In this paper we present NewsFlash, an adaptive search system that assists a searcher to efficiently search a library of stored TV news reports. The system automatically summarises the closed-caption subtitles embedded in the TV broadcasts and selects appropriate sentences to best describe report content in respect to the searchers query. During interaction the system selects useful terms from these summaries and uses these terms to update the display and potentially update a stored searcher profile. We evaluate the worth of our approach with real searchers and realistic information seeking scenarios. A novel means of testing the worth of a permanent profile of searchers general interests is also proposed. Our results show that the adaptive techniques we propose can work well in multimedia search environments.
展开▼
