I love the plethora of content on Spotify and Netflix, but sometimes get lost with so many choices.
Most of these products are equipped with a mature personalized recommendation mechanism, and users' personal preferences can be deduced by simply clicking on the simple interaction of likes and dislikes, playback frequency and friend information.
In addition to individual users, the system also aggregates and analyzes what people with similar preferences generally like.
These recommendations are very helpful, my weekly Spotify playlist finds some great music recommendations for me, and YouTube is constantly pushing relevant content for me. However, there is still a lot of room for improvement in these personalized recommendation services.
Solution
I suggest that personalized recommendations can obtain more detailed information, so as to make recommendations more personalized.
The key here is to understand the reasons behind what people like or dislike.
Long press like/dislike to select the reason for the operation
In the animation above, I used Spotify as an example to demonstrate the implementation. When long-pressing the "dislike" button, the interface displays options for the user to choose a reason:
Is it playing too many times?
Is it not suitable for the current scene atmosphere?
Reminds you of a bad meeting?
The same goes for "likes":
Reminds you of a happy memory?
Like the meaning of the lyrics?
Did you go to all the concerts of this band?
If the recommendation algorithm can achieve this level of granularity, I can imagine how deeply it will understand the user. For example, a song on your most recent single loop will remind you of this time in the future. Maybe there's a type of music you only listen to when you're falling asleep and don't mix it up in your daily life. This is far more useful than a pure binary like/dislike.
interface details
In this case, I used the existing interactive form of the product and borrowed the "long press preview" function for reference. Allows users to provide more information in a non-intrusive way while retaining the simple action of clicking "like"/"dislike".
Although long-pressing is a Spotify-existing operation, I wanted to provide a little guidance so that this feature isn't limited to advanced users. If you use the existing prompt popup, after clicking "Like", you will see the following:
Simple operation guide
On the current play page of Spotify's weekly playlist discovery, the "Like" and "Dislike" actions are replaced by "Shuffle" and "Loop". But I think it's more important to understand user preferences when listening to recommended playlists and stations. So when playing recommended playlists and radio stations, the playback order can be moved to the "More Actions" menu. For the play page of the user-created playlist, the current form can still be maintained.
At present, the buttons on both sides of Spotify's playback controls are "Shuffle" and "Loop"
Try putting "Shuffle" and "Repeat" in the "More Actions" menu
Summarize
Although this case is limited by the existing b2b data Spotify framework, it is believed that the theoretical views behind it are applicable to many content push platforms such as YouTube, Netflix, and Podcasts. To gain a deep understanding of the timing and reasons for personalized recommendations, user voices are also invaluable. This becomes even more important when looking at trends through big data. In any case, I hope these ideas will play a role in improving recommendation algorithms.
——
Z Yuhan: My translation of the article does not necessarily mean that I absolutely agree with the original author's point of view, and the original author does not necessarily believe that it is absolutely correct when he puts forward a point of view. There are still many possibilities for personalized recommendation in the future. This article proposes a hypothesis. Whether you agree with it or not, welcome to discuss it in the comment area.