Wordnet Ontology as a Wordnet Ontology as a Geographical - PowerPoint PPT Presentation

Wordnet Ontology as a Wordnet Ontology as a Geographical Information Geographical Information Resource Resource Davide Buscaldi, Davide Buscaldi, Dpto. Sistemas Informáticos y Dpto. Sistemas Informáticos y (DSIC) Computación (DSIC) Computación Universidad Politécnica de Universidad Politécnica de Valencia Valencia Valencia, Nov. 15th 2005 Valencia, Nov. 15th 2005

Plan of the talk Plan of the talk • The Geographical Information Retrieval The Geographical Information Retrieval • task task • WordNet (in brief) WordNet (in brief) • • Exploiting WordNet: Exploiting WordNet: • – Query Expansion Query Expansion – – Index Terms Expansion – Index Terms Expansion • Results Results • • Conclusions Conclusions •

The Geographical Information The Geographical Information Retrieval Task Retrieval Task • Actually GIR is ambiguous: Actually GIR is ambiguous: • – (Geographic Information) Retrieval** (Geographic Information) Retrieval** – – Geographical (Information Retrieval)* Geographical (Information Retrieval)* – • In this case: In this case: • – “ – “Retrieval of information involving some kind of Retrieval of information involving some kind of spatial awareness ” ”* (Fred Gey @ GeoCLEF 2005) * (Fred Gey @ GeoCLEF 2005) spatial awareness – E.g. E.g. “ “Find news about Find news about riots riots in France. in France.” ” – • Not to be confused with GIR as a particular Not to be confused with GIR as a particular • aspect of Spatial Information Retrieval** aspect of Spatial Information Retrieval** – E.g. E.g. “ “What is the What is the river river flowing through Paris? flowing through Paris?” ” –

Common GIR issues (1) Common GIR issues (1) • (Almost) The same Geographical Entity can (Almost) The same Geographical Entity can • be indicated in several different (and be indicated in several different (and sometimes ambiguous) manners: sometimes ambiguous) manners: • United Kingdom of Great • United Kingdom of Great Britain and Northern Britain and Northern Ireland Ireland • • United Kingdom, UK, U.K. United Kingdom, UK, U.K. + Ireland, Eire + Ireland, Eire • Great Britain, GB + Ireland • Great Britain, GB + Ireland • Reino Unido, Gran • Reino Unido, Gran Bretagna Bretagna • British Isles • British Isles

Common GIR Issues (2) Common GIR Issues (2) • Missing Missing explicit explicit geographical geographical • information: information: – E.g., consider the following text: E.g., consider the following text: – “On Sunday mornings, the covered market opposite On Sunday mornings, the covered market opposite “ the station in the leafy suburb of Aulnay-sous-Bois Aulnay-sous-Bois - - the station in the leafy suburb of barely half an hour's drive from central Paris Paris - spills - spills barely half an hour's drive from central opulently on to the streets and boulevards.” ” opulently on to the streets and boulevards. Whereas the text is talking about events Whereas the text is talking about events in France, the GE France France itself is never itself is never in France, the GE mentioned. mentioned.

The WordNet Ontology The WordNet Ontology • Lexical resource containing nouns, verbs, Lexical resource containing nouns, verbs, • adjectives and adverbs organized into adjectives and adverbs organized into synonym sets (synsets (synsets ) ) synonym sets – each synset represents one underlying lexical each synset represents one underlying lexical – concept. concept. – various relations link the synonym sets various relations link the synonym sets – • Hypernymy (is-a relation) Hypernymy (is-a relation) • • Meronymy (has-part relation) Meronymy (has-part relation) • • Holonymy (part-of relation) Holonymy (part-of relation) • • Available at Available at • – http://wordnet.princeton.edu/perl/ http://wordnet.princeton.edu/perl/webwn webwn –

Geographical Conceptual Geographical Conceptual Networks in WordNet Networks in WordNet British Isles Ireland UK Great Britain (Hibernia) Ireland N. Ireland (Eire) Wales Scotland Holonym England Meronym

