Emotion classification in arousal-valence dimension using discrete affective keywords tagging

摘要

Human emotional recognition is a recent topic and an essential stage in Human-Computer Interaction issue. Recently, investigated researches have proved that the analysis of physiological signals is a relevant modality to assess the affective states. This work aims to classify the emotion into three defined areas in arousal-valence model using self-reported emotional keywords tagging instead of discrete rating value. Thus, we explored the peripheral physiological signals collected in the MAHNOB-HCI Tagging database. In this dataset, there are bodily responses of 24 participants after watching 20 stimuli videos to evoke their emotions. We preprocessed the data, extracted features and classified the states using the Support Vector Machine method. The obtained classification rates are 59.57%, 57.44% in arousal and valence, respectively. These results are promising compared to recent related work.