0
$\begingroup$

I'm working on an hybrid music recommender system project, my goal is to create recommendation playlists in accordance with users tastes. I already implemented the first part which use a collaborative filtering algorithm, and I am now working on the content-based filtering part.

In order to make my recommender system accurate I read dozens of research papers talking about evaluating recommender system. There are many variables taken in account in those evaluations (such as coverage etc..)

Some of those papers talk about user confidence in recommender systems, defining it as one of the most important parameters to take in account. I read that for 10 items recommended if the user like something like 3 or 4 of them, it's enough to gain his confidence.

There is a point that i could'nt find in all those papers:
How many times can we recommend the same item to an user ?

To explain my question, during the playlists generation process there is a risk that the same music appear in two different playlists and I'm wondering what will be the impact on the users confidence in my recommender system in this case.

$\endgroup$
0
$\begingroup$

I don't think it matters if you recommend the same song in multiple playlists, just that your recommendations were accurate.

For example, imagine a user has two playlists a rock playlist and a classical playlist. If you recommended Crazy Train - Ozzy Osbourne in both playlists, you could theoretically have a "successful recommendation" in the rock playlist, and an bad recommendation in the country playlist.

The number of times a song appears is a design decision/variable - there is no right answer. You could potentially optimize on in the future using A/B testing (if I limit a song to being recommended x times, do those users then value my recommendations more than the general population?).

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.