Exploiting WordNet Exploiting WordNet • WordNet can help in addressing most of GIR WordNet can help in addressing most of GIR • issues issues • Solve Solve synonymy synonymy : : • – E.g. synset corresponding to E.g. synset corresponding to “ “ U.K. U.K. ” ”: : – • {United Kingdom, UK, U.K., Great Britain, GB, Britain, {United Kingdom, UK, U.K., Great Britain, GB, Britain, • United Kingdom of Great Britain and Northern Ireland} United Kingdom of Great Britain and Northern Ireland} • Find missing (geographical) information: Find missing (geographical) information: • – Meronymy ( Meronymy (“ “has member/part has member/part” ” relationship) relationship) – – Holonymy ( Holonymy (“ “is member/part of is member/part of” ”) ) – • Two solutions tested: Two solutions tested: • – Query Expansion (QE) Query Expansion (QE) – – Index Terms Expansion (ITE) Index Terms Expansion (ITE) –

Query Expansion Query Expansion • Expand the geographical terms of the Expand the geographical terms of the • query with their synonyms and (some) query with their synonyms and (some) meronyms meronyms – Geographical terms are identified through Geographical terms are identified through – the WordNet ontology (words having the the WordNet ontology (words having the synset {region, location} among their synset {region, location} among their hypernyms hypernyms – Meronyms containing the word Meronyms containing the word “ “ capital capital ” ” in in – the definition ( gloss gloss ) or in the meronym ) or in the meronym the definition ( synset itself synset itself

Query Expansion - Example Query Expansion - Example • “ “Foreign minorities in Germany Foreign minorities in Germany” ” • – “ “Germany Germany” ” appears in the synset: appears in the synset: – {Germany, Federal Republic of Germany, {Germany, Federal Republic of Germany, Deutschland, FRG} Deutschland, FRG} – The following meronyms contain the word The following meronyms contain the word – “capital capital” ”: : “ • Berlin, german Berlin, german capital capital • • Bonn (was the Bonn (was the capital capital of Germany between of Germany between • 1949 and 1989) 1949 and 1989) • Munich, Muenchen ( Munich, Muenchen (capital capital of Bavaria) of Bavaria) • • Aachen, Aken, Aix-la-Chapelle (formerly Aachen, Aken, Aix-la-Chapelle (formerly • Charlemagne northern capital capital) ) Charlemagne northern

Index Terms Expansion Index Terms Expansion • Find geographical terms in the text collection Find geographical terms in the text collection • – openNLP openNLP Named Entities detector Named Entities detector – (http://opennlp.sourceforge. http://opennlp.sourceforge.net net) ) ( • Put all their holonyms and synonyms into a Put all their holonyms and synonyms into a • special geo geo index index special – Search Engine used: Lucene Search Engine used: Lucene – (http://lucene.jakarta.org http://lucene.jakarta.org) ) ( • Label geographical terms in the query with the Label geographical terms in the query with the • geo search field: search field: geo – E.g. E.g. “ “riots in France riots in France” ” -> text:riots geo:France -> text:riots geo:France –

Index Terms Expansion - Index Terms Expansion - Example Example “On On Sunday mornings Sunday mornings, the , the covered market opposite covered market opposite “ the station station in the in the leafy suburb leafy suburb of of Aulnay-sous-Bois Aulnay-sous-Bois - - the barely half an hour's half an hour's drive drive from from central central Paris Paris - - spills spills barely opulently on to the on to the streets streets and and boulevards boulevards. .” ” opulently From WordNet: From WordNet: Paris, Paris, French capital French capital, , capital of France capital of France, , city of light city of light � � France, French Republic France, French Republic � � Europe Europe � � � Northern hemisphere Northern hemisphere � - To standard index - To standard index - To geographical index - To geographical index

Experiment Setup Experiment Setup • GeoCLEF 2005 collection and queries GeoCLEF 2005 collection and queries • – Los Angeles Times 1994 Los Angeles Times 1994 – – Glasgow Herald 1995 Glasgow Herald 1995 – • “ “Topic Description Topic Description” ” runs: runs: • – Typical TD from queries: Typical TD from queries: – • “ “Shark attacks near California and Australia Shark attacks near California and Australia” ” • • “ “Vegetable exporters of Europe Vegetable exporters of Europe” ” • • “ “Holidays in the Scottish Trossachs Holidays in the Scottish Trossachs” ” • • 1000 results returned for each query 1000 results returned for each query •

Results - Query Expansion Results - Query Expansion Clean System with QE 100% 80% Precision 60% 40% 20% 0% 0 1 2 3 4 5 6 7 8 9 10 Recall levels