When: Thursday, October 05, 1:00 p.m.

Where: 1109 Newell-Simon Hall

Vasco Pedro

LTI Ph.D Thesis Proposal

Abstract:
In the last decades the number of available ontologies has grown considerably. These resources offer the promise of easily-accessible, open-domain ontological information, but the existence of such diverse ontologies raises the issue of information merging and reuse. A comparison of available ontologies reveals both redundant and complementary coverage, but the variety of frameworks and languages used for ontology development makes it a challenge to merge query results from different ontologies. This research proposes to address this problem by building an ontological middleware level for only small fragments of ontologies in an on-demand basis by querying multiple ontologies and merging the query results from multiple knowledge base systems. We then follow ontological chains and inferences across ontologies, using partial query results from one ontology to query another. This is a more complex version of cross-data-base joins, where the data schemas are sufficiently compatible. An initial evaluation used the federated ontology search for answer type checking in question answering. The results of the evaluation show that it is possible to obtain results that outperform querying ontologies independently in both precision and recall. Solutions to additional problems in querying multiple ontologies such as concept identification and ontology selection will be provided using string and structural similarity measures. We propose the creation of an evaluation framework using the question and answer sets in the TREC evaluations to account for a wider set of ontological operations.

<< Back