240 likes | 258 Views
Explore passage extraction and spatial compatibility through syntactic and semantic analysis in geographic documents, focusing on time, phenomenon, and space queries.
E N D
Geographic reference analysis for geographic document querying F.Bilhaut , T.Charnois, P.Enjalbert & Y.Mathet {bilhaut, charnois, enjalbert, mathet}@info.unicaen.fr GREYC, CNRS UMR 6072 University of Caen
The "GéoSem" project • Passage extraction from geographical documents • From a query to a ranked set of passages • Queries are concerned with : - time - phenomenon - space
Excerpt from "Hérin" corpus From 1965 to 1985, the number of high-school students has increased by 70%, but at different rythms and intensities depending on academies and departments. Lower in South-West and Massif Central, moderate in Brittany and Paris, the rise has been considerable in Mid-West and Alsace. […] Also occurs the schooling duration increase which was more important in departments where, in the middle of the 60's, study continuation after primary school was far from beeing systematic.
Excerpt from "Hérin" corpus From 1965 to 1985, the number of high-school students has increased by 70%, but at different rythms and intensities depending on academies and departments. Lower in South-West and Massif Central, moderate in Brittany and Paris, the rise has been considerable in Mid-West and Alsace. […] Also occurs the schooling duration increase which was more important in departments where, in the middle of the 60's, study continuation after primary school was far from beeing systematic. Time
Excerpt from "Hérin" corpus From 1965 to 1985, the number of high-school students has increased by 70%, but at different rythms and intensities depending on academies and departments. Lower in South-West and Massif Central, moderate in Brittany and Paris, the rise has been considerable in Mid-West and Alsace. […] Also occurs the schooling duration increase which was more important in departments where, in the middle of the 60's, study continuation after primary school was far from beeing systematic. Time Phenomenon
Excerpt from "Hérin" corpus From 1965 to 1985, the number of high-school students has increased by 70%, but at different rythms and intensities depending on academies and departments. Lower in South-West and Massif Central, moderate in Brittany and Paris, the rise has been considerable in Mid-West and Alsace. […] Also occurs the schooling duration increase which was more important in departments where, in the middle of the 60's, study continuation after primary school was far from beeing systematic. Time Phenomenon Space
Queries • Which passages address educational difficulties in west of France in the 50's ? • Which passages address variations of the number of pupils in rural areas ? • Which passages address Calvados district?
Queries • Which passages address educational difficultiesin west of Francein the 50's? • Which passages address variations of the number of pupilsin Paris area? • Which passages address Calvados district?
Some Signifiant Spatial Expressions Paris in north of France from south of Loire Some seabord towns The quarter of The districts in north of France Fifteen All Some seabord towns of Normandy The most rural districts situated from south of Loire
The type "zone"a georeferenced area anchored in a named place Paris in north of France Normandy From Normandy to Alsace from south of Loire
The ‘LocGeo’ type • The canonical form: [quantification]+[type]+[zone] Quant Type Zone qualification administrative Position named geo. entity The quarter of / districts in north of France Fifteen / All / Some seabord towns of Normandy The most rural districts situated from south of Loire Some seabord towns
The ‘LocGeo’ type quant type zone Quant Type Zone qualification administrative Position named geo. entity The quarter of / districts in north of France Fifteen / All / Some seabord towns of Normandy The most rural districts situated from south of Loire Some seabord towns
Semantic Representation « Paris » ty_zone: town egn: nom: Paris zone: loc: internal Lat: 45.633333 coord: Long: 5.733333
Semantic Representation « Some seabord towns in north of Normandy » type: relative quant: ty_zone: town type: geo: seabord locgeo: ty_zone: region egn: nom: Normandy zone: loc: internal position: north
Implementation and (first) Results • A tokenisation and a morphological analysis • A DCG to perform altogether syntactic and semantic analysis• the grammar contains 160 rules• an internal lexical base of 200 entries• a gazetteer of 100000 named places (France) • 9OO expressions recognised and analysed from a geographical corpus (200 text pages) • Good results but a precise and quantitative evaluation to be done
Semantic matching : Why ? corpora […] the south of a Bordeaux-Genève line […] Text A […] the northern half of France […] 3 a query 1 […] In Paris and Toulouse […] "Which passages address Paris ?" 2 Text B […] In Ile de France region […]
Semantic matching : How ? • Spatial compatibility : Is the zone denoted by the passage spatially compatible with the one of the query? (is there, at least, an intersection?) • Relevance degree : if this zone is compatible, how relevant is it w.r.t.the query? - probability - granularity
Compatibility computation • Q1) Which passages address Paris ? • P1) […] the capital city […] • P2) […] big cities in France. • P3) […] the northern half of France […] • P4) […] South of a Bordeaux-Genève line. YES gazetteer YES gazetteer + computation YES NO gis+computation
GIS GIS Relevance degree (1)Quantification Query= "Calvados" (french district) P1= "The quarter of districts in north of France" P2= "All districts in north of France" P3= "Some districts in north of France" P4= "Fifteen districts in north of France" rank 3 r=25% 1 r=100% 4 r=i/n=5/52=9.6% 2 r=i/n=15/52=29%
Relevance degree (2)Granularity country region district city "zone" ’the northern half of France’ "Basse Normandie" "Caen" "Calvados"
locgeo(locgeo:(det:Det..type:Type..Zone)) --> #prep, det(Det), type(Type), zone(Zone). det(Sem) --> [X],{lexique(X,[X|R],det,Sem)}. type(X) --> typeQualif(X). type(ty_zone:N) --> nomtype(N). typeQualif(ty_zone:N..Q) --> option, nomtype(N), #prep, qualif(Q). nomtype(Sem) --> [X], {lexique(X,[X|R],nom,Sem)}. zone(X)--> egn(X). egn(egn:(ty_zone:T..nom:Y..coord:C)) --> --> ls_lexiconExtDCG(np, type_sem:egn..type_zone:T..nom:Y..coord:C ). egn(egn:(ty_zone:T..nom:Y)) --> [X],{lexique(X,[X|R],np, type_sem:egn..type_zone:T..nom:Y)}.
lexique(quelque,[quelque],det,type_sem:relatif..type:relatif_qualifielexique(quelque,[quelque],det,type_sem:relatif..type:relatif_qualifie ..nb:'qualitatif:faible'). lexique(tout,[tout,le],det,type_sem:exhaustif). lexique(région,[région],nom,type_sem:zone(administrative) ..nom_zone:région). lexique(ville,[ville],nom,type_sem:zone(administrative) ..nom_zone:ville). Lexique('Bretagne',['Bretagne'], np,type_sem:egn..type_zone:région..nom:'Bretagne').