Spotify Pattern Mining Recommendations
Market basket analysis on 1M+ Spotify listening records to surface high-confidence music recommendation rules
Source private — academic coursework. Available upon request.
Overview
- Frames each user's listening history as a "basket" and mines association rules to find artists and tracks that are reliably co-listened — the basis for a simple recommendation engine
- Built three transaction sets (artist, track, artist+track) from 742K cleaned Spotify records using user-level market basket framing
- Benchmarked Apriori vs. ECLAT across speed, memory, and rule quality — Apriori won both benchmarks
- Surfaced top rules by lift (e.g. Daft Punk catalog co-listens, hip-hop and pop trios) with an actionable recommendation pitch
Tech Stack
| Layer | Tools |
|---|---|
| Data Prep | R, dplyr |
| Mining | arules (Apriori & ECLAT; support/confidence/lift tuning) |
| Visualization | arulesViz, ggplot2, plotly |
| Profiling | Rprofmem, proc.time |
| Reporting | Quarto |
Skills Developed
- Association rule mining and parameter tuning
- Algorithm benchmarking (runtime and memory)
- Translating statistical output into business recommendations