Aiming at the requirement of intelligent user interface state awareness and intention prediction, a method of interactive state classification and intention prediction based on eye tracking is proposed. 5 states of interaction are defined, which are monitoring state, tracking state, decision state, burst state and off-loop state, design induced experiments were conducted to collect eye movement data in each state, a single factor analysis of variance shows that the 7 eye movements in the 5 interactive states has a significant difference. The prediction accuracy of the SVM classification model under category 5 conditions is 77.2%, while the accuracy rate of the classification under category 4 conditions is 85.9%, individual differences also have important effects on prediction accuracy, the accuracy rate of single subjects under category 5 conditions are more than 84%, while under category 4 conditions up to 90%. The research results are of reference value to the design and application of intelligent user interface.
Read full abstract