Summary Generation for Temporal Extractions

Recent advances in knowledge harvesting have enabled us to collect large amounts of facts about entities from Web sources. A good portion of these facts have a temporal scope that, for example, allows us to concisely capture a person’s biography. However, raw sets of facts are not well suited for presentation to human end users. This paper develops a novel abstraction-based method to summarize a set of facts into natural-language sentences. Our method distills temporal knowledge from Web documents and generates a concise summary according to a particular user’s interest, such as, for example, a soccer player’s career. Our experiments are conducted on biography-style Wikipedia pages, and the results demonstrate the good performance of our system in comparison to existing text-summarization methods.