Collective Intelligence
summary
Focus on recommendation algorithms: a simple friends-of-friends intersection recommender (counts mutual followers) proves effective by checking how many of the top-10 are already followed; a nearest-neighbors implementation uses ukian distance and cosine similarity on the Netflix sample (177k movies, ~450k users) and a top-500 subset; linear algebra work uses the linalg gem (Fortran/Lea) for SVD (rank-4). Prototypes run on a laptop (one movie vs 500 takes ~30 minutes), produce reasonable neighbors (Matrix → Terminator 2, Fifth Element; SVD → Lord of the Rings) but expose memory and IO scaling limits.
AI-written summary · about
chapters (17)
- 0:00Intro
- 1:33Examples
- 4:02Google Ads
- 5:00Amazon
- 6:00Follow Recommender
- 11:11Matrix
- 16:31Example
- 23:36Results
- 24:20Problems
- 26:21Methods
- 29:53SVG SVG
- 30:03Tracking
- 31:29Ruby Matrix
- 34:14Movie Recommendations Code
- 34:51Future Work
- 38:53JRuby
- 40:04ActiveRecord
related talks
01h 31mHands on - Build a Recommender system
37mA primer on recommendation systems.
01h 21mUnlocking Personalization: A Deep Dive into Modern Recommendation Algorithms
38mBuilding Recommender System in PHP8
58mRecommender Systems – A Walkthrough
33mPicking What to Watch Next: Build a Recommendation System
01h 48mRecommendation engines explained
15mWhat they didn’t teach you in Grad School #bbuzz