Building Recommender Systems with Machine Learning and AI

9h 7mBeginner2020-04-02

Authors

Frank Kane

Founder of Sundog Education and Subdog Software LLC

Course details

Automated recommendations are everywhere: Netflix, Amazon, YouTube, and more. Recommender systems learn about your unique interests and show the products or content they think you’ll like best. Discover how to build your own recommender systems from one of the pioneers in the field. Frank Kane spent over nine years at Amazon, where he led the development of many of the company’s personalized product recommendation technologies. In this course, he covers recommendation algorithms based on neighborhood-based collaborative filtering and more modern techniques, including matrix factorization and even deep learning with artificial neural networks. Along the way, you can learn from Frank's extensive industry experience and understand the real-world challenges of applying these algorithms at a large scale with real-world data. You can also go hands-on, developing your own framework to test algorithms and building your own neural networks using technologies like Amazon DSSTNE, AWS SageMaker, and TensorFlow.

Learning objectives
Top-N recommender architectures
Types of recommenders
Python basics for working with recommenders
Evaluating recommender systems
Measuring your recommender
Reviewing a recommender engine framework
Content-based filtering
Neighborhood-based collaborative filtering
Matrix factorization methods
Deep learning basics
Applying deep learning to recommendations
Scaling with Apache Spark, Amazon DSSTNE, and AWS SageMaker
Real-world challenges and solutions with recommender systems
Case studies from YouTube and Netflix
Building hybrid, ensemble recommenders

Skills covered

Machine LearningArtificial Intelligence FoundationsPythonArtificial Intelligence (AI)Open SourceOne-Off

Concepts

1. Getting Started

01 - Install Anaconda, review course materials, and create movie recommendations
02 - Course roadmap
03 - Understanding you through implicit and explicit ratings
04 - Top-N recommender architecture
05 - Review the basics of recommender systems

2. Introduction to Python

06 - Data structures in Python
07 - Functions in Python
08 - Booleans, loops, and a hands-on challenge

3. Evaluating Recommender Systems

09 - Train test and cross-validation
10 - Accuracy metrics (RMSE and MAE)
11 - Top-N hit rate - Many ways
12 - Coverage, diversity, and novelty
13 - Churn, responsiveness, and A B tests
14 - Review ways to measure your recommender
15 - Walkthrough of RecommenderMetrics.py
16 - Walkthrough of TestMetrics.py
17 - Measure the performance of SVD recommendations

4. A Recommender Engine Framework

18 - Our recommender engine architecture
19 - Recommender engine walkthrough, part 1
20 - Recommender engine walkthrough, part 2
21 - Review the results of our algorithm evaluation

5. Content-Based Filtering

22 - Content-based recommendations and the cosine similarity metric
23 - K-nearest neighbors (KNN) and content recs
24 - Producing and evaluating content-based movie recommendations
25 - Bleeding edge alert - Mise-en-scene recommendations
26 - Dive deeper into content-based recommendations

6. Neighborhood-Based Collaborative Filtering

27 - Measuring similarity and sparsity
28 - Similarity metrics
29 - User-based collaborative filtering
30 - User-based collaborative filtering - Hands-on
31 - Item-based collaborative filtering
32 - Item-based collaborative filtering - Hands-on
33 - Tuning collaborative filtering algorithms
34 - Evaluating collaborative filtering systems offline
35 - Measure the hit rate of item-based collaborative filtering
36 - KNN recommenders
37 - Running user- and item-based KNN on MovieLens
38 - Experiment with different KNN parameters
39 - Bleeding edge alert - Translation-based recommendations

7. Matrix Factorization Methods

40 - Principal component analysis (PCA)
41 - Singular value decomposition (SVD)
42 - Running SVD and SVD++ on MovieLens
43 - Improving on SVD
44 - Tune the hyperparameters on SVD
45 - Bleeding edge alert - Sparse linear methods (SLIM)

8. Introduction to Deep Learning

46 - Deep learning introduction
47 - Deep learning prerequisites
48 - History of artificial neural networks
49 - Playing with TensorFlow
50 - Training neural networks
51 - Tuning neural networks
52 - Introduction to TensorFlow
53 - Handwriting recognition with TensorFlow, part 1
54 - Handwriting recognition with TensorFlow, part 2
55 - Introduction to Keras
56 - Handwriting recognition with Keras
57 - Classifier patterns with Keras
58 - Predict political parties of politicians with Keras
59 - Intro to convolutional neural networks (CNNs)
60 - CNN architectures
61 - Handwriting recognition with CNNs
62 - Intro to recurrent neural networks (RNNs)
63 - Training recurrent neural networks
64 - Sentiment analysis of movie reviews using RNNs and Keras

9. Deep Learning for Recommender Systems

65 - Intro to deep learning for recommenders
66 - Restricted Boltzmann machines (RBMs)
67 - Recommendations with RBMs, part 1
68 - Recommendations with RBMs, part 2
69 - Evaluating the RBM recommender
70 - Tuning restricted Boltzmann machines
71 - Exercise results - Tuning a RBM recommender
72 - Auto-encoders for recommendations - Deep learning for recs
73 - Recommendations with deep neural networks
74 - Clickstream recommendations with RNNs
75 - Get GRU4Rec working on your desktop
76 - Exercise results - GRU4Rec in action
77 - Bleeding edge alert - Deep factorization machines
78 - More emerging tech to watch

10. Scaling It Up

79 - Introduction and installation of Apache Spark
80 - Apache Spark architecture
81 - Movie recommendations with Spark, matrix factorization, and ALS
82 - Recommendations from 20 million ratings with Spark
83 - Amazon DSSTNE
84 - DSSTNE in action
85 - Scaling up DSSTNE
86 - AWS SageMaker and factorization machines
87 - SageMaker in action - Factorization machines on one million ratings, in the cloud

11. Real-World Challenges of Recommender Systems

88 - The cold start problem (and solutions)
89 - Implement random exploration
90 - Exercise solution - Random exploration
91 - Stoplists
92 - Implement a stoplist
93 - Exercise solution - Implement a stoplist
94 - Filter bubbles, trust, and outliers
95 - Identify and eliminate outlier users
96 - Exercise solution - Outlier removal
97 - Fraud, the perils of clickstream, and international concerns
98 - Temporal effects and value-aware recommendations