Weekly Jams recycles unlistened suggestions (with figure!)

biocv · February 3, 2024, 2:14pm

Since I have figured out how to resolve the Weekly Jams to my local collection, I have faithfully been downloading the playlists. The rate at which I was able to resolve suggestions to my collection was pretty good at first (~90%) but has been dropping steadily over the past weeks to roughly ~50%:

The reason is that Weekly Jams retains suggestions I have been unable to resolve. This causes the list to gradually clog up with unlistened suggestions. Of course, there are several hacks to work around that, but I feel that the Weekly Jams feature would be greatly improved if it were to offer all fresh suggestions each week. This should be easy to implement by introducing some cooldown period or a penalty for recent suggestions.

Thanks!

rob · February 5, 2024, 1:57pm

Hi!

Two thoughts:

Such a cool-down feature is not as easy to implement as you suggest.
We need to look at why this is happening before we dream up a solution. My take on the situation is that we can generate a finite number of recommendations given the number of people in our community right now. And if you continually listen to your weekly jams (I listened to my daily jams each day for months!) you’ll notice exactly what you are suggesting – over time things get less interesting and less on point because you’re reaching further and further down the list of recommended tracks. Putting out all-new fresh suggestions each week would accelerate the pace at which things get worse.

From where I sit there are no easy fixes for this problem – the best fix is to get more people sending their listens – ideally people who do not listen to the same music we are already listening to.

The team currently doesn’t have the bandwidth to examine these issues and make improvements here – certainly not in the next 6 months. If you’d like to dive and make see if you can improve things, please do:

listenbrainz-server/listenbrainz_spark/recommendations at master · metabrainz/listenbrainz-server · GitHub

and

github.com

metabrainz/troi-recommendation-playground/blob/main/troi/patches/periodic_jams.py

from datetime import datetime, timedelta

import troi.filters
import troi.listenbrainz.feedback
import troi.listenbrainz.listens
import troi.listenbrainz.recs
import troi.musicbrainz.recording_lookup
from troi import Playlist
from troi.playlist import PlaylistMakerElement

DAYS_OF_RECENT_LISTENS_TO_EXCLUDE = 60  # Exclude tracks listened in last X days from the daily jams playlist
DAILY_JAMS_MIN_RECORDINGS = 25  # the minimum number of recordings we aspire to have in a daily jam, this is not a hard limit
BATCH_SIZE_RECS = 100  # the number of recommendations fetched in 1 go
MAX_RECS_LIMIT = 1000  # the maximum of recommendations available in LB

WEEKLY_JAMS_DESCRIPTION = """<p>The ListenBrainz Weekly Jams playlist aims to be a playlist of tracks that we believe
    that you'll like. Designed to be a review playlist, it suits well for playing whenever you need to have comfortable
    music to listen to that does not require active listening.</p>

    <p>The playlist contains tracks that you've listened to before and that our collaborative filtering algorithm believes

This file has been truncated. show original

I’d be happy to review your PRs!

biocv · February 5, 2024, 6:43pm

Not sure I am following. The issue is not that the recommendations are not interesting, but that I appear to be increasingly unable to resolve them to my collection. Are you saying that if more recommendations would be generated the unresolvable tracks would be purged?

Nor did I expect them to. I just wanted to give feedback in the hope that it will help you guys develop the various LB features. I have gotten a lot of good stuff out of LB.

I am wondering where I gave you the impression I am a programmer . As it happens I do have some (very modest) coding skills in python so I am happy to take a look. I am rather curious to see how the recommendation engine actually works TBH.