Using Abductive Inference and Dynamic Indexing to Retrieve Multimedia SGML Documents

Abstract

The retrieval of complex multimedia items such as SGML-structured texts can be facilitated by means of a formal representation of knowledge about these data. These information sources must be aggregated dynamically at the time of query processing. In this paper, an interactive, probabilistic retrieval system is proposed, comprising an extended Bayesian network, a multimedia indexing component and an abductive retrieval engine. The inference process exploits and controls the index structure of the network. The prototype has been tested on a collection of SGML structured dictionary articles. An example is presented in the last section of the paper.