Hacker News new | past | comments | ask | show | jobs | submit login
NYT obtains internal TikTok document describing recommendation algorithm (nytimes.com)
39 points by perihelions on Dec 6, 2021 | hide | past | favorite | 10 comments



> the TikTok Algorithm Knew My Sexuality Better Than I Did,” reads one in a series of headlines about people marveling at the app’s X-ray of their inner lives.

I think its disingenuous to describe it like that. The concept is so simple that it’s absolutely mundane. The app isn’t X-raying your life. It’s just measuring how long you look at a video, and if you press like or comment. It’s brain dead simple, and the only reason so many are astonished by it is because they assume that spending 20 min looking at half naked people of one or the other gender won’t affect the suggestions “because I didn’t press like or comment” but obviously it can still tell that after being shown the same content for 20 min straight, your still not scrolling by, and will keep suggesting it to you because of that.


Regardless of whether they are biased or not, and whether they are positive or negative, I think it’s unethical for the NYT to publish articles regarding the social media industry without disclosing their multi-million dollar annual payments from Facebook/Meta.


I don't really know what you are talking about, could you expand on that or link to a source?


Facebook pays publishers to use their content.


Is X here a "cross product", as used in other systems like Sibyl and TFX?

Plike X Vlike + Pcomment X Vcomment + Eplaytime X Vplaytime + Pplay X Vplay

If so this whole algorithm and what's being done are pretty standard fare, the entire industry works to maximize <metric> (often watch time). These companies are paperclip maximizers and don't have ethics or moral systems per se; they are trying to maximize metric to make $.


The usual companies may not have ethics as you write, but politically motivated company can easily embed "ethics" in a recommendation engine like that.


The recommendation/selection algorithm isn't their "secret sauce" but rather their content identification and classification systems. Any algorithm is useless without good data and only having one part of it (like user watch time and interactions) makes for some rather boring results, but if you can categorize the content perfectly you've struck gold.


There's no redeemable "user value" to addictive videos. It's monopolization of attention with a drug.



Surprisingly light on detail and new information.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: