-
Task
-
Resolution: Fixed
-
Normal
-
None
-
None
-
None
We want to import recommendation feedback to spark cluster to use it in generating better recommendations for the next batch:
- The recommendation feedback should be stored in parquets at /recommendation/feedback/<year>/<month>/<day>/data.parquet
/recommendation/feedback/2021/02/27/data.parquet
which means there should be a folder for every year inside /recommendation/feedback and each folder should contain 365 files, one for each day
2. Since the recordings in recommendation feedback (Lemmy) have already been mapped to the msid-mbid mapping, the following keys in the dump json will suffice
mb_recording_mbid, user_name, feedback, rated_at
The task is only for integrating recommendation feedback into the process of generating recommendations, I am yet to think about listen feedback since it uses recording msids and uses integral feedback (-1, 1).
Done when:
feedback dumps are created and hosted on FTP