Mood Recognition based on Biosignals

Abstract

Affective computing has been dominated by research on emotion. This research contributes to the field by shifting its focus to mood, a more diffused and long-term affective state. Knowing a person’s mood could positively contribute to the human- computer and human-robot interaction. The Continuously Annotated Signals of Emotion (CASE) data set was used, which allowed recognizing mood through the use of biosignals. Electrodermal Activity (EDA), Electromyography (EMG), and the continuous annotation of participants’ valence and arousal were analyzed. Using this continuous annotation, we selected appropriate mood windows and labeled them with four discrete mood states. To the author’s knowledge, the methodol- ogy presented for continuous annotation processing is new and contributes to fu- ture mood-related research. The extracted mood labels showed a high construct validity. A generalized mood classification model was built, which showed a clas- sification accuracy of 63.64%. This result is comparable to existing state-of-the-art mood recognition models. A leave-one-participant-out cross-validation underlined the vulnerability of pattern recognition in small data sets.