Few-Shot Learning for Detecting Affective States from Keyboard and Mouse Data

Around 49% of people working with a computer for more than 4 hours a day and 30% of those working with a computer for 2–4 hours a day report stress due to computer use [1]. Affects such as stress and fatigue have been shown to cause attention loss, slow reaction, and tiredness leading to diminished productivity and burnout [2]. Decrease in productivity has been observed in subjects during a negative emotional state. The study in [3] found that 70% of users showed decreased typing speed in negative emotional state and 83% of users showed increase in typing speed in positive emotional state compared to the neutral state. With the increase in popularity of work from home, the blurring of lines between professional and personal lives in work-from-home may cause new forms of stress and other emotional afflictions [4]. Accurate reading of human affective states could help make decisions that improve employee well-being and productivity. Developing tools to detect emotional states and enabling proper emotional well-being strategies in workplace is thus of vital importance [5].