My advisor, Oren Etzioni, gave the second keynote at WSDM today. His main focus is on his latest work which is a new paradigm in search using open extraction.
His hypothesis: Use Machine Reading - Information Extraction + tractable inference. For example, who did what? E.G.
IE(sentence) = who did what? - speaker(Alon Halevy, UW)
Inference = uncover implicit information - Will Alon visit Seattle? (maybe… is UW University of Washington or University of Waterloo?)
His argument, in part based on systems like Opine (done by Ana-Maria Popescu, another of Oren’s students and now at Yahoo Research), is information extraction enables very rich applications to be built.
How do you make it work on the Web? Open IE - Self-supervised Information Extraction (Banko, Cafarella, Soderland, et al, IJCAI ‘07). This leads him to find triples of (Noun-Phrase, Verb, Noun-Phrase).
More later (possibly even references… y’know, auto-tagging on blogs would actually be really useful…)
