University Louis Pasteur Strasbourg I

Louis Pasteur

Description

Niveau: Supérieur, Doctorat, Bac+8

• dissertation

University Louis Pasteur Strasbourg I University of West Bohemia in Pilsen Doctoral Dissertation under Joint Supervision 2007 Dalibor FIALA

• faculté des sciences appliquées


• external reviewer


• roland de guio


• insa de strasbourg


• lubomír popelínsk?

Sujets

• Dissertation
• Louis Pasteur University
• Collet
• Pasteur
• Lubomir
• Rousselot

Informations

• Publié par: profil-zyak-2012
• Nombre de lectures: 53
• Poids de l'ouvrage: 2 Mo

Extrait

University Louis Pasteur Strasbourg I

University of West Bohemia in Pilsen














Doctoral Dissertation
under Joint Supervision




















2007 Dalibor FIALA University Louis Pasteur Strasbourg I
LGeCo, INSA Strasbourg

University of West Bohemia in Pilsen
Faculty of Applied Sciences



WEB MINING METHODS FOR THE
DETECTION OF AUTHORITATIVE
SOURCES

by
Dalibor FIALA


A dissertation under joint supervision submitted in partial
fulfillment of the requirements for the degree of Doctor of
Philosophy in “Computer Science” and “Computer Science
and Engineering”


Presented and defended publicly on November 30, 2007 before the board of examiners.




Pierre COLLET internal reviewer University Louis Pasteur
François JACQUENET external reviewer University of Saint-Etienne
Lubomír POPELÍNSKÝ external reviewer Masaryk University Brno
Bernard KEITH examiner INSA Strasbourg
Roland DE GUIO examiner INSA Strasbourg
Václav MATOUŠEK examiner University of West Bohemia
François ROUSSELOT supervisor INSA Strasbourg
Karel JEŽEK supervisor University of West Bohemia





Strasbourg / Pilsen 2007 Université Louis Pasteur Strasbourg I
LGeCo, INSA Strasbourg

Université de la Bohême de l’Ouest à Plzeň
Faculté des Sciences Appliquées



LES MÉTHODES DE LA FOUILLE DU WEB
POUR LA DÉTECTION DES SOURCES
FAISANT AUTORITÉ

par
Dalibor FIALA


Thèse en cotutelle présentée pour l’obtention du grade de
Docteur de l’Université Louis Pasteur Strasbourg
(spécialité Informatique) et de l’Université de la Bohême de
l’Ouest (spécialité Informatique et ingénierie)


Soutenue publiquement le 30 novembre 2007 devant la commission d’examen.




Pierre COLLET rapporteur interne Université Louis Pasteur
François JACQUENET rapporteur externe Université Saint-Etienne
Lubomír POPELÍNSKÝ rapporteur externe Université Masaryk Brno
Bernard KEITH examinateur INSA Strasbourg
Roland DE GUIO examinateur INSA Strasbourg
Václav MATOUŠEK examinateur Université de Plzeň
François ROUSSELOT directeur de thèse INSA Strasbourg
Karel JEŽEK directeur de thèse Université de Plzeň





Strasbourg / Plzeň 2007 Université Louis Pasteur Strasbourg I
LGeCo, INSA Strasbourg

Západočeská univerzita v Plzni
Fakulta aplikovaných věd



METODY WEB MININGU PRO
VYHLEDÁVÁNÍ AUTORITATIVNÍCH
ZDROJŮ


Ing. Dalibor FIALA


Disertační práce pod dvojím vedením
k získání akademického titulu doktor
v oboru “Informatika” a “Informatika a výpočetní technika”


Předneseno a obhájeno veřejně před zkušební komisí dne 30. listopadu 2007.





