It’s always funny to me when someone talks about how awesome the tech behind recommender-systems is and what complex problems had to be solved to make it work but in the end it’s still just absolute garbage.
AI/ML covers a ton of algorithms, some of them are that boring, some of them aren’t.
Re above. Take all users who viewed all items. Run a MapReduce to segregate them into pairs. Calculate the frequency of pairs and store the result. That clear enough for you?
Reducing the computational cost is what makes it complex… but why am I even discussing this here anyway, I was mocking the topic in the first place. Your disregard of the problems in the details is kinda amusing though, because that’s probably the reason most recommender engines are as crap as they are.
It’s always funny to me when someone talks about how awesome the tech behind recommender-systems is and what complex problems had to be solved to make it work but in the end it’s still just absolute garbage.
It’s not really that interesting, you find hot spots where interest between items is correlated.
yeah similarly to AI right? Also not really interesting, you just do some math and boom: AI!
AI/ML covers a ton of algorithms, some of them are that boring, some of them aren’t.
Re above. Take all users who viewed all items. Run a MapReduce to segregate them into pairs. Calculate the frequency of pairs and store the result. That clear enough for you?
Reducing the computational cost is what makes it complex… but why am I even discussing this here anyway, I was mocking the topic in the first place. Your disregard of the problems in the details is kinda amusing though, because that’s probably the reason most recommender engines are as crap as they are.
Well, there’s problems that are complex at the center, and there’s problems that aren’t. This one isn’t.