Title

Using probabilistic finite state models to evaluate human-computer interfaces to case systems

Date of Completion

January 1992

Keywords

Computer Science

Degree

Ph.D.

Abstract

This project involves the development of a technique for modeling and analytically evaluating human-computer interfaces. The modeling technique is supported by an extensive set of theorems and algorithms which describe the various model building phases. The theorems and algorithms show that the representations of each human-computer interface converges to a single probabilistic finite state model which accurately represents the human-computer interaction. The resulting probabilistic finite state models are then used as predictors to generate data which can be statistically analyzed to yield a comparison between alternative human-computer interfaces.^ The technique uses an approach in which probabilistic finite state models of the prospective human-computer interfaces are constructed. Specifications of the human-computer interfaces provide for the partial definitions of initial probabilistic finite state models. Initially, the missing components of the probabilistic finite state models are the time distributions which specify how much time is spent in each state and the probabilities associated with the state transitions. Experiments are required to determine the time distributions and state transition probabilities. For this project a set of experiments was performed using simulated human-computer interfaces to a CASE system component which calculates time performance of software for parallel architectures. The data collected during the experiments were used to complete and refine the probabilistic finite state models. The refined models were then exercised as predictors and the generated data was statistically analyzed to compare the interfaces. The results demonstrated that the modeling and evaluation technique can be used effectively in the design and refinement of human-computer interfaces. ^