the real search results start.
I just would like to see the results of a recommendation algorithm that gives you something that it thinks you definitely won’t like, say, 20 percent of the time.
Because a lot of times in my endless scrolling I just end up with the same old drivel. Throw me something challenging occasionally, jeeeez.
This is what made YouTube rabbit holes fun a few years ago. Now we’re all in our own echo chambers.
Personally, I don’t like the idea of any recommendation or advertising algorithm using personal information of any kind. Though I can understand why location would be needed for advertising: in order to ensure ads for regional services are not shown outside of that region.
The kinds of data that I think should be used:
-Recommendations:
-Like history
-Watch (or view) history - specifically (and only) if I click on a post or watch more than some reasonable percentage of a video that would indicate I watched the entirety of the video.
-Advertisements:
-Location (based solely on IP)
-The content currently being viewed, based on a general categorization of the content. If I’m watching a video about technology I don’t need to be seeing ads for financial services.