When disasters strike, an urgent task for decision makers and the general public is to consolidate scattered, probably chaotic and incomplete information and then to plan the next step. The information needed to reach a decision or develop a policy is very likely not provided by a single document or a single piece of text, but resides in multiple sources. In other words, a document might be complemented or further elaborated upon by other documents, and all of them, only as a whole, provide a complete and satisfying answer. This project will create a new information chain of support for emergency management. InfoChain is expected to significantly improve the way policy/decision makers, disaster researchers and the general public search emergency-related information by automatically connecting the dots in information. All research results gleaned from this project will be disseminated to the research community through publications, tutorials and open-source software. Data collections will be available for other researchers to use in evaluating their own approaches.
Today's search engines, or more general Information Retrieval (IR) systems are designed for simple information needs that can be satisfied by one document. Documents are ranked independently with regard to their relevance scores to a query. A single document is expected to cover all the query concepts. Unfortunately, this assumption offers little help for connecting different pieces of relevant information, which is crucial to emergency management. The key novelty of this project is the capability to search document tuples, a group of documents that collectively satisfy an information need. Tuple-based retrieval models and result organization that define collective relevance of a document tuple with respect to an information need will help users efficiently examine search results. The document tuple is a more natural and desirable retrieval unit for complicated information needs, and its incorporation into IR retrieval models for emergency management leaps over the single-document-oriented IR practices and is the first such effort. For further information see the project web site at: www.eecis.udel.edu/~hfang/infochain.html
