Kno wledge Disco v ery in Spatial Data b y Means of ILP - PowerPoint PPT Presentation

Kno wledge Disco v ery in Spatial Data b y Means of ILP Lub o� Pop el�nsk� Masaryk University Brno and CTU Pr ague, Cze chia Email: popel@fi.muni.cz Motiv ation � Inductiv e logic programm ing and inductiv e query languages � Description of (ma yb e) inexactly de�ned geographic ob jects

Kno wledge Disco v ery in Spatial Data b y Means of ILP Lub o� Pop el�nsk� Masaryk University Brno and CTU Pr ague, Cze chia Email: popel@fi.muni.cz Outline 1. Inductiv e query language 2. Metho d & WiM 3. Examples 4. Discussion & F uture researc h

BRIDGE LINEAR PLANAR Object1 Geometry Geometry Object 2 FORESTRY BUILDING HIGHWAY_BRIDGE RIVER ROAD RAILWAY Named Named Ob ject-orien ted database sc hema State FOREST_HOUSE FOREST WOOD Importance Forest

Ra w data 4

inductiv e query class descriptions (in F-logic) ob ject descriptions (in F-logic) ? TRANSLA TE � @ � @ R Bac kground kno wledge Example set T yp e de�nitions @ � @ R � WiM ? result of inductiv e query (Horn clauses) GW iM sc hema 5

WiM inductiv e learner e�cien t searc hing for the re�nemen t graf shift of syntactic bias generator of near-misses oracles needs from 2 to 4 examples for most of the ILP b enc hmark predicates (list pro cessing) smaller dep endency on the qualit y of the example set in comparison to some of ILP programs has b een tested b oth on go o d examples and on randomly c hosen example sets. 6

Inductiv e language extract characteristic rule �� �� extract < KindOfRule > rule extract discriminate rule for < NameOfT arget > �� �� from [ < ListOfClasses > ] �� �� [ < Constrain ts > ] [ from p oin t of view < Domain > ] extract dependency rule �� �� �� �� 7

Discrimination of F orests and W o o ds Find a di�erence b et w een forests and w o o ds from the p oin t of view of area. ar ea is the name of set of predicates lik e ar ea ( Geometr y ; Ar ea ) . extract discriminate rule forest(F) :- for isF orest from forest geometry(F,GForest), area(GForest,Area), in con trast to w o o d 100 < Area. from p oin t of view area. Relation b et w een F orests and W o o ds Find a relation b et w een forests and w o o ds from the p oin t of view of area. ar ea is the name of set of predicates lik e ar ea ( Geometr y ; Ar ea ) . extract dep endency rule forestOrWood(F,W) :- for forestOrW o o d geometry(F,GF),area(G F,F A), geometry(W,GW), area(GW,WA), from forest, w o o d WA<GA. from p oin t of view area. 8

Discussion 1. The query language is quite p o w erful ) quite complex queries can b e form ulated. Ho w ev er , the price that user has to pa y for is sometim es to o big. 2. Ho w to pro cess large amoun t of data F uture researc h � In terface to P ostgreSQL ob ject-relational DBMS � Geographic domain kno wledge 9