Building Recommendation Engines with PySpark

This chapter will show you how powerful recommendations engines can be, and provide important distinctions between collaborative-filtering engines and content-based engines as well as the different types of implicit and explicit data that recommendation engines can use. You will also learn a very powerful way to uncover hidden features (latent features) that you may not even know exist in customer datasets.

In this chapter you will review basic concepts of matrix multiplication and matrix factorization, and dive into how the Alternating Least Squares algorithm works and what arguments and hyperparameters it uses to return the best recommendations possible. You will also learn important techniques for properly preparing your data for ALS in Spark.

In this chapter you will be introduced to the MovieLens dataset. You will walk through how to assess it's use for ALS, build out a full cross-validated ALS model on it, and learn how to evaluate it's performance. This will be the foundation for all subsequent ALS models you build using Pyspark.

In most real-life situations, you won't not have "perfect" customer data available to build an ALS model. This chapter will teach you how to use your customer behavior data to "infer" customer ratings and use those inferred ratings to build an ALS recommendation engine. Using the Million Songs Dataset as well as another version of the MovieLens dataset, this chapter will show you how to use the data available to you to build a recommendation engine using ALS and evaluate it's performance.