3 ms·

Wow, that is horrible. And equally horrible is the fact that you don't seem to find this problematic at all. If the features were/are regularly used is not real

Wow, that is horrible. And equally horrible is the fact that you don't seem to find this problematic at all. If the features were/are regularly used is not really relevant: "algorithmically selected" implies a certain non-bias in the process, but if this is not how it really works then one should be much much more skeptical about the validity of any "trending" topics.

emn13 10y ago
It's a common misconception that "algorithmically selected" implies a certain non-bias. Perhaps this myth stems from examples used to explain simple algorithms (e.g. drawing straws, or pulling names from a hat). Complex algorithms certainly do exhibit bias, and not only that - it's intentional. The whole point of a "trending" topic is to discriminate (i.e. inject bias) trendy topics from less trendy topics. So you're talking about only adding the bias you want it to add - but I hope you can imagine that's an impossible cause to get 100% correct. There's no rigorous definition of trending; nor is there a rigorous definition of the topics to be selected from, nor are lots of other matters here rigorously defined. There's no way you're going to do better than some fuzzy algorithms trained on real data. Critically, that tends to mean real data generated by... biased humans.
- _Understated_ 10y ago
  I would have thought that the algorithm would be entirely agnostic and based on incoming data only without humans to distort it, e.g. I share details with a dozen friends about aliens landing in Central Park, they then share it on and on. Facebooks algorithm will spot this and also others who did the same and then say "Hey, looks like there is a trend growing about aliens in Central Park" and from there would act upon it based on that data only, e.g. this topic was talked about 5 million times in the last minute and therefore bump it up the list to its corresponding position based on other "trending" topics. Having humans involved will distort that process, thus meaning true trending does not exist. Well, on Facebook at least.

emn13 10y ago

It's a common misconception that "algorithmically selected" implies a certain non-bias. Perhaps this myth stems from examples used to explain simple algorithms (e.g. drawing straws, or pulling names from a hat). Complex algorithms certainly do exhibit bias, and not only that - it's intentional. The whole point of a "trending" topic is to discriminate (i.e. inject bias) trendy topics from less trendy topics. So you're talking about only adding the bias you want it to add - but I hope you can imagine that's an impossible cause to get 100% correct. There's no rigorous definition of trending; nor is there a rigorous definition of the topics to be selected from, nor are lots of other matters here rigorously defined. There's no way you're going to do better than some fuzzy algorithms trained on real data. Critically, that tends to mean real data generated by... biased humans.

_Understated_ 10y ago
I would have thought that the algorithm would be entirely agnostic and based on incoming data only without humans to distort it, e.g. I share details with a dozen friends about aliens landing in Central Park, they then share it on and on. Facebooks algorithm will spot this and also others who did the same and then say "Hey, looks like there is a trend growing about aliens in Central Park" and from there would act upon it based on that data only, e.g. this topic was talked about 5 million times in the last minute and therefore bump it up the list to its corresponding position based on other "trending" topics. Having humans involved will distort that process, thus meaning true trending does not exist. Well, on Facebook at least.