User-Adapted Image Descriptions from Annotated Knowledge Sources Berardina De Carolis and Fiorella de Rosis Intelligent Interfaces, Department of Computer Science, University of Bari, Italy {decarolis, derosis}@di.uniba.it Abstract. We present the first results of a research aimed at generating user- adapted image descriptions from annotated knowledge sources. This system employes a User Model and several knowledge sources to select the image at- tributes to include in the description and the level of detail. Both 'individual' and 'comparative-descriptions' may be generated, by taking an appropriate 'ref- erence' image according to the context and to an ontology of concepts in the domain to which the image refers; the comparison strategy is suited to the User background and to the interaction history. All data employed in the gen- eration of these descriptions (the image, the discourse) are annotated by a XML-based language. Results obtained in the medical domain (radiology) are presented, and the advantage of annotating knowledge sources are discussed. 1 Introduction The amount of heterogeneous information available on the Web is growing exponen- tially; this growth makes increasingly difficult to find, access, present and maintain information. From research about how to make these tasks easier, methods for mak- ing machine understandable multimedia web resources have emerged: these methods require associating semantics to information, through the use of metadata. The de- scription of such metadata is typically based on a domain conceptualization and a definition of a domain-specific annotation language. An annotation can be loosely defined as “any object that is associated with another object by some relationship” (from the W3C Annotation Working Group). In particular, XML is a standard, pro- posed by the W3C, to create mark-up languages for a wide variety of application domains; developing such languages favours universal storage and interchange formats, re-use and share of resources for web distributed knowledge representation and programming [11]. Metadata annotation of web resources is essential for applying AI techniques for searching and extracting relevant information (by improving a semantic contextual- ized search), for maintaining web resources (by keeping them consistent, correct and