Deriving event relevance from the ontology constructed with formal concept analysis

Abstract

In this paper, we present a novel approach to derive event relevance from event ontology constructed with Formal Concept Analysis (FCA), a mathematical approach to data analysis and knowledge representation. The ontology is built from a set of relevant documents and according to the named entities associated to the events. Various relevance measures are explored, from binary to scaled, and from symmetrical to asymmetrical associations. We then apply the derived event relevance to the task of multi-document summarization. The experiments on DUC 2004 data set show that the relevant-event-based approaches outperform the independent-event-based approach. © Springer-Verlag Berlin Heidelberg 2006.