Recommendations — Let me help you help me

Heiko Maiwand
4 min readAug 22, 2017

I love Spotify. I love Netflix. I am a user of services like these because I enjoy access to their massive catalog of content. However, this scale creates the issue that I am lost amongst so much choice.

To address this, most of these services have developed sophisticated recommendation systems, built on deriving user preferences from a simple thumbs up or thumbs down interaction. I imagine they use many implicit factors as well, such as: “How often do you play this” or “What do your friends like?”

Services like these cross-reference the aggregate — meaning they figure out what people with your preferences and patterns usually consume. They combine this with the individualized implicit information mentioned above, and the ‘likes’ and ‘dislikes’ they get.

These recommendations are useful and impressive. My Spotify’s Discover Weekly playlist finds me some gems, and YouTube quite consistently finds me relevant content. However, often these recommendations still leave a lot to be desired.

The solution

I am proposing how a service can grow their knowledge about an individual to a very nuanced level to make recommendations much more personal.

This fundamentally comes down to understanding why someone likes or dislikes a piece of content.

Long-pressing on thumbs up/down reveals the new feature

In the above animation (full resolution video available here), I used Spotify as an example to embody how this could be achieved. When someone long-presses the ‘thumbs down’ button, the interface reveals options to let the listener tell them why:

  • Is it being overplayed on the radio?
  • Does it just not fit the mood right now?
  • Remind you of a sad time in your life?

Same goes for ‘liking’ a song:

  • Does the song remind you of a happy, nostalgic time?
  • Do you like the message in the lyrics?
  • Do you go to every concert the band puts on?

If a recommendation algorithm had access to this level of detail, I can only imagine what kind of customer understanding it could derive. The song that you overplayed now to the point of exhaustion will probably be a great song to remind you of this point in your life a few years from now. Maybe there is a certain genre that you listen to only to fall asleep to, that you don’t need mixed in your Daily Mix. Either way, it is much more useful than just a binary like/dislike.

UI Details

For this exercise, I used their existing interaction paradigms. It borrows from their existing ‘long-press to preview’ feature. (Note: this feature in Spotify has been deprecated since publishing this article)
It also still retains the simple one-tap to easily ‘like’ or ‘dislike’ but allows users to give richer information in a non-intrusive way.

Even though the long-press is an existing action within Spotify, I would like to see some extra learnability added into it so it doesn’t remain a feature only for power-users. Using the existing pop-up shown after ‘liking’ a song can be repurposed like this (Video available here):

The simple action gives users a clue about the extended functionality

Today, when listening to the Discover Weekly playlist in Spotify, the ‘thumbs up’ or ‘thumbs down’ buttons are superseded by the ‘shuffle’/‘repeat’ buttons. I propose that understanding customers’ preferences when listening to generated playlists and radio stations is more valuable. For generated playlists, the ‘shuffle’/‘repeat’ buttons can be moved to the ‘more options’ drawer (as shown below). User-generated playlists would still retain the current layout to emphasize this priority.

Relocate ‘shuffle’/‘repeat’ for generated playlists and stations

Conclusion

This concept was done within the confines of the existing Spotify UI, but I believe this overall concept would be very valuable across many different services: YouTube, Netflix, Podcasts, etc. Giving customers a voice would be invaluable to deeply understand why and when to promote content. This becomes infinitely more powerful when trends are observed across the aggregate. Ultimately, I want to enhance the power of the recommendation engines.

What do you think?

I’d love to hear thoughts and comments. It would also be interesting to hear if there are already services that have a similar mechanism and how they implemented it.

The concept, animations, icons, and images were created by Heiko Maiwand

Edit: I have iterated on the title a couple of times because it was originally over-critical of Spotify. This was not my intention. I know that the secret weapon of any content provider is data about its usage, and I wanted to illuminate a missed opportunity. What will set services apart — aside from content — is customer understanding and recommendations.

--

--