Plan for today Introduction Graph Matching Method Theme Recognition Comparison Conclusion
Introduction Fact: growth of information sources Problem: impossible to read everything Assumption: documents are structured Solution: automated summary!
Graph Matching Mani & Bloedorn, 1997, MITRE company - Word is a node -Adjacency links -For each phrase: -Find different -Find common -FSD algorithm: -C- and D-score -Phrase links Articles Graphs Words Summary? Phrases -Name links -Same links -Spreading act. -Select best ones
Theme Recognition McKeown et al., 1999, Columbia Univ. -break into paragraphs -find similarities -preprocess phrases -match phrase trees -make grammar trees -break into phrases -construct sentences -cluster similar para’s Articles Themes Phrases Summary Sentences
Comparison 5 issues –Content Representation –Information Fusion –Semantics Preservation –Scalability –Domain Independence
Content Representation Graph Matching –keeps doc’s apart –scale: word Theme Recognition –all doc’s on one pile –scale: paragraph
Information Fusion Graph MatchingTheme Recognition
Semantics Preservation Graph MatchingTheme Recognition
Scalability Graph MatchingTheme Recognition
Domain Independence Graph MatchingTheme Recognition
Conclusion