Data 607 Discussion 11: Recommender System Analysis (Spotify)

##1. Scenario Design Analysis ###Viewpoint: The Customer (Parent & DS Student) Who are the users? A household with mixed listeners (adults and young children).

What are their goals? * To have distinct listening experiences: high-energy or nostalgic music for the parent (the “2007 vibe”), and educational/fun songs for the children.

To discover new music based on mood and genre rather than just repeating the same artist.

How does the system help/hinder these goals?

Hinders: The system fails to separate “child” and “adult” data, leading to an annoying mix of Meladze and children’s nursery rhymes.

Hinders: The algorithm is stuck in an “Artist Loop.” If I listen to one song by an artist, it forces the whole album on me instead of finding similar “vibe” songs (like moving from Meladze to Vintage or VIA Gra).

Viewpoint: The Organization (Spotify) Who are they? A global streaming giant competing for paid subscribers.

What are their goals? * Monetization: Force users into Premium by making the free tier unbearable (3 ads in a row).

Retention: Keep users inside the app by promoting podcasts and new releases.

How does the system help/hinder these goals?

Hinders: By over-pushing ads and providing poor recommendations, they are driving users away to competitors like Yandex Music or VK.

Helps: Using “Auto-play” within the same album keeps “engagement” metrics high, even if the user experience is poor.

##2. Reverse Engineering the System ###Based on the interface, Spotify seems to rely heavily on Item-to-Item Collaborative Filtering with a high weight on Artist Metadata.

The Flaw: Instead of analyzing the genre or mood (e.g., “Russian Pop/Nostalgia”), the system sees you like “Artist A” and assumes you want everything by “Artist A.”

Data Pollution: The system doesn’t realize that “Baby Shark” and “Valera” belong to two different personas. It treats the account as one person with a very confused taste.

##3. Recommendations for Improvement ###As a Quality Management professional, I propose the following “Corrective Actions”:

“Kids Mode” Toggle: A simple switch in the UI that prevents specific songs from affecting the main recommendation algorithm. This would solve the “data pollution” problem immediately.

“Surprise Me” (Discovery) Button: A feature to break the “Filter Bubble.” Instead of showing more songs by the same artist, this would pull tracks with similar acoustic profiles from different artists (e.g., matching the tempo and mood of Meladze’s “Two Rivers” with other 2000s pop icons).

Tiered Advertising: Instead of “3 ads in a row,” implement a more “humane” frequency. Aggressive advertising to parents with children creates a negative brand association that leads to churn (switching to Yandex/VK).

UI Simplification: Remove the “Podcast” clutter for users who only want music. The interface is currently “over-engineered,” making it hard to find a simple playlist.