Is Fidel Castro Really an American President? On Set Expansion An - PowerPoint PPT Presentation

The SEAL System Fetcher Extractor Ranker Seeds Web pages Mentions Suggestions Google Wrapper Graph

Ranker

Ranker • Problem: Noise

Ranker • Problem: Noise • Solution: Similarity measure between seeds and mentions

Ranker • Problem: Noise • Solution: Similarity measure between seeds and mentions • Ranked output

Ranker • Problem: Noise • Solution: Similarity measure between seeds and mentions • Ranked output • Understand relation

Analysis

Analysis

Analysis

Analysis

Analysis

Analysis • 4 source types: • document • seed • wrapper • mention

seeds find doc Castro derive derive extract prt..x fgA..gz extract extract JFK

Graph-walk (Page rank) seeds find doc Castro derive derive extract prt..x fgA..gz extract extract JFK

Graph-walk (Page rank) P(find|doc) = 0.5 seeds find doc Castro derive derive extract prt..x fgA..gz extract extract JFK

Graph-walk (Page rank) P(find|doc) = 0.5 seeds P(derive|doc) = 0.5 find doc Castro derive derive extract prt..x fgA..gz extract extract JFK

Graph-walk (Page rank) P(find|doc) = 0.5 seeds P(derive|doc) = 0.5 find P(seeds|doc,find) doc Castro = 1 derive derive extract prt..x fgA..gz extract extract JFK

Graph-walk (Page rank) P(find|doc) = 0.5 seeds P(derive|doc) = 0.5 find P(seeds|doc,find) doc Castro = 1 derive derive extract P(prt..x|doc,derive) prt..x = 0.5 fgA..gz extract extract JFK

Graph-walk (Page rank) P(find|doc) = 0.5 seeds P(derive|doc) = 0.5 find P(seeds|doc,find) doc Castro = 1 derive derive extract P(prt..x|doc,derive) prt..x = 0.5 fgA..gz P(fgA..gz|doc,derive) extract extract = 0.5 JFK

Graph-walk (Page rank) P(find|doc) = 0.5 seeds P(derive|doc) = 0.5 1/2 P(seeds|doc,find) doc Castro = 1 1/4 1/4 extract P(prt..x|doc,derive) prt..x = 0.5 fgA..gz P(fgA..gz|doc,derive) extract extract = 0.5 JFK

Graph-walk (Page rank) seeds find doc Castro Transitions derive 1/2 in both ways 1/4 prt..x fgA..gz extract 1/4 JFK

Transition Matrix s s d w1 w2 m1 m2 find s d m2 d derive derive extract w1 w2 w1 w2 extract extract m1 m1 m2

Transition Matrix x s s d w1 w2 m1 m2 find s d m2 d derive derive extract y w1 w2 w1 w2 extract extract m1 m1 m2

Transition Matrix x s s d w1 w2 m1 m2 find s d m2 d derive derive extract y w1 w2 w1 w2 extract extract m1 m1 m2 (x,y) = P(x → y)

Transition Matrix s s d w1 w2 m1 m2 find s 0 d m2 d derive derive extract w1 w2 w1 w2 extract extract m1 m1 m2

Transition Matrix s s d w1 w2 m1 m2 find s 0 d m2 d 1 derive derive extract w1 w2 w1 w2 extract extract m1 m1 m2

Transition Matrix s s d w1 w2 m1 m2 find s 0 d m2 d 1 derive derive extract w1 0 w2 w1 w2 0 extract extract m1 0 m1 m2 0

Transition Matrix s s d w1 w2 m1 m2 find s 0 ½ d m2 d 1 0 derive derive extract w1 0 ¼ w2 w1 w2 0 ¼ extract extract m1 0 0 m1 m2 0 0

Transition Matrix s s d w1 w2 m1 m2 find s 0 ½ 0 0 0 0 d m2 d 1 0 ½ ½ 0 0 derive derive extract w1 0 ¼ 0 0 ½ 1 w2 w1 w2 0 ¼ 0 0 ½ 0 extract extract m1 0 0 ¼ ½ 0 0 m1 m2 0 0 ¼ 0 0 0

Transition Matrix with lazyness factor λ =0.01 seeds s d w1 w2 m1 m2 find s 0 ½ 0 0 0 0 doc Castro d 1 0 ½ ½ 0 0 derive derive extract w1 0 ¼ 0 0 ½ 1 prt..x fgA..gz w2 0 ¼ 0 0 ½ 0 extract extract m1 0 0 ¼ ½ 0 0 JFK m2 0 0 ¼ 0 0 0

State Vector s s find d d m2 w1 derive derive extract w2 w2 w1 m1 extract extract m2 m1

State Vector s s 1 find d 0 d m2 w1 0 derive derive extract w2 0 w2 w1 m1 0 extract extract m2 0 m1

Transition Matrix and State Vector

Transition Matrix and State Vector · =

Transition Matrix and State Vector · =

Iterated Multiplication · =

Iterated Multiplication · = 1000x

Iterated Multiplication · = 1000x