4 ms·

> Your argument is basically "only group X dislikes Y" and it's based on an anecdote. It's very unfortunate that you (and presumably others) read it like that.

by bsnnkv 3y ago

> Your argument is basically "only group X dislikes Y" and it's based on an anecdote. It's very unfortunate that you (and presumably others) read it like that. I am not making an argument or a generalized statement in my post. If in doubt, you can also check my post history to see that I do not have a history of making any such posts. I am expressing curiosity about how the recommendation algorithm might be tailoring content not just based on interests but on the inferred ethnic backgrounds of people (because interests and ethnic backgrounds are not correlated with each other). I'm open to feedback on how this could have been phrased differently.

vorpalhex 3y ago
It's the entire premise that's flawed. As a basic rule, you can't use anecdotes to draw assumptions about wide group trends. That my particular Korean friends like the cheesesteak place up the street says nothing about Koreans, cheesesteaks or that restaurant. This is because at a basic level my friend group just doesn't look like any kind of general population. If nothing else it's geographically restricted, a tiny subset of classes and interests - and likely indirectly education levels and income. > because interests and ethnic backgrounds are not correlated with each other They totally are, speaking for populations! This is how de-anonymization typically works. They also correlate with class and income. And yes, if you target anime, you'll select the Japanese and if you target Webtoons you'll select mostly Korean and Chinese. Your flaw here is not your language but the direction of your reasoning. You can't reason from the specific anecdata up to population level - "my Black friends like tiktok therefore tiktok must target Black people". That's a "the parking lot is wet, it must be raining" sort of statement. You CAN reason down. "Tiktok really does a good job showing content from Black influencers, they probably have a large Black audience".
- bsnnkv 3y ago
  > As a basic rule, you can't use anecdotes to draw assumptions about wide group trends. I think you missed the point again, because my post was not trying to use anecdotes to draw assumptions about wide group trends. It was an expression of curiosity and a desire to see data, which at least in my mind is quite opposed to drawing assumptions (or making an argument, or a generalized statement as you had previously written). Nevertheless, this is an interesting post with some good information and I appreciate the time you took to write it out. > They totally are, speaking for populations! This is how de-anonymization typically works. They also correlate with class and income. This part in particular was very informative; I'll be doing some more reading on this thanks to your explanation.

by bsnnkv 3y ago

vorpalhex 3y ago

It's the entire premise that's flawed. As a basic rule, you can't use anecdotes to draw assumptions about wide group trends. That my particular Korean friends like the cheesesteak place up the street says nothing about Koreans, cheesesteaks or that restaurant. This is because at a basic level my friend group just doesn't look like any kind of general population. If nothing else it's geographically restricted, a tiny subset of classes and interests - and likely indirectly education levels and income. > because interests and ethnic backgrounds are not correlated with each other They totally are, speaking for populations! This is how de-anonymization typically works. They also correlate with class and income. And yes, if you target anime, you'll select the Japanese and if you target Webtoons you'll select mostly Korean and Chinese. Your flaw here is not your language but the direction of your reasoning. You can't reason from the specific anecdata up to population level - "my Black friends like tiktok therefore tiktok must target Black people". That's a "the parking lot is wet, it must be raining" sort of statement. You CAN reason down. "Tiktok really does a good job showing content from Black influencers, they probably have a large Black audience".

bsnnkv 3y ago
> As a basic rule, you can't use anecdotes to draw assumptions about wide group trends. I think you missed the point again, because my post was not trying to use anecdotes to draw assumptions about wide group trends. It was an expression of curiosity and a desire to see data, which at least in my mind is quite opposed to drawing assumptions (or making an argument, or a generalized statement as you had previously written). Nevertheless, this is an interesting post with some good information and I appreciate the time you took to write it out. > They totally are, speaking for populations! This is how de-anonymization typically works. They also correlate with class and income. This part in particular was very informative; I'll be doing some more reading on this thanks to your explanation.