Organization and Retrieval in Affectively Annotated K-Line Indexed Media Repositories

A.A. Toptis and A. Dubitski (Canada)


Artificial Intelligence, Multimedia Systems, Affective Computing.


We present a method that organizes a media repository upon consultation of a user’s individualized mood conditions and content preferences. The described method is applicable to any type of media content, as long as the content is available prior to the commencement of the organization process. Examples of such media are music, video, pictures, text, etc. We describe our method when the media type is music. Based on the widely accepted emotion theory of Ekman our method captures a user’s song selection preferences when the user is under various emotional states and it forms “memory experiences” by employing a data structure called K-lines, proposed by Minsky. The captured preferences and memory experiences are then normalized to generate a tolerance spectrum within which certain mood conforming song selections are deemed to be desirable. As a result, the song repository is transformed to a personalized mood-laden K-line mesh which can then be tapped to satisfy any future affectively annotated song listening requests. We implement our method and present evaluation results of testing with human participants. The evaluation clearly indicates that the proposed method is robust, and significantly outperforms the most typical paradigm of random song selection.

Important Links:

Go Back