Flow ignores genre balance and gets stuck on one genre

I've been using Deezer as my main music streaming service for over a year. I transferred my entire library, spent a lot of time liking and disliking tracks to train the recommendation system, and tried to help it understand my tastes. My goal wasn't even to get perfect recommendations — I just wanted it to occasionally suggest phonk alongside the rest of my music.

The problem is that Deezer now recommends phonk in about 90% of cases, even though only around 10% of my favorite tracks are phonk. It feels like the algorithm has overcorrected and is ignoring the overall balance of my listening habits.

Almost everyone I know who has used Deezer says the recommendations are much worse than those from Spotify and especially Yandex Music, and unfortunately I completely agree. If Deezer wants to gain a meaningful share of the market, improving recommendations should be a much higher priority.

There are only two reasons why I still use Deezer:

1. Flow is a great feature.
2. Deezer is included in my mobile carrier plan.

Otherwise, I would probably switch to Spotify. Spotify at least has track-based recommendations, while Deezer's Track Mix feature has been broken for a long time (see: https://en.deezercommunity.com/features-feedback-44/track-mix-is-empty-81318?tid=81318&postid=264734#post264734).

Please invest in recommendation quality and hire people who can build a system that consistently surfaces relevant music based on a user's complete listening profile and all of their likes, rather than overfocusing on a single genre.

For inspiration, I would point to Yandex Music. Its catalog is more limited, but its recommendation engine is excellent. One of its strengths is finding users with similar listening habits and recommending tracks that are popular among those listeners.

The recently added feature that allows users to exclude genres does not solve this problem. Excluding a genre entirely is too extreme. What would be much more useful is a set of preference sliders that let users control genre balance — allowing them to hear some genres more often and others less often, without removing them completely from recommendations.