A note to the team designing recommendations on google ads:
Just because I search for deep learning concepts for personal learning, does not mean I will be purchasing every paid online data science course on the planet.

But you might purchase one, so there obviously must be a competition to ensure you know about all of them so you don't miss one out of ignorance while choosing. 😉

Google does this all the time. Especially with YouTube recommendations. I watch one video on Jenkins because I'm trying to figure out how to use a poorly documented plugin, and now my recommended videos are almost all about Jenkins.

Oh boy, what a perfect time to link this blog:

https://apenwarr.ca/log/20190201
"Forget privacy: you're terrible at targeting anyway"

@EmberQuill This is one of YouTube's most annoying algorithm "features." It clearly is over-trained to offer similar video recommendations, irrespective of any particular user's viewing habits, meaning that any side-trip out of curiosity or for troubleshooting/research renders the recommendations almost unusable for days.

I had a binge session for a couple days of a particular type of time-wasting video, and it took weeks of telling YouTube that I wasn't interested on every single recommendation before the videos were purged from my feed.

We forget that the algorithm exists to serve Google (maximize watch time) rather than serve us (give us what we actually want, not just stuff similar to what we have already watched), and it's things like this that remind us.

@powerfulparadox ugh, yeah, I had the same sort of problem. I watched some of those awful "robotic voice reads funny reddit posts for people who can't be bothered to read themselves" videos and it took several weeks to stop recommending thrm, even after hiding every single recommended video multiple times in a row.

YouTube has over ten years of data on my watching habits so you'd think they would be better at catching outliers at this point.

I also have the problem with Amazon recommending things after I've bought one and most likely don't need more. Like YouTube, recommendations seem to be based almost entirely on the last few weeks of data, with a couple of random recommendations based on things from five years ago.