chapter eleven

11 Interfaces for Data Annotation

 

This chapter covers:

  • Understanding basic Human-Computer Interaction (HCI) principles.
  • Applying HCI principles in annotation interfaces.
  • Combining human and machine intelligence to maximize the strengths of each.
  • Implementing interfaces with different levels of machine learning integration
  • Adding machine learning to applications without disrupting existing work practices.

In the last 10 Chapters, we have covered everything about Human-in-the-Loop Machine Learning, except for the vital component of the human-machine interface! This chapter covers how to build interfaces that maximize the efficiency and accuracy of the annotations. This chapter also covers the trade-offs: there is no one set of interface conventions that can be applied to every task, so you must make an informed decision about what is best for your task and your annotators.

11.1        Basic Principles of Human-Computer Interaction

11.1.1    Introduction to Affordance, Feedback and Agency

11.1.2    Designing Interfaces for Annotation

11.1.3    Minimizing eye movement and scrolling

11.1.4    Keyboard shortcuts and input devices

11.2        Breaking the rules effectively

11.2.1    Scrolling for batch annotation

11.2.2    Foot pedals

11.2.3    Audio inputs

11.3        Priming in Annotation Interfaces

11.3.1    Repetition Priming

11.3.2    Where priming hurts

11.3.3    Where priming helps

11.4        Combining Human and Machine Intelligence

11.4.1    Annotator Feedback

11.4.2    Maximizing Objectivity by asking what other people would annotate

11.4.3    Recasting continuous problems as ranking problems

11.5        Smart interfaces for maximizing human intelligence

11.5.1    Smart Interfaces for Semantic Segmentation

11.5.2    Smart Interfaces for Object Detection

11.5.3    Smart Interfaces for Language Generation

11.5.4    Smart Interfaces for Sequence Labeling

11.6        Machine Learning to assist human processes

11.6.1    The perception of increased efficiency

11.6.2    Active Learning for increased efficiency