Scalable Recommender Systems: Where Machine Learning Meets Search

Abstract

This tutorial provides an overview of how search engines and machine learning techniques can be tightly coupled to address the need for building scalable recommender or other prediction-based systems. In particular, we will review ML-Scoring, an open source framework, created by the authors that tightly integrates machine-learning models into Elasticsearch, a popular search engine that is distributed, scalable, highly available with real-time search and analytic functionalities. The fundamentals and basic methods in information retrieval and machine learning will be explained. Accompanying the theory, practical examples will illustrate their applications with a series of hands-on exercises. These will demonstrate how to load a dataset into Elasticsearch, how to train a model in an external software framework such as Spark, Weka, or R, and finally how to load the trained models as a ML-Scoring plugins created for Elasticsearch.