This work is directed toward the creation of a system for automatically summarizing documents by extracting selected sentences. Several heuristics including position, cue words, and title words are used in conjunction with lexical chain information to create a salience function that is used to rank sentences for extraction. Compiler technology, including the Flex and Bison tools, is used to create the AutoExtract summarizer that extracts and combines this information from the raw text. The WordNet database is used for the creation of the lexical chains. The AutoExtract summarizer performed better than the Microsoft Word97 AutoSummarize tool and the Sinope commercial summarizer in tests against ideal extracts and in tests judged by humans.
展开▼
