liveBook · Manning

Priming

As explained in Human-in-the-Loop Machine Learning: Active learning and annotation for human-centered AI

Priming is the influence of context or preceding events on a person's perception.

Priming occurs when context or preceding events influence how a person interprets an item. In annotation, task design, item order, and annotator background can all shape judgments, so the same item may be labeled differently depending on what came before or surrounds it. section 1.4.2 section 11.3

Priming can help or harm, depending on the task:

  • Repetition can skew subjective ratings or close categorical decisions, such as labeling a person pushing a bicycle as a pedestrian or cyclist. section 11.3.2
  • Context can help resolve ambiguity: the theme of a health transcription task may make “patients” more likely than “patience.” Familiarity can also make annotators faster, while a positive emotional state may support speed and accuracy. section 11.3.3
  • When designing annotation workflows, consider how task order and context interact with annotator backgrounds. Ground truth and interannotator agreement can help monitor annotation quality; for high-volume batch labeling, scrolling can reduce priming bias when balanced data cannot be displayed. section 11.3 section 11.8

A cue outside the task

People interpreted an accent as more New Zealand-like when a kiwi bird was in the room, and more Australia-like when a kangaroo was present—even though the toys were not mentioned. Context can influence judgments without being part of the explicit task. section 1.4.2

How Human-in-the-Loop Machine Learning: Active learning and annotation for human-centered AI covers it

How deep it goes
Defines priming and distinguishes repetition, contextual, positive, and affective effects through task examples and interface implications.
Examples
Uses sentiment sequences, bicycle labels, accent perception, ambiguous transcription, and music as examples of priming effects.
What you can do after
Apply interface-design considerations by examining task order, context, judgment subjectivity, and whether priming helps or harms annotation.
How it connects
Connects priming to annotation interfaces, sentiment analysis, categorical labeling, transcription, and annotator background.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Human-in-the-Loop Machine Learning: Active learning and annotation for human-centered AI. The section links point to the parts of the book it draws on.