An example of SVD++ for implicit dataset feedback?

[echatzikyriakidis]

How SVD++ can work with implicit feeback?

An example would be perfect.

So far I was using the implicit library for implicit CF.

Does anyone has an example to share?

Thank you

[NicolasHug]

SVD++ uses the existence of a rating (whatever its actual value) as implicit feedback.

It can be extended to support "native" implicit ratings but this is not supported in surprise. Details may be found in the original paper(s) which are linked in the docs.

[echatzikyriakidis]

Hi,

I have only interactions of users and items.

No expicit ratings. No negative feedback.

I can create two dataset versions:

1) userId, itemId, 1

2) userId, itemId, number_of_times_the_interaction_happened

Will it work with any of the above approaches?

Some implicit algorithms (like ALS) set by default 0 to all interactions that haven't exist. This is how the implicitlib works. Will SVD++ will work also that way?

Thank you.

[NicolasHug]

There are extensions of SVD to handle pure implicit feedbacks but they're not implemented in surprise. You'll have better luck with other libraries like implicit which are specifically designed for these use-cases.

[echatzikyriakidis]

Thank you @NicolasHug !

I will go with implicit.

By the way implicit is a great library for implicit datasets but it does not have the out of the box candies you have:

Cross validation, GridSearch, etc.

You need to implement them manually.

Stay safe!