Diversity and Data: Cases in the Music Industry
summary
PPL reports a repertoire database with over 10 million recordings and receives ~919,000 recordings this year so scaling is real. A recommendation engine is built using Python, Spark on HDFS, Redshift and visualization in Gephi, running fuzzy matching, TF‑IDF + fuzzy matching + DBSCAN clustering and bootstrapped outlier detection to infer aliases and yearly core lineups. Collaborative filtering alone fails because of noisy titles, misspellings and empty lineups. The pipeline finishes fuzzy-matching and recommendations in about 3–4 hours.
AI-written summary · about
chapters (6)
- 0:00PPL and data mission
- 6:31PPL’s data science turn
- 11:57Performer recommendation engine
- 21:12Entity disambiguation in music data
- 27:44Band membership changes over time
- 31:07Q&A
related talks
40mBuilding A Folk Music Recommendation System with LLMs
38mmusic2vec: A smart music recommendation and discovery
42mSpotify: Horizontal scalability for great success
20mSo tell me, what is your musical taste?
59mClosing Keynote: Paul Lamere
32mSpotify and Python: love at first sight
38mKeynote: Using the Python Data Science Stack to Determine
19mMusic Streaming in the Creative Economy -