Summary

In this chapter, we have described the general strategy of recommender systems and implemented in Java an early version developed at Amazon. We first explored the notion of similarity measures, including cosine similarity. We saw how user ratings are used in recommender systems. We looked at the general idea of sparse matrices, which is the likely mathematical structure for a utility matrix, and then saw how they could be implemented using a random access file. Finally, we reviewed the Netflix prize competition, which raised the level of interest in recommender systems among data scientists.