Mean normalization and movie rating logic

Course Q&A Machine Learning Specialization Unsupervised Learning, Recommenders, Reinforcement
1

Hello everyone,

This is perhaps a philosophical/semantics question but I was just wondering about this whole approach of designating ratings for someone who had not provided any ratings before (as discussed in the video).

More specifically, I can understand the mathematical footwork to provide starting guesses for Eve in the example provided.
However, is it not a stretch to assume that we have sufficient intuition about someone when there is absolutely no record of any rating? I would hate being bunched up with the average when my movie tastes might be very different to the mean.
Perhaps this is really just a starting point when you have nothing else to go on but this assumption still leaves me uncertain.

What to you think?

2

Agreed.

So when collaborative filtering can’t give us any better insights into one’s perference, we might want to go for the content-based recommendation option that will make use of any initial knowledge about the user. However, there is always a limit.