K-DIME: An Affective Image Filtering System

Even though extracting the semantic content of information is a critical feature in information search, today’s Web sites still have poor capabilities in this regard, especially when it comes to multimedia data. While most Web sites provide visual, and sometimes, auditory content, they don’t provide any tools to process the resulting subjective information, and by doing so, they fail to support the user’s decision process. Often, users will select a picture or a piece of music based on the impression or feeling it conveys from a pool of many others satisfying the same objective criteria, such as percentage of a given color or type of object. In fact, in fields such as design, advertising, or entertainment, affective content is the main criterion for selection. Therefore, it’s crucial that users can fully interact with the multimodal components of the information, beyond a simple search or viewing of file names, verbal descriptions, or even indexes of objective content. My group in the Database Systems Lab at the University of Aizu is investigating how to endow a computer system with the ability to handle affective and subjective criteria. In this article, I present our framework for software agents to adapt to their users and perform Web-based multimedia search on the basis of objective and subjective criteria. We expect such systems to continuously tune the filtering criteria by exploiting users’ feedback, thus providing flexibility against the intrinsic variability of affective experiences.