search

irecsys / CARSKit

Java-Based Context-aware Recommendation Library
https://carskit.github.io/
GNU General Public License v3.0
124 stars 53 forks source link

Using CARSKit with implicit feedback #10

Closed MatthiasKirsch closed 5 years ago

MatthiasKirsch commented 6 years ago

Hi :)

I am using CARSKit with BPR for top-N recommendation. My dataset is purchase data so I have only implicit feedback (more precisely I just know what the user bought). I added some cool stuff like weather information and demographic stuff about the cities of the stores etc.

When using CARSKit I only applied BPR with Pre-filtering techniques (User-, item-, uisplitting) so far because of the implicit feedback.

  1. Is there another approach I should try out? For example CSLIM?
  2. Is it possible to use a hybrid filtering approach like DCR/DCW/BPSO in combination with BPR?

Many thanks in advance!! :)

irecsys commented 6 years ago

Sounds good.

But unfortunately, we do not know which algorithm will perform the best by given a data. You have to try multiple algorithms to find the best one.

Based on my experience, if you'd like to see a good results on top-N recommendations, you can try SLIM based approach or CAMF_ICS, CAMF_MCS, CAMF_LCS. For SLIM-based approach, you need to carefully to find the optimal parameters. CAMF_ICS works good in most cases, CAMF_MCS can work even better if you carefully tune up the parameters.

In terms of the hybrid filtering approach like DCR/DCW/BPSO, you cannot try BPR, they were built based on the user-based collaborative filtering.

MatthiasKirsch commented 6 years ago

Thank you very much for your response!

My problem is that I have implicit feedback so my rating matrix has only values 0 and 1. Because of this I cannot use algorithms like CAMF and some other context-aware algorithms, I think.

I read about that CSLIM might be able to handle this type of implicit feedback. Do you have experience with such cases?

Many thanks in advance :)

irecsys commented 6 years ago

I do not think so. SLIM is another item-based collaborative filtering, actually.

On Fri, Oct 13, 2017 at 2:09 AM, MatthiasKirsch notifications@github.com wrote:

Thank you very much for your response!

My problem is that I have implicit feedback so my rating matrix has only values 0 and 1. Because of this I cannot use algorithms like CAMF and some other context-aware algorithms, I think.

I read about that CSLIM might be able to handle this type of implicit feedback. Do you have experience with such cases?

Many thanks in advance :)

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/irecsys/CARSKit/issues/10#issuecomment-336369496, or mute the thread https://github.com/notifications/unsubscribe-auth/AHDB5xIsKvRKChGibhFP_89PGhZDcRwRks5srwymgaJpZM4P3dE9 .