Users of popular streaming services invariably encounter their recommendation systems. YouTube suggests new videos to watch, Netflix recommends movies and series tailored to the viewer's preferences, and Spotify plays precisely the tunes you'd feel like dancing to.
But how do these systems genuinely discern what will appeal to each of their users at a given moment? Special recommendation mechanisms are deployed, which bear a striking resemblance regardless of the specific streaming service in use at that time.
User Activity History. For platforms like YouTube, Netflix, or Spotify to aptly determine which content would appeal to a specific user, they gather the user's activity history. This includes a list of viewed videos or movies and listened tracks.
Activity history is indeed of paramount importance. If you've been watching a plethora of travel videos on YouTube lately, it will inevitably suggest videos on new cities and countries. The same applies to music genres on Spotify or movies and series on Netflix.
For YouTube, the time spent on a particular video also matters. If you watched a lengthy video to the end, it's a positive signal for the system—it will suggest more of such content. Conversely, if you stopped the video in the early minutes, it indicates that similar videos should be recommended less frequently.
User Interaction with the Platform. By user interaction with the platform, we refer to likes and dislikes given to specific content, comments (if they are facilitated for content on the platform), and the addition of videos, movies, series, or music to the favourites list.
The favourites list feature is prevalent in the majority of streaming services, including YouTube, Netflix, and Spotify. When specific videos, movies, series, or tracks are added to the liked list, it significantly influences the recommendation system.
Spotify also pays close attention to the playlists that a user creates. Tracks added to these playlists are considered high priority. This is also true for radio stations that are based on a specific track; they serve as an excellent gauge of a user's musical taste.
Similar Traits of Preferred Content. Popular streaming services also deeply analyze additional information about the content that a user prefers. This pertains not only to genres or artists but also to tags and other specific characteristics.
Did you watch several thrillers in a row on Netflix? The streaming service's recommendation system will undoubtedly notice and begin suggesting movies and series in the same genre. This also applies to preferences concerning actors, directors, and other content production contributors.
Spotify takes it a step further. It considers not just genres or artists but also the tempo of specific tracks in BPM (beats per minute). Moreover, the streaming service's algorithms can even analyse the musical instruments used in liked compositions.
Similar Traits in User Groups. Another crucial aspect of recommendation systems of modern streaming services is the analysis of interests within user groups. This mechanism nicely complements the others mentioned earlier in the text.
If several users, grouped by shared interests, like the same content from a hypothetical category one, they are highly likely to also appreciate videos, movies, series, or music from another category.
For instance, if multiple Spotify users from the same electronic music enthusiasts group begin actively listening to country music, the remaining users will also see this genre in their recommendations. This applies to content on YouTube, Netflix, and other platforms as well.
Of course, in reality, recommendation systems employ many more mechanisms and even artificial intelligence. But most of them are kept under wraps. All this aims to make the user experience of any service as comfortable and engaging as possible.