![D E E P L E A R N I N G F O R R E C O M M E N D E R S
• Predict the next item the user will want to interact with
Recommendations as sequence prediction
[a] -> b
[a, b] -> c
[a, b, c] -> d
[0, 0, 0, a] -> b
[0, 0, a, b] -> c
[0, a, b, c] -> d](https://image.slidesharecdn.com/sparkforrecommendersystemssorinpe-180524194219/85/Spark-for-Recommender-Systems-21-320.jpg)
Spark for Recommender Systems
- 1. Using Spark’s Machine Learning Library to Make Product Recommendations Sorin Pește Technology Solutions Professional, Data & AI Microsoft source: xkcd.com
- 2. (demo)
- 3. A PA C H E S PA R K A unified, distributed, open source engine for large-scale data processing Spark Structured Streaming Stream processing Spark MLlib Machine Learning Spark Core Engine Spark SQL Interactive Queries Yarn Mesos Standalone Scheduler Spark MLlib Machine Learning Spark Streaming Stream processing GraphX Graph Computation
- 4. S PA R K : A B R I E F H I S T O R Y
- 5. S PA R K D ATA F R A M E S A distributed collection of data that’s conceptually equivalent to a table
- 6. S P A R K M A C H I N E L E A R N I N G ( M L L I B ) Offers a set of parallelized machine learning algorithms for ML Supports Model Selection (hyperparameter tuning) using Cross Validation and Train-Validation Split. Supports Java, Scala or Python apps using DataFrame-based API Enables Parallel, Distributed ML for large datasets on Spark Clusters
- 7. S P A R K M L L I B A L G O R I T H M S Spark MLlib Algorithms
- 8. S P A R K M L L I B P I P E L I N E S
- 9. C O L L A B O R A T I V E F I L T E R I N G
- 10. C O L L A B O R A T I V E F I L T E R I N G User Latent Factors Item Latent Factors
- 11. A L T E R N A T I N G L E A S T S Q U A R E S ( A L S )
- 13. A L S : E X P L I C I T V S I M P L I C I T F E E D B A C K Explicit feedback — user rates items Implicit feedback — system records user activity Browses a product page Watches a movie trailer Plays a song Shares on social media etc Implicit feedback is generally used in real-world implementations
- 14. A L S : H Y P E R P A R A M E T E R T U N I N G Hyperparameters which can be adjusted: rank = the number of latent factors in the model maxIter = the maximum number of iterations regParam = the regularization parameter
- 15. A L S : H Y P E R P A R A M E T E R T U N I N G
- 16. A L S : W H A T A B O U T R E A L - T I M E ? Near real-time computation of ALS algorithm may be unfeasible Streaming variant of ALS, using Stochastic Gradient Descent https://github.com/brkyvz/streaming-matrix-factorization • Oryx Framework (http://oryx.io ) also offers streaming ALS
- 17. B E Y O N D A L S ALS-learned latent factors can be useful as input for other algorithms
- 18. D E E P L E A R N I N G A set of machine learning techniques that use multiple layers of non-linear processing units to learn useful data representations of input
- 19. D E E P L E A R N I N G W I T H S P A R K Integrations with existing DL libraries • Microsoft CNTK (mmlspark) • TensorFlow (TensorFlowOnSpark) • DeepLearning4J • Caffe (CaffeOnSpark) • Keras (Elephas) • mxnet • Paddle • and more… Implementations of DL on Spark • BigDL • DeepDist • SparkCL • SparkNet • Deep Learning Pipelines (Databricks) • and more… Distributed Hyperparameter Tuning
- 20. D E E P L E A R N I N G F O R R E C O M M E N D E R S • Neural Collaborative Filtering (He et al, 2017) https://arxiv.org/abs/1708.05031 https://github.com/hexiangnan/neural_collaborative_filtering Neural Collaborative Filtering
- 21. D E E P L E A R N I N G F O R R E C O M M E N D E R S • Predict the next item the user will want to interact with Recommendations as sequence prediction [a] -> b [a, b] -> c [a, b, c] -> d [0, 0, 0, a] -> b [0, 0, a, b] -> c [0, a, b, c] -> d
- 22. D E E P L E A R N I N G F O R R E C O M M E N D E R S • Predict the next item the user will want to interact with Recommendations as sequence prediction
- 23. D E E P L E A R N I N G F O R R E C O M M E N D E R S Session-based Recommendations with Recurrent Neural Networks (Hidasi et al., 2015) https://arxiv.org/abs/1511.06939 https://github.com/hidasib/GRU4Rec Recommendations as sequence prediction
- 24. D E E P L E A R N I N G F O R R E C O M M E N D E R S https://arxiv.org/pdf/1510.01784.pdf Featurize product images