Mediator y e r u Q d e t a l u Query m r o e f K L - PowerPoint PPT Presentation

Motivation Approach Search Scoring Experiments Related Work Conclusions O r b i t z F l i g h t lowestFare(“MXP”,“HYD”) S e a r c h Mediator y e r u Q d e t a l u Query m r o e f K L M R Reformulated Query O n l i n e SELECT MIN(price) calcPrice(“MXP”,“HYD”,”economy”) FROM flight Q a n t a s WHERE depart=“MXP” S p e c i a l s AND arrive=“HYD” Source Definitions: - Orbitz Flight Search c e : e r v i w S N e - KLM Online Alitalia - Qantas Specials Generate Model of Service?

Motivation Approach Search Scoring Experiments Related Work Conclusions K K K n o n o n o w n w n w n S S S o u r o u r o u r c e c e c e 1 2 3 New Source 4 distance zipcode

Motivation Approach Search Scoring Experiments Related Work Conclusions K K K n o n o n o w n w n w n S S S o u r o u r o u r c e c e c e 1 2 3 centroid(zip1, lat1, long1), centroid(zip2, lat2, long2), greatCircleDist(lat1, long1, lat2, long2, dist2), convertKm2Mi(dist1, dist2). source1(zip1, lat1, long1), source1(zip2, lat2, long2), source2(lat1, long1, lat2, long2, dist2), source3(dist2, dist).

Motivation Approach Search Scoring Experiments Related Work Conclusions match 80210 90266 842.37 843.65 60601 15201 410.31 410.83 10005 35555 899.50 899.21

Motivation Approach Search Scoring Experiments Related Work Conclusions e w N Target e u r c Tuples S o Known Source Known Source Candidate Tuples

Motivation Approach Search Scoring Experiments Related Work Conclusions k ! r o W s u o i v e P r Lerman, Plangprasopchok and Knoblock. Automatically labeling data used by web services. AAAI’06. e w N Target e u r c Tuples S o Known Source Known Source Candidate Tuples

Motivation Approach Search Scoring Experiments Related Work Conclusions Expressive Language Sufficient for modeling most online sources 1. Sample the new source Invoke target with set of random inputs; Add empty clause to queue ; while ( queue not empty) v := best definition from queue ; forall ( v’ in Expand( v ) ) if ( Eval( v’ ) > Eval( v ) ) insert v’ into queue ; 2. Best-first search through space of candidate definitions

Motivation Approach Search Scoring Experiments Related Work Conclusions New Source 5 Input Output <zip1, dist1> <zip2, dist2> Non-empty randomly Result generated input Empty tuples Result

Motivation Approach Search Scoring Experiments Related Work Conclusions New Source 5 source5 (_,_,_,_). Expand source5( $zip1,$dist1,zip2,dist2) source5 (zip1,_,_,_) :- source4 (zip1,zip1,_). source5 (zip1,_,zip2,dist2) :- source4 (zip2,zip1,dist2). source5 (_,dist1,_,dist2) :- < (dist2,dist1). …

Motivation Approach Search Scoring Experiments Related Work Conclusions New Source 5 source5 (_,_,_,_). Expand source5( $zip1,$dist1,zip2,dist2) source5 (zip1,_,_,_) :- source4 (zip1,zip1,_). source5 (zip1,_,zip2,dist2) :- source4 (zip2,zip1,dist2). source5 (_,dist1,_,dist2) :- < (dist2,dist1). Expand … source5 (zip1,dist1,zip2,dist2) :- source4 (zip2,zip1,dist2), source4 (zip1,zip2,dist1). source5 (zip1,dist1,zip2,dist2) :- source4 (zip2,zip1,dist2), < (dist2,dist1). …

Motivation Approach Search Scoring Experiments Related Work Conclusions Standard techniques Non-standard technique

Motivation Approach Search Scoring Experiments Related Work Conclusions Input Target Output Clause Output <$zip1, $dist1> <zip2, dist2> <zip2, dist2> No Overlap No Overlap Overlap!

Motivation Approach Search Scoring Experiments Related Work Conclusions At least half of input tuples are non-empty invocations of target forall (tuple in InputTuples ) T_target = invoke (target, tuple) T_clause = execute (clause, tuple) if not (| T_target |=0 and | T_clause |=0) Jaccard similarity Average results only fitness = when output is returned return average( fitness )

Motivation Approach Search Scoring Experiments Related Work Conclusions 10.6 km ≈ 10.54 km Google Inc. ≈ Google Incorporated Mon, 31. July 2006 ≈ 7/31/06

Motivation Approach Search Scoring Experiments Related Work Conclusions Inductive search bias: • Max clause length: 7 • Predicate repetition: 2 • Max variable level: 5 • Executable candidates • No variable repetition Equality Approximations: • 1% for distance , speed , temperature & price • 0.002 degrees for latitude & longitude • JaroWinkler > 0.85 for company , hotel & airport • hand-written procedure for date .

Motivation Approach Search Scoring Experiments Related Work Conclusions Distinguished forecast from current conditions current price = yesterday’s close + change

Motivation Approach Search Scoring Experiments Related Work Conclusions

Motivation Approach Search Scoring Experiments Related Work Conclusions

Motivation Approach Search Scoring Experiments Related Work Conclusions g e r m b o l o B Yahoo c y e n r r C u Exchange s e R a t Rates e i d d w r l W o US Hotel s e a l D e l H o t Rates l s o t e H 5 * Hotels By e t a t S B y Zipcode e n c a i s t D n Great Circle e e Centroid w B e t s of Zipcode d e Distance c o i p Z g l e o G o e l o t H Government c h a r S e Hotel List

Motivation Approach Search Scoring Experiments Related Work Conclusions