Computational Cognitive Modelling for UX Prediction

Description

Builds runnable simulations of the cognitive processes users employ, predicting performance, errors and learning on a design before testing with people.

Computational cognitive modelling expresses hypotheses about perception, memory, attention and decision-making as runnable programs (symbolic architectures such as ACT-R and EPIC, or connectionist networks) and simulates users interacting with a design. Because the model is explicit, it yields quantitative predictions of task times, error likelihoods and learning trajectories, and it explains why a design performs as it does. In UX the method allows many variants to be screened before recruiting participants and supports decisions in domains where testing is expensive or risky. Rigour depends on grounding the model in established architecture parameters, validating predictions against at least some human data, and reporting where the model's assumptions are likely to break down, such as novel strategies or emotional states.

Attributes

Loss type
Extinct
Stage
Design
Effort
Weeks

Steps

  1. Select a cognitive phenomenon relevant to the interface:
  2. For navigation: use a spreading activation model to predict which labels users will associate.
  3. (2) For decision-making: use a drift-diffusion model to predict decision time and error rates for different option set sizes.
  4. (3) For learning: use a reinforcement learning model to predict which feature interaction sequences users will learn. Run the model with parameters fitted from prior research. Use model predictions to design the study's hypotheses.

Sources

  • Computational Experiments in Cognitive Neuroscience — Yuko Munakata
  • Cognitive Modeling: Nine Paradigms (slide deck)

Also known as

  • Cognitive simulation
  • Connectionist modelling
  • ACT-R modelling

Categories

  • User Research
  • Task Analysis & Cognitive Modelling