Kasra Rahjerdi
@jc4p
every single morning when i open the app this person or their sister is at the very top of my feed. every single day. i wonder what’s going on with the algorithm that causes this
44 replies
5 recasts
104 reactions
triumph
@triumph
if an app has an algo i kinda feel they should probably have a data engineer or two it…does not feel like the case here
1 reply
0 recast
18 reactions
Kasra Rahjerdi
@jc4p
imo the root issue is all their algorithms are supervised (they label and tag things and give them scores and adjust) vs what everyone else has been doing for 10+ years which is unsupervised (the system learns the labels and rules) — the current approach 1. doesn’t scale 2. doesn’t work with other languages
3 replies
0 recast
12 reactions
akshaan
@akshaan
I think you're conflating supervised vs. unsupervised training with manual vs. learned models. We used supervised training, because the model needs feedback from users to know what they like. We don't use manual heuristics, but instead train models that infer what users like from the feedback they're given. This does scale because they feedback is generated when users click, like or otherwise engage with content. It is not generated by us manually labeling things.
1 reply
0 recast
1 reaction
Kasra Rahjerdi
@jc4p
appreciate the response and totally acknowledge the oversimplification! my issue (which i have no way of knowing it's a wild guess) is that the models are trained on global (and possibly handpicked) features, the vibe i get is there isn't clustering and then each cluster gets its own features
1 reply
0 recast
0 reaction
akshaan
@akshaan
the features are personalized to each user (so not global) but are hand-picked. The model does some implicit clustering based on the features we feed it, but we don't have explicit clustering of topics or users today. Something we're thinking about actively rn.
1 reply
0 recast
3 reactions
