Towards Contextual and Structural Relevance Feedback in XML Retrieval Lobna Hlaoua IRIT (Institut de...

18
Towards Contextual and Structural Relevance Feedback in XML Retrieval Lobna Hlaoua IRIT (Institut de Recherche en Informatique de Toulouse) Equipe SIG-RI (Systèmes d’Informations Généralisées) 118, route de Narbonne - 31062 Toulouse cedex 04

Transcript of Towards Contextual and Structural Relevance Feedback in XML Retrieval Lobna Hlaoua IRIT (Institut de...

Page 1: Towards Contextual and Structural Relevance Feedback in XML Retrieval Lobna Hlaoua IRIT (Institut de Recherche en Informatique de Toulouse) Equipe SIG-RI.

Towards Contextual and Structural Relevance Feedback in

XML Retrieval

Lobna HlaouaIRIT (Institut de Recherche en Informatique de Toulouse)

Equipe SIG-RI (Systèmes d’Informations Généralisées)

118, route de Narbonne - 31062 Toulouse cedex 04

Page 2: Towards Contextual and Structural Relevance Feedback in XML Retrieval Lobna Hlaoua IRIT (Institut de Recherche en Informatique de Toulouse) Equipe SIG-RI.

Outline

• Context

• Relevance Feedback in XML Retrieval

• Contextual Relevance Feedback

• Structural Relevance Feedback

• Conclusion & prospects

Page 3: Towards Contextual and Structural Relevance Feedback in XML Retrieval Lobna Hlaoua IRIT (Institut de Recherche en Informatique de Toulouse) Equipe SIG-RI.

Context: XML Retrieval

-- --- -----------------

-- --- ------------------- ---

-----------------

-- --- ----------------- -- ---

-----------------

-- --- -----------------

-- --- -----------------

-- --- -----------------

-- --- -----------------

-- --- -----------------

-- --- -----------------

-- --- -----------------

-- --- -----------------

-- --- -----------------

Traditional IR

- Document is atomic unit

- user can be submerged by noisy subjects

Page 4: Towards Contextual and Structural Relevance Feedback in XML Retrieval Lobna Hlaoua IRIT (Institut de Recherche en Informatique de Toulouse) Equipe SIG-RI.

Context: XML Retrieval

<book date-publi=‘2000’><title> Ontologies </title><author> J.Dupond</author><chapter>

<title> history of ontology </title><section num= "1" >

<title> Introduction</title><para> ...ontology should be seen only as

an interdiscipline... </para></section><section num= " 2 " > <title> What is Ontology</title>

<para> ...An ontology is an explicit specification of a conceptualization... </para>

<para> In the philosophical sense, we may refer to an ontology as a particular system of categories accounting for a certain vision of the world….. </para>

</section></chapter><chapter> …. </chapter></book>

XML Retrieval

- different granularities

Page 5: Towards Contextual and Structural Relevance Feedback in XML Retrieval Lobna Hlaoua IRIT (Institut de Recherche en Informatique de Toulouse) Equipe SIG-RI.

Context: XML Retrieval

<article date-publi=‘2000’><title> Ontologies </title><author> J.Dupond</author><chapter>

<title> history of ontology </title><section num= "1" >

<title> Introduction</title><para> ...ontology should be seen only as

an interdiscipline... </para></section><section num= " 2 " > <title> What is Ontology</title>

<para> ...An ontology is an explicit specification of a conceptualization... </para>

<para> In the philosophical sense, we may refer to an ontology as a particular system of categories accounting for a certain vision of the world….. </para>

</section></chapter><chapter> …. </chapter></article>

« ontologies case study »

- CAS (Content And Structure) Ex:

- CO (Content Only) Ex:

« //article[about(.,ontologies)]//sec[about(., ontologies case study)] »

«//article[about(.,ontologies)]»

«//article[about(.,ontologies)]//para[about(., ontologies case study)] »

<para> In the philosophical sense, we may refer to an ontology as a particular system of categories accounting for a certain vision of the world….. </para>

<section num= " 2 " > <title> What is Ontology</title>

<para> ...An ontology is an explicit specification of a conceptualization... </para>

<para> In the philosophical sense, we may refer to an ontology as a particular system of categories accounting for a certain vision of the world….. </para>

</section>

<article date-publi=‘2000’><title> Ontologies </title><author> J.Dupond</author><chapter>

<title> history of ontology </title><section num= "1" >

<title> Introduction</title><para> ...ontology should be seen only as an

interdiscipline... </para></section><section num= " 2 " > <title> What is Ontology</title>

<para> ...An ontology is an explicit specification of a conceptualization... </para>

<para> In the philosophical sense, we may refer to an ontology as a particular system of categories accounting for a certain vision of the world….. </para>

</section></chapter><chapter> …. </chapter></article>

Page 6: Towards Contextual and Structural Relevance Feedback in XML Retrieval Lobna Hlaoua IRIT (Institut de Recherche en Informatique de Toulouse) Equipe SIG-RI.

Relevance Feedback (RF)

• RF in traditional IR consists in enriching the initial query using terms extracted from relevant documents.

• How RF can be used in XML retrieval ?

Page 7: Towards Contextual and Structural Relevance Feedback in XML Retrieval Lobna Hlaoua IRIT (Institut de Recherche en Informatique de Toulouse) Equipe SIG-RI.

Relevance Feedback (RF) in XML

Retrieval • Problems

– How extracting terms from retrieved elements having different semantic

• element could (title, section, paragraph, etc.)

