Combining multiple sources of evidence in XML multimedia documents: An inference network incorporating element language models

Abstract

This work makes use of the semantic structure and logical structure in XML documents, and their combination to represent and retrieve XML multimedia content. We develop a Bayesian network incorporating element language models for the retrieval of a mixture of text and image. In addition, an element-based collection language model is used in the element language model smoothing. The proposed approach was successfully evaluated on the INEX 2005 multimedia data set. © Springer-Verlag Berlin Heidelberg 2007.