Practical Machine Learning: Innovations in Recommendation

Chapter 6. Example: Music Recommender

One of the best ways to learn how a recommender works is to put your hands on one. With that in mind, we developed a concrete example of a recommender for a machine-learning course developed by MapR Technologies, a distributed computing platform company, with help from a training and consulting company, Big Data Partnership. The recommender is for a mock business, Music Machine. We explore it here to illustrate what we’ve covered so far.

Business Goal of the Music Machine

This mock music company wants to increase stickiness for their web-based music-listening site by offering visitors enticing music recommendations that will keep them on the site longer and keep them coming back.

The business is a figment of the authors’ imagination, but the music-recommendation engine and non-public Music Machine website (see Figure 6-1) are real. They provide a working example of a simple in-production recommender built according to the design we’ve been discussing.

Figure 6-1. Screenshot of the mock music-listening website for which a real Mahout-Solr recommender was built.

Online businesses with similar goals are quite real and widespread, and you have most likely encountered them yourself, whether the items of interest were music, books, new cars, destinations, or something else.

Data Sources

Two types of data are needed for the music recommender: metadata about ...