4. IMPLEMENTATION
4.1 Gathering of data
4.1.1 Valence-Arousal Test
Motivated by the Valence-Arousal methodology to classify emotions, the aim of this first experiment was to track the reactions of the subject while exposed to different types of videos. These videos were selected to evoke variations specifically in the Interest, Ex- citement, Focus and Relaxation. Following the descriptions of the metrics, it can be de- fined as a hypothesis that the Interest metric should represent the Valence and the Ex- citement the Arousal [91].
The first four videos were considered as “relaxation” videos, so following with the Va- lence-Arousal approach, they were trying to evoke low levels of Excitement and middle levels of Interest. In order to see if the values given by the headset matched the feelings of the user, the subject was asked to define his state on target metrics. The definition consisted of four classification levels: low, mid-low, mid-high and high. Being low not feeling the emotion at all and high feeling completely that mental state.
There were different conclusions extracted from this first group of videos. The first one was that the Excitement metric was quite accurate with the values defined by the user. In most of the videos, the metric was defined as mid-low and the values recorded by the headset were in a low range too. This could match with the Valence-Arousal approach as these videos where evoking a calm and relaxed feeling which are related to a low arousal level. Regarding the Interest, the subject addressed a mid-low feeling being the metric varying around 50 in this first group of videos. Regarding the Focus metric, it was addressed as mid-low and mid-high in different videos, being possible to see the varia- tions on the metric through the videos but they never reached high values.
The second group of videos was selected to induce high arousal and valence responses in the user. The content of the videos was related to highlights moments of different
sports. In comparison to the first group of videos, considered a “low-arousal” class, the results were as expected. The Excitement was addressed as high by the subject in the moments of each highlight and mid-high between them. That reaction was completely verified watching the metric arriving at values around 90 in the highlights and around 50 in the meantime. It was interesting too to see how the Interest was defined as high by the user during the videos and the metric itself is around 80 most of the time even if the Excitement was lower. About the Focus, it was pointed as high by the user too, and as the Interest, it remained in high values around 75 even if the Excitement was low. The Relaxation was defined as mid-low by the user, but opposite as with the first videos the values were higher.
The third and last group of videos was focused on increasing the grade of aversion feel- ings on the subject. Here there was a bit more confusion on the metrics. The first video was pointed as mid-low in Excitement terms by the user due to the fact that they were generating a great degree of sadness and compassion in the subject. The metrics record low levels of the metric. At the same time, it was addressed as mid-low interest due to the great grade of aversion, but the Interest metric remained in middle levels. The Focus was defined as mid-low, matching with the registered metrics. The second video gener- ated a greater value of Excitement, addressed as high arousal by the user due to irrita- tion, what is completely verified on the registered data, but the grade of Interest was pointed as mid-low due to the aversion and the metrics show a middle-high grade of Interest through the video. The Focus was defined as mid-high which fits with the ob- tained values. In both videos, the Relaxation was addressed as low but the values suffer plenty of variation.
One of the most important aspects of this last experiment was the Stress metric. In this last case, it worked perfectly and it addressed high values, what fits with the grade of aversion that the subject was suffering.
The extracted conclusion of this experiment was that the Excitement metric represents very well the arousal of the approach, but the Interest can be confused, so it is necessary more data in order to verify that it represents the emotional valence. Regarding other metrics, the Relaxation is not completely related to the relaxation at the moment itself so it could be confusing to using it. But as a method to know if the user is relaxed or not it is possible to use the Stress metric. Regarding the Focus and the Engagement, it is necessary to perform more experiments to know how to apply them and which are they optimal ranges in the subject.