Contextual information retrieval using concept chain graphs

This paper discusses concept chain graphs, a new framework for information retrieval that supports sophisticated contextual queries. Concept chain graphs subsume traditional information retrieval models, but extend them by supporting (i) more sophisticated content representation reflecting information extraction output, and (ii) more sophisticated retrieval algorithms including probabilistic graph models and graph mining. Concept chain graphs are designed specifically for applications involving unapparent information revelation (UIR). UIR manifests itself when information generated by multiple authors working independently at different times may together reveal more information than apparent. A key to UIR is connecting information trails that span multiple documents. This requires the support of sophisticated models of context, including cross-document context. Three types of queries are discussed in this paper: (i) concept-based queries using ontologies, (ii) concept chain queries which find the best evidence trail connecting two concepts across documents, and (iii) concept graph queries which reflect more complex patterns. Examples from processing the 9-11 corpus are discussed.