Dumb question: how do ratings on a scale of 1-5 map to positive examples?

[zachmayer]

Is any item a user gave a rating considered a positive example? Or only items they rated as a 5?

[maciejkula]

I think in this case you have to decide it yourself: what rating counts as positive? It's common to say that any rating higher than some threshold counts as positive; what that is probably depends on your use case.

On 29 Aug 2017 20:02, "Zach Mayer" notifications@github.com wrote:

Is any item a user gave a rating considered a positive example? Or only items they rated as a 5?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/maciejkula/triplet_recommendations_keras/issues/12, or mute the thread https://github.com/notifications/unsubscribe-auth/ACSCAytCG0a-rQIUYEykiF_umjHTkEdQks5sdFJOgaJpZM4PGVJN .

[zachmayer]

And then would I throw away the known, negative ratings that are below that threshold?

[maciejkula]

Yes! You could also model rating and implicit feedback jointly, but I haven't implemented it here.

On 29 Aug 2017 21:04, "Zach Mayer" notifications@github.com wrote:

And then would I throw away the known, negative ratings that are below that threshold?

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/maciejkula/triplet_recommendations_keras/issues/12#issuecomment-325763944, or mute the thread https://github.com/notifications/unsubscribe-auth/ACSCA9QovUEl1hK-Iutk70w9TsbjxFLuks5sdGDTgaJpZM4PGVJN .

[zachmayer]

Thanks! Any ideas for modeling them jointly? One thing I've tried is treating it as a multiclass problem with 6 classes (class 0 = no rating, class 1-5 correspond to ratings).