So during last week’s meeting, I told Karsten about my progress regarding sensor research. He instructed me to dive even deeper into this research area and construct a couple of matrices in order to summarize my findings. Furthermore, since I had also been reading up on the APIs of the HCI department’s available brainwave sensors (i.e. the Muse headband and the MindWave headset), I suggested to take a closer look into that by experimenting with one of the devices in the weekend.
Firstly, there are two general ways to express human emotion: discretely and dimensionally. A discrete model of emotion represents different emotions as distinct elements in the emotion space (e.g. happy, sad, angry, scared, …), whereas a dimensional model plots these emotions in multiple dimensions (e.g. valence, arousal, dominance, ….) without the need to label them. Although the academic world has apparently not yet reached consensus about which model is best, most researchers do prefer Russell’s circumplex model of affect when opting for a dimensional model of emotion. The dimensions that define this model are valence (i.e. pleasure/displeasure) and arousal.
My first matrix shows a collection of discrete emotions on the horizontal axis, accompanied by the signals through which they can be measured on the vertical axis. An X depicts that a measurement relation is present in literature, whereas the absence of an X means that proof of such a relation is yet to be found.
My second matrix summarizes the relation between some dimensions in the emotion space on the horizontal axis and the signals through which they can be measured on the vertical axis. Again, an X depicts that a measurement relation is present in literature, whereas the absence of an X means that proof of such a relation is yet to be found.
The third and final matrix shows the aforementioned signals on the vertical axis, accompanied by their measurement tools on the vertical axis. Yet again, an X depicts that a measurement relation is present in literature, whereas the absence of an X means that proof of such a relation is yet to be found.
Next to the construction of these matrices – which took a really long time because of the insane amount (121) of research papers I already found on the subject of emotion recognition – , I also took a closer look at the Muse headband API. I again relied on the device service of the HCI department to provide me with a Samsung SII for the weekend, accompanied with a Muse headband this time. At a first glance, the API didn’t appear all that difficult to understand, especially when taking into account that the accompanying website has an example Android project available for its developers. However, after experimenting for a while with this project, it came to my attention that understanding the API was going to be harder than I thought… It was documented, of course, but poorly. But of course, not all is lost, since there is still another brainwave sensor in the HCI department to explore!