Heuristic Determination of the Most Influential Affective Attributes in K-Line Indexed Media Collections

A.A. Toptsis and A. Dubitski (Canada)


Artificial Intelligence, Affective Computing, Intelligent Systems, Multimedia Systems.


We propose a methodology that can be used to make an educated guess of which emotion attributes, among a large set of emotion attributes, may be the most influential for the development of a media related affective computing system. Our method determines the most suitable emotion attributes for building such a system. As a result, any system developer can apply our method and determine the most influential emotion attributes for the system under development. Based on such findings, the developer can then calibrate the affective space to be comprised of the most influential emotion attributes rather than all the attributes that come with a typical off-the-self emotion classification and thus build the final production grade system using only the significantly smaller affective space that is recommended by our method. The obvious benefit if that the production-grade system can be built much faster and also perform faster due to using a significantly reduced number of emotion attributes. We describe our method and present findings from its application on two different media collections. The findings indicate that the method can successfully distinguish among different emotion attributes and recommend the most significant ones for the development of the underlying system.

Important Links:

Go Back