Interpreting Keystrokes to Ascertain Human Mood

Abstract

The human state of mind can be reflected in interactions with computers, such as the time taken to type a word, the number of times that a correction is made, the average time taken to press a set of keys, etc. In this research, we developed an application that captures keystroke-based human-computer interactions while gathering user mood (pleasant vs. unpleasant) information utilizing a pre-established survey instrument – the Brief Mood Introspection Survey (BMIS). Using keystroke-based features and saliency measurements of the features, we constructed models to differentiate between pleasant and unpleasant moods. Once unpleasant moods are detected, possible interventions can be applied. For unpleasant mood detection, generalized neural network (GRNN), probabilistic neural network (PNN), and Levenberg-Marquardt neural network (LMNN) algorithms provided the best F1-scores, whereas decision tree (DT) algorithm provided the best recall score.