Discovering aspect-based correlation of Web contents for cross-media information retrieval

The main issue regarding cross-media information retrieval is the determination of correlations between different types of media objects. The conventional approach derives the correlations based on common properties extracted from media contents or synchronous presentation of multiple media pre-authored in a scheduled scenario. We propose a novel approach for determining the cross-media correlation derived from the referential contexts of media objects in the Web. A Web page links to the media objects distributed over the Web so that it aggregates them with respect to the page content. Our approach extracts the referential context by analyzing the logical structure of the Web and discovers the aspect of a media object, which means the latent semantics of the referential context. The aspect-based correlation reveals the relation between media objects regarding their reputations on the Web. In this paper, we propose an approach for discovering aspect-based correlations with an experimental implementation