Topic-driven multi-document summarization with encyclopedic knowledge and spreading activation

摘要: Information of interest to users is often distributed over a set documents. Users can specify their request for information as query/topic -- one or more sentences questions. Producing good summary the relevant relies on understanding query and linking it with associated To "understand" we expand using encyclopedic knowledge in Wikipedia. The expanded linked its documents through spreading activation graph that represents words grammatical connections these topic activated nodes are used produce an extractive summary. method proposed tested DUC summarization data. system implemented ranks high compared participating systems competitions, confirming our hypothesis useful addition system.