SINAI-GIR A Multilingual Geographical IR System University of Jaén (Spain) José Manuel Perea Ortega CLEF 2008, 18 September, Aarhus (Denmark) Computer.

SINAI-GIRSINAI-GIR

A Multilingual Geographical IR SystemA Multilingual Geographical IR System

University of Jaén (Spain)

José Manuel Perea Ortega

CLEF 2008, 18 September, Aarhus (Denmark)

Computer Science Department

Introduction

• Preliminary work of SINAI in GeoCLEF: – 2006: query expansion using gazetteers and

thesaurus [García-Vega et al., 2007]– 2007: filtering documents based on manual rules

[Perea-Ortega et al., 2007]

• GeoCLEF 2008:– Filtering documents using new manual rules and

new approachs (query reformulation, keywords and hyponyms extraction, query geo-expansion)

GeoCLEF 2008, Aarhus

Multilingual Query

English collection

IR SubsystemIR Subsystem

GeoNames

Final Re-Ranked Documents retrieved

TRANSLATORTRANSLATOR QUERY ANALYZERQUERY ANALYZER

English Query (Q)

Q1

Q2Q3

Collection Collection PreprocessingPreprocessing

subsystemsubsystem

GeoNames

VALIDATORVALIDATOR

Documents retrieved

Keywords and geo-information

extracted


extracted

SINAI-GIR System overview

Multilingual Query

English collection


GeoNames



English Query (Q)

Q1

Q2Q3


subsystemsubsystem

GeoNames

VALIDATORVALIDATOR

Documents retrieved


extracted


extracted


Translates the queries from other languages into English

We have used SINTRAM (SINai TRAnslation Module) [García-Cumbreras et al., 2007]

It works with different online machine translators

Multilingual Query

English collection


GeoNames



English Query (Q)

Q1

Q2Q3


subsystemsubsystem

GeoNames

VALIDATORVALIDATOR

Documents retrieved


extracted


extracted


Preprocessing: stemming, stopwords, POS The toponyms are extracted (NER) Two indexes are generated:

• Locations• Keywords

Multilingual Query

English collection


GeoNames



English Query (Q)

Q1

Q2Q3


subsystemsubsystem

GeoNames

VALIDATORVALIDATOR

Documents retrieved


extracted


extracted


Query Preprocessing: stemming, stopwords, removes irrelevant information

The toponyms are extracted (NER) Spatial relations finder based on manual rules Query reformulation based on POS tagging and

query parsing subtask Geo-expansion using a gazetteer Keywords/Hyponyms detection

Multilingual Query

English collection


GeoNames



English Query (Q)

Q1

Q2Q3


subsystemsubsystem

GeoNames

VALIDATORVALIDATOR

Documents retrieved


extracted


extracted


Lemur as index-search engine

Okapi with PRF as weighting function

Multilingual Query

English collection


GeoNames



English Query (Q)

Q1

Q2Q3


subsystemsubsystem

GeoNames

VALIDATORVALIDATOR

Documents retrieved


extracted


extracted


Filter the list of documents recovered by the IR subsystem, applying different manual rules and using the geographical data detected in the query

Re-rank the documents using predefined weights for each rule and the keywords/hyponyms detected in the query

Experiments description

• SINAI has participated in mono and bilingual tasks with a total of 15 experiments15 experiments:– MONO-EN: 9 experiments– BILI-X2EN: 6 experiments

• Combining the content of topic labels: TD or TDN• BaselineBaseline: Q1 without applying any filtering or re-

ranking process• Other experimentsOther experiments:

– Filtering and re-ranking of the fusion list of the documents recovered by the Q1, Q2 and Q3

– Using keywords and/or hyponyms in the re-ranking process


MONO-EN results


Best result: baselinebaseline (no filtering and no re-ranking)

In some filtering experiments the use of keywords improves the results

Best results using only the TD topic labels

BILI-X2EN results


Best result: baselinebaseline (no filtering and no re-ranking) with Portuguese topics

Best results using only the TD topic labels

Conclusions

• The baseline experiment seems to work well because we include the geo-information in the retrieval process

• The filtering of documents does not seem to work well because we include the geo-information in the query and we are re-ranking documents which maybe are not relevant with respect to their content

• The use of keywords for re-ranking the documents retrieved could be interesting because in some experiments it improves the results obtained without using them

• Query reformulation could be also interesting because for some topics it retrieves valid documents which are not retrieved with the default query


TextMESS at GeoCLEF 2008

• Spanish TextMESS projectTextMESS project (Intelligent, Interactive and Multilingual Text Mining based on Human Language Technologies): joint participation by the Polytechnic University of Valencia and University of Jaén (SINAI)

• Method employed: merging algorithm based on merging algorithm based on fuzzy Borda voting schemefuzzy Borda voting scheme, taking as input the , taking as input the two document lists returned by both systemstwo document lists returned by both systems

• Second best result in the monolingual English task


Thank you


sinai.ujaen.es

• References

– García-Vega, Manuel and García-Cumbreras, Miguel A. and Ureña-López, L.A. and Perea-Ortega, José M. GEOUJA System. The first participation of the University of Jaén at GEOCLEF 2006. In LNCS, volume 4730, pages 913-917. Springer-Verlag, 2007.

– Perea-Ortega, Jose M. and García-Cumbreras, Miguel A. and García-Vega, Manuel and Montejo-Ráez, Arturo. GEOUJA System. University of Jaén at GEOCLEF 2007. In Proceedings of the Cross Language Evaluation Forum (CLEF 2007), page 52, 2007.

– García-Cumbreras, Miguel A. and Ureña-López, L. Alfonso and Martínez-Santiago, Fernando and Perea-Ortega, José M. BRUJA System. The University of Jaén at the Spanish task of QA@CLEF 2006. In LNCS, volume 4730, pages 328-338. Springer-Verlag, 2007.


http://sinai.ujaen.es

SINAI-GIR A Multilingual Geographical IR System University of Jaén (Spain) José Manuel Perea Ortega CLEF 2008, 18 September, Aarhus (Denmark) Computer.

Documents

query expansion

query rerank

query geoexpansion geoclef

filtering documents

list of documents

aarhus slide

locations keywords

subtask geoexpansion