langreiter.com plain, simple: etzioni-airntkilfg-notes

LANGREITER.COM plain, simple

START

INDEX

etzioni-airntkilfg-notes

CREATED BY chris • LAST EDITED BY chris 922 days AGO

Video: All I Really Need to Know I Learned from Google

Greg Linden's notes

- how to build a program that reads?
- how to accumulate massive knowledge?
- [create Opine]: review miner
- KnowItAll - generate & test (using [create PMI]), bootstrap classifier (@@ how does that come in, exactly?)
- templates & relations
- private index of nontrivial portion of web - quite feasible
- traditional index too slow: built Bindings Engine, 136x faster than Nutch
- noisy-or is lousy! - their urns model much better
- [create KnowItNow]: 33 days runtime (KnowItAll) down to 3 minutes
- problems: no conjunctions, no inferences
- so: [create TextRunner]
- extract as much data as possible (1.2 triples/sentence?)
- local inference through [create Spreading Activation]
- 90m pages, 0.65s/doc, 68.5% precision
- open problems: probabilistic semantics, n-ary relations, precision & recall