Description
Models users as agents choosing actions to maximise expected reward under cognitive and perceptual bounds, and derives predicted optimal behaviour for a given interface.
An MDP specifies states, actions, transition probabilities and rewards; a POMDP adds partial observability, which suits users who cannot see everything. Computational rationality assumes users adapt their strategies optimally to the interface, given bounds such as limited visual acuity and memory. Solving the model (often by reinforcement learning) yields predicted behaviour: where users look, when they give up, which menu strategies they adopt. Comparing interfaces then means comparing the optimal policies and their expected costs. This gives UX a principled theory of why users behave as they do, and a way to explore design spaces in simulation before testing.
Attributes
- Discipline
- Task Analysis & Cognitive Modelling
- Loss type
- Latent
- Stage
- Design
- Data
- Analytical / expert
- Effort
- Months
Steps
- Define the task as states, actions, observations and rewards, including time costs and task success value.
- Specify bounded cognitive and perceptual constraints (foveal vision, memory decay, motor noise) based on literature.
- Solve for the optimal or near-optimal policy using dynamic programming or reinforcement learning.
- Generate predicted behaviour (fixations, clicks, abandonment, strategy choice) for each design variant.
- Validate against empirical data from a small study or logs, checking qualitative patterns and quantitative fit.
- Use the model to explore further design variants and to explain observed behaviour.
Sources
- Oulasvirta, Kristensson, Bi & Howes (eds.), Computational Interaction (2018)
- Lewis, Howes & Singh, Computational rationality: Linking mechanism and behavior through bounded utility maximization (2014), Topics in Cognitive Science
- Sutton & Barto, Reinforcement Learning: An Introduction (2nd ed., 2018)
Also known as
- MDP
- POMDP models of users
- Computational rationality
- Reinforcement learning user models
Categories
- User Research
- Task Analysis & Cognitive Modelling