Back to projectPublic page

Content-Based Filtering: Item Features

Understand how platforms analyze what you like and find items with similar characteristics

Part of How Netflix, Spotify, TikTok Know You

4 blocks0 nested pages
Last updated Oct 29, 2025. Clone to remix or explore the blocks below.
5396c4e5...
2cebb560...
content

Pandora's 450-Feature Obsession

Educational content slides

Pandora's 450-Feature Obsession

Pandora's Music Genome Project:

Trained musicians manually analyze every song

450 attributes measured:

  • Melody complexity
  • Harmonic structure
  • Rhythm patterns
  • Vocal style (raspy, smooth, energetic)
  • Instrumentation (synthesizers, guitars, drums)
  • Lyrical themes

Takes 20-30 minutes per song. A human listens, tags, scores.

Example for Billie Eilish: "whispered intimate vocals, minimal instrumentation, dark pop production, unconventional song structure"

This is content-based filtering: analyzing the THING itself, not user behavior.

063c3150...
quiz

Quiz: 3 Questions

Test your understanding with this quiz.

0 / 3

Pandora recommends obscure indie artists you've never heard of, while Spotify mostly suggests popular artists. Why the difference?

bbc21dc0...
feedback

Build Your Content-Based Filter

Complete this exercise and get AI-powered feedback.

Build Your Content-Based Filter

Content-based filtering excels at niche interests where user data is sparse.

Your task: Define features for something you're passionate about, then use those features to make predictions.

This reveals: How feature selection determines recommendation quality. Good features = good recommendations.