New book review for Practical Machine Learning: Innovations in Recommendation, by Ellen Friedman and Ted Dunning, O'Reilly, 2014, reposted here:
This work is a short, white paper sized text that walks the reader through practical aspects of machine learning recommendation, specifically for those who are new to this space. After discussing the build versus buy dilemma with respect to incorporation of a recommendation system into the enterprise, the authors stress simplicity for ease of adoption and maintenance, and state that smart simplification in the case of recommendation is the focus of this paper, focusing on user behavior, co-occurrence, and text retrieval. Following this introductory discussion, the authors delve into a high-level discussion of choosing and collecting data for input into a recommendation system, building the recommendation model using co-occurrence, how Apache Mahout builds a model, and use of Apache Mahout with Apache Lucene (composed of Solr and Lucene-Core).
For readers already somewhat familiar with this problem space, it is not until Chapter 6 ("Example: Music Recommender") that the authors walk through an example that is used for a machine learning course developed by MapR Technologies. While still at a high level, this example discusses data sources, recommendations at scale using an Apache Hadoop based cluster, use of search to make recommendations, dithering, anti-flood measures, and multimodal and cross recommendation. While some technologists will likely be inclined to initiate work in this area by hitting community websites for the open source projects discussed, this book provides an accessible introduction to machine learning recommendation, and references to commercial implementations such as LucidWorks Search and the MapR distribution of Hadoop can be largely ignored.