chapter four

4 Algorithmic amplification: Measuring RecSys

 

This chapter covers

  • The challenges in measuring recommender systems—and their algorithmic amplification—at scale
  • The limitations of how recommender systems learn users’ preferences
  • The socio-technical nature of recommender systems

Recommender systems are responsible for showing us content in the hope that we find it valuable and continue using the platform, either by scrolling social media or by continuing to listen to songs or other media. They are not only distributing our attention to various pieces of content but also, when successful, even increasing the level of our attention—or at least the amount of our time—dedicated to platforms.

4.1 Quantifying amplification

4.2 Implicit learning

4.3 Users’ preferences and proxy metrics

4.4 Testing recommender systems: A/B tests

4.5 Lack of a baseline

4.6 Socio-technical nature of technology

4.6.1 Feedback loops

4.6.2 Feedback loops in recommender systems and social media

4.6.3 The rich-getting-richer effect

4.7 Labeling

4.7.1 Labeling content

4.7.2 Labeling users

4.8 Summary