• … whereas in IR only document units are considered

– How structural constraints can be extracted from relevant elements

– How enriching XML queries : • adding structural constraints And/OR keywords in both

CO and CAS queries?

Page 8: Towards Contextual and Structural Relevance Feedback in XML Retrieval Lobna Hlaoua IRIT (Institut de Recherche en Informatique de Toulouse) Equipe SIG-RI.

Our approach

• Contextual RF– expand the query with expressive words according to

the context of the judged component from different granularities.

• Structural RF– select the more appropriate generative structure from

judged components and adding to CO query.

Page 9: Towards Contextual and Structural Relevance Feedback in XML Retrieval Lobna Hlaoua IRIT (Institut de Recherche en Informatique de Toulouse) Equipe SIG-RI.

Process of RF in XML Retrieval

Relevant components

Extraction of expressive terms

Extraction of relevant structure

Initial query

Relevance Feedback

Reformulated query

Results RC+ NRC

Page 10: Towards Contextual and Structural Relevance Feedback in XML Retrieval Lobna Hlaoua IRIT (Institut de Recherche en Informatique de Toulouse) Equipe SIG-RI.

• Objective: select the more expressive words

• Let’s consider Er={er1, er

2, ..., erk,... er

m} , erk

={l1,..., lj,.ln} – assign a score to terms (ti) occurring in each leaf node (lj) of the relevant elements.

– Compute the score of terms of in each element (erk).

– Select the best terms according to number of occurrence in relevant elements.

Contextual Relevance Feedback

)(

),(),(

j

jiji lcard

ltnfltscore

),(),( jel

irki ltscoreetscore

kr

j

),()( kEe

ii ttscoretscorerr

k

Page 11: Towards Contextual and Structural Relevance Feedback in XML Retrieval Lobna Hlaoua IRIT (Institut de Recherche en Informatique de Toulouse) Equipe SIG-RI.

Structural Relevance Feedback

• Objective is to select the more appropriate generative structure – retrieve the smallest common ancestor– attribute scores for each structure

dn

i iSSscore Si is the structure of relevant element having a joint base with the candidate structuren is a number of the relevant elementsd is the distance which separates nodes is a constant vaying in [0,1]

Page 12: Towards Contextual and Structural Relevance Feedback in XML Retrieval Lobna Hlaoua IRIT (Institut de Recherche en Informatique de Toulouse) Equipe SIG-RI.

Example of RF in CO query

• Initial query Q: “information retrieval”

• Structural RF – we suppose that relevant structures have the following

scores:• /book/chapter/section/subsection (0.4)

• /book/chapter/section/subsection/para (0.4)

• /book/chapter/section/title (0.2)

Q2: “article//sec[about(., information retrieval)] ”

Page 13: Towards Contextual and Structural Relevance Feedback in XML Retrieval Lobna Hlaoua IRIT (Institut de Recherche en Informatique de Toulouse) Equipe SIG-RI.

Conclusion & prospects

• Outlined the problem of Relevance Feedback in XML Retrieval.

• New challenge in IR : up until now only very few related works

• Our investigations can be considered as a first step of a long hard work.

• The main idea behind theses investigations:– how keywords and structural constraints can be selected and added

to CO queries.

• Experiments will be carried out “soon” in INEX framework

Page 14: Towards Contextual and Structural Relevance Feedback in XML Retrieval Lobna Hlaoua IRIT (Institut de Recherche en Informatique de Toulouse) Equipe SIG-RI.

THANK YOU

Page 15: Towards Contextual and Structural Relevance Feedback in XML Retrieval Lobna Hlaoua IRIT (Institut de Recherche en Informatique de Toulouse) Equipe SIG-RI.

Contextual Relevance Feedback

book

num=1

Introduction Internet knowledge ….

Search engine

chapter

Yahoo... Google…..

author

J. Dupont

title

section

title para

section

title para

para

chapter

title

Web access

Leaf node Node

Search engine

title

num=2

num=2

date-publi=2000

Page 16: Towards Contextual and Structural Relevance Feedback in XML Retrieval Lobna Hlaoua IRIT (Institut de Recherche en Informatique de Toulouse) Equipe SIG-RI.

Contextual Relevance Feedback

book

Introduction Internet knowledge ….

Search engine

chapter

Yahoo... Google…..

author

J. Dupont

title

section

title para

section

title

para

para

chapter

title

Web access

Search engine

title

date-publi=2000

?

?

?

?

Subsection

Subsection

?

Page 17: Towards Contextual and Structural Relevance Feedback in XML Retrieval Lobna Hlaoua IRIT (Institut de Recherche en Informatique de Toulouse) Equipe SIG-RI.

Contextual Relevance Feedback

book

Introduction Internet knowledge ….

Search engine

chapter

Yahoo... Google…..

author

J. Dupont

title

section

title para

section

title

para

para

chapter

title

Web access

Search engine

title

date-publi=2000

S2=0.4S3=0.2

subsection

subsection

S2=0.4

0.73

0.46

0.58

section

Page 18: Towards Contextual and Structural Relevance Feedback in XML Retrieval Lobna Hlaoua IRIT (Institut de Recherche en Informatique de Toulouse) Equipe SIG-RI.

Structural Relevance Feedback

book

chapterauthortitle

section

title para

section

title para para

chapter

titletitle ...

anc[9]

des[4]

sca[11,13]

1

1110

7 8

4

6

32

9

5

141312