Building your own Distributed Streaming Recommendation Engine
summary
A distributed streaming ALS implemented with stochastic gradient descent updates per incoming rating, using Apache RDDs and Apache Spark DStreams and Kafka for 5s windows (1,000 records per mini‑batch). Model stores user/item latent vectors plus global and per‑entity biases, initializes unseen IDs with random vectors and uses full‑outer joins to merge new and existing factors. Evaluation on MovieLens (~26M ratings) shows streaming RMSE trends toward batch ALS but highlights cold‑start, hyperparameter search cost, and shuffle/lookup performance issues. MovieLens 26M
AI-written summary · about
chapters (39)
- 0:00Introduction
- 0:43Recommendation engines
- 1:40Collaborative filtering
- 3:16Factorization
- 4:49Derivation
- 5:54Regional Ratings Matrix
- 6:12Demo
- 6:52Assigning Ratings
- 7:02Ratings Matrix
- 8:05Rating Factorization in Streaming
- 9:10Stochastic Gradient Descent SGD
- 9:33Biases
- 10:44Stochastic gradient descent
- 11:31The practical difference
- 13:48RDDs
- 14:20Spark Sommelier
- 16:10Streaming Recommendation
- 17:28MLE API
- 18:12Factor class
- 18:36Summary
- 19:20Special Scenario
- 19:57Limit Case
- 21:16Random Feature Vector
- 21:42User and Product Weighting Factors
- 22:37Product Weighting Factors
- 23:00Global Bias
- 23:46Predicted Rating
- 24:13Getting the Error
- 25:00radiance
- 26:21new observations
- 26:43weighting factors
- 27:43data set
- 29:07training batch LS model
- 29:40validation RDD
- 31:35results
- 32:33problems
- 34:46streaming version
- 35:12grid search
- 35:57performance considerations
related talks
01h 48mRecommendation engines explained
37mBuild a Recommender System in Apache Spark and Integrate it Using Akka
45mReal-time, Streaming Advanced Analytics....
32mBuilding an recommendation system for IPTV
48mBuild a Recommender System in Apache Spark and Integrate It Using Akka
40mIntroduction to Machine Learning
41mRecommendations for Building Machine Learning Software
34mActionable Insights with Real-time Streaming Analytics of Customer Reviews
