Multiple Ontology-based Indexing of Multimedia Documents on the World Wide Web

Abstract

In order to cope with the growing need to search multimedia documents with precision on the Web, we propose a multimedia conceptual indexing framework incorporating semantic relations between annotation words. To do this, we utilize our DOM Tree-based Webpage segmentation algorithm to automatically extract surrounding textual information of the multimedia documents in Webpages. Next, we employ knowledge represented in multiple ontologies to discover the latent semantic dimensions of the surrounding textual information. As a consequence, indexes (represented as semantic networks) are constructed where nodes of each network capture words that exist in the ontologies and edges represent the semantic relations that hold between those words. To address the semantic heterogeneity problem between the produced networks, we employ a multi-level merging algorithm that combines heterogeneous networks into a more coherent network. Additionally, we utilize concept-relatedness measures to address the issue of unrecognized entities by the ontologies. We evaluate the techniques of the proposed framework using three different multimedia dataset types. Experimental results indicate that the proposed techniques are effective and precise.