Chapter 5. Case study: click prediction for online advertising

This chapter covers

  • A real, large-scale intelligent system
  • Targeting users with web-browsing data
  • Ranking users’ propensity to interact using logistic regression

Online click prediction is a specific problem in the world of advertising and is an excellent example of a high-velocity, high-throughput problem where decisions must be made with a very low latency. In general, this class of problems has a large number of applications. Online trading, website optimization, social media, the Internet of Things, sensor arrays, and online gaming all generate high-velocity data and can benefit from fast processes that make decisions in light of the most recent information possible.

Every time you open a browser and navigate to a web page, thousands if not millions of decisions are made in order to determine what ads to put in front of you. These decisions are made by communicating across many data stores and by understanding whether a specific ad will have a positive impact on the user. This has to happen quickly, because decisions must be made before the page can finish loading.

5.1. History and background

5.2. The exchange

5.3. What is a bidder?

5.4. What is a decisioning engine?

5.5. Click prediction with Vowpal Wabbit

5.6. Complexities of building a decisioning engine

5.7. The future of real-time prediction

5.8. Summary