A Novel Method to Build and Validate an Affective State Prediction Model from Touch-Typing

Abstract

Affective systems are supposed to improve user satisfaction and hence usability by identifying and complementing the affective state of a user at the time of interaction. The first and most important challenge for building such systems is to identify the affective state in a systematic way. This is generally done based on computational models. Building such models requires affective data. In spite of the extensive growth in this research area, there are a number of challenges in affect induction method for collecting the affective data as well as for building models for real-time prediction of affective states. In this article, we have reported a novel method for inducing particular affective states to unobtrusively collect the affective data as well as a minimalist model to predict the affective states of a user from her/his typing pattern on a touchscreen of a smartphone. The prediction accuracy for our model was 86.60%. The method for inducing the specific affective states and the model to predict these states are validated through empirical studies comprising EEG signals of twenty two participants.