search

dsmilab / WSDM2018-KKBox-MRC

:musical_score: Predict the chances of a user listening to a song repetitively after the first observable listening event within a time window was triggered.
https://www.kaggle.com/c/kkbox-music-recommendation-challenge
GNU General Public License v3.0
7 stars 0 forks source link

Combine CF method #3

Open howeverforever opened 6 years ago

howeverforever commented 6 years ago

I don't know how useful it is. But we need a try.

ctjoy commented 6 years ago

In my case, using CF is still useful. It improved my score about 0.02% in the test dataset. On the other hand, it improved the AUC score 0.71 to 0.84 in the validation dataset. I think the main problem on this issue is cold start. You can see further information on this thread: https://www.kaggle.com/c/kkbox-music-recommendation-challenge/discussion/42795