Prof. Pierre COLLET Univ Louis Pasteur
Prof. François JACQUENET Univ Saint-Etienne
Doc. RNDr. Lubomír POPELÍNSKÝ, Ph.D. Masarykova univerzita
Prof. Bernard KEITH INSA Strasbourg
Prof. Roland DE GUIO INSA Strasbourg
Prof. Ing. Václav MATOUŠEK, CSc. KIV ZČU v Plzni
Dr. François ROUSSELOT INSA Strasbourg
Doc. Ing. Karel JEŽEK, CSc. školitel KIV ZČU v Plzni





Strasbourg / Plzeň 2007













To Anna for her love and patience
and to all of my family for their support and encouragement















































The work on this doctoral thesis was supported in part by the Ministry of Education of the
Czech Republic under Grant 2C06009 COT-SEWing. Declaration

I submit this dissertation for review and defence in partial fulfillment of the requirements for
the degree of Doctor of Philosophy at the University of West Bohemia in Pilsen, Czech
Republic and at the University Louis Pasteur Strasbourg, France.

I hereby declare that this doctoral thesis is completely my own work and that I used only the
cited sources.


______________________
Pilsen, August 30, 2007 Dalibor Fiala
Abstract

The development of information society in recent decades has enabled collecting, filtering and
storing huge amounts of data. These data must be further processed to gain valuable
information and knowledge. The scientific field dealing with extracting information and
knowledge from data has evolved rapidly to cope with the extent and growth of information
sources the number of which has geometrically increased with the appearance of the World
Wide Web. All traditional approaches in information retrieval, knowledge acquisition, and
data mining must be adapted for the dynamic, heterogeneous, and unstructured data on the
Web. Web mining has come into being as a fully-fledged research discipline.

The Web brings much specificity with it. The most salient feature is its link structure. The
Web is a dynamic, linked network of nodes. Web pages contain links to other pages with
similar contents, of a specific or more general interest, or otherwise related. Soon it was
discovered that the link structure of Web is a vast resource of information and that it presents
a wonderful field for applications from the social network domain as well as from the
mathematical graph theory. Brin and Page have submitted the interlinkage of Web pages to an
extensive research which resulted in the appearance of the now famous article “The anatomy
of a large-scale hypertextual Web search engine” in 1998 introducing Google – a search
engine for day-to-day usage by the whole Web community. The success of Google has been
very much due to the underlying algorithm called PageRank, which makes use of the
interconnection of billions of Web pages recursively so as to identify popular, prestigious,
significant, or authoritative sources on the Web. The description of PageRank has been
published and this results in a steady flow of new research papers on link-based methods that
finally introduce a completely new group of algorithms – ranking algorithms. Each technique
has its particular properties and is aimed at coping with specific problems. Although
originally conceived for the Web, ranking algorithms are usable in every environment that can
be modelled as a graph.

The innovative portion of this doctoral thesis deals with the definitions, explanations and
testing of modifications of the standard PageRank formula adapted for bibliographic
networks. The new versions of PageRank take into account not only the citation but also the
co-authorship graph. We verify the viability of the new algorithms by applying them to the
data from the DBLP digital library and by comparing the resulting ranks of the winners of the
ACM SIGMOD E. F. Codd Innovations Award. The rankings based on both the citation and
co-authorship information turn out to be better than the standard PageRank ranking. In
another part of the disseration, we present a methodology and two case studies for finding
authoritative researchers by analyzing academic Web sites. In the first case study, we
concentrate on a set of Czech computer science departments’ Web sites. We analyze the
relations between them via hyperlinks and find the most important ones using several
common ranking algorithms. We then examine the contents of the research papers present on
these sites and determine the most authoritative Czech authors. In the second case study, we
do exactly the same with French academic computer science Web sites to find the most
significant French researchers in the field. We also discuss the weak points of our approach
and propose some future improvements. To the best of our knowledge, it is the only attempt
ever made at discovering authoritative researchers from the above countries by directly
mining from Web data.

Keywords: Web mining, Web crawling, ranking algorithms, bibliographic networks,
citations, co-authorships, authorities, bibliographic PageRank. Résumé

Le récent développement de la société de l’information a permis de collecter, de filtrer et de
stocker de grandes masses de données. Le problème est maintenant d’exploiter ces données
pour obtenir des informations et des connaissances pertinentes. Les techniques d’extraction
des informations et des connaissances à partir de données ont rapidement évol