A SOCIAL APPROACH FOR DOCUMENT ANALYSIS
Ibrahim Bounhas and Yahya Slimani
Department of Computer Science, Faculty of Sciences of Tunis, University of Tunis, 1060, Tunis, Tunisia
Keywords: Authority Study, Document Usage Type, Document Model, Logical Structure.
Abstract: This paper discusses the social constraints that should be taken into account in document analysis. In fact, a
document is viewed and analysed as the object of a transaction between realizers and beneficiaries. Then,
the first condition for the success of the document reading process is insuring confidence between the two
parties. Thus document analysis should help in authority evaluation which is based on identifying the names
and the roles of the realizers and also studying their behaviours. Our proposal approach is social-based.
Besides authority study, it is based on identifying document usage types. A usage type defines how a
community of users can access to a document.
Designing a document parser requires to solve complex problems. We argue that authority evaluation
requirements and document usage types are important parameters that can reduce this complexity. Thus, the
first step of our approach consists in identifying social parameters. Then documents are fragmented
regarding their logical structure. We distinguish two levels: the macro-logical level and the micro-logical
level. These two levels allow us to design a generic approach that could be applied to documents having
different styles and different types.
1 INTRODUCTION
The semantic Web project brought solutions to the
limits of the actual Web at the semantic level. In the
last few years, Zacklad (2007) introduced the socio-
semantic Web with the aim to study the social
interactions and how they lead to the creation of
explicit and semantically rich knowledge
representations. From the system point of view,
these interactions can affect many steps in document
analysis. As mentioned by Zacklad (2007), the
document should be analyzed as a transaction
between actors (realizers and beneficiaries) involved
in an exchange process.
In this paper, we focus on the problem of
analyzing semi-structured documents like scientific
papers and books. We believe that reflections
presented by Zacklad (Zacklad, 2007) should change
our design of approaches and tools of document
analysis. Many researchers have and continue to
work on analyzing complex documents, like
scientific papers and books. Some researchers
identify only high level fragments (chapters or
sections titles). Others perform detailed analysis but
consider only some parts of the document (Connan
and Omlin, 2000). Approaches that allow the
analysis of the whole document are limited to a
document type or style. For example, Rangoni and
Belaïd (2006) analyze papers belonging to the same
conference.
Despite these works, we think that many
questions are still without responses. How to
determine elements that should be identified by a
document parser? How to define the best granularity
level to split a document? Which activity or practice
some fragment is useful to?
We believe that a social study can contribute to
answer to these questions. Thus, we consider two
social dimensions in the document analysis process.
The first dimension concerns the “authority notion”
considered as a social issue. In fact, the credibility of
actors having produced or transmitted a document is
one of the relevance criteria. The second dimension
is related to the need of personalization based on the
user practice.
The remainder of this paper is as follows.
Section 2 overviews the problem of document
analysis under the social constraints. Our proposed
approach is presented in Section 3. Sections 4 and 5
give two applications that illustrate our approach.
Finally, Section 6 concludes our paper and suggests
some directions for future research.
95
Bounhas I. and Slimani Y. (2009).
A SOCIAL APPROACH FOR DOCUMENT ANALYSIS.
In Proceedings of the International Conference on Knowledge Management and Information Sharing, pages 95-102
Copyright
c
 SciTePress

2 PROBLEM DISCUSSION
The great number of information suppliers on the
Internet and the enormous quantity of the available
information caused anxieties towards the
information reliability. Information validation,
which was done by authors, editors and libraries, is
henceforth done by the user. The later is in many
situations unable to identify the source of
information or to judge its credibility. This difficulty
is amplified when many actors having different
privileges or prerogatives participate in producing
and/or transmitting the information (Zacklad, 2007).
Another change related to the generalisation of the
Web is the diversification of the users’ needs. From
the social point of view, the main phenomenon is the
document reading manner or what is called
"document usage" (Aussenac-Gilles and
Condamines, 2004).
Aussenac-Gilles and Condamines (2004) argue
that we should model both the text and the usages
and that the document usages are not as many as
users. Thus, the document usage notion allows
viewing the document in a collective perspective
which means that a community of users share the
same manner of reading a document. Zacklad (2007)
confirms this fact by arguing that the actual Web
caused the multiplication of document centred
collective practices. That's why we consider
document usage as a social phenomenon. In fact,
user's needs and actions on information (or
documents) are related to his/her membership in a
community of practice. According to Wenger
(1998), a community of practice is “a group of
professionals, informally bound to one another
through exposure to a common class of problems,
common pursuit of solutions, and thereby themselves
embodying a store of knowledge”. According to his
membership in a community, a user is interested in
some fragments of a document but not in others.
In addition, document analysis should reply to
the needs of mediators (such as terminologists)
whose task is to analyze its content to produce
terminological or ontological resources used to
support the research information process and
facilitate the end user work.
Thus, we can distinguish three types of actors
involved in the documentarisation process:
 Actors participating in document
production or transmission,
 Beneficiaries regrouped into communities,
 Mediators: many practices related to these
actors such as summarisation, key-word
extraction and name entity study.
Our goal is to suggest solutions to the problems
of authority evaluation (cf. section 2.1) and the
impact of actors practices in document segmentation
(cf. section 2.2).
2.1 Authority Evaluation
Authority can be defined as the set of indicators that
prove (or can be used to study) the faithfulness of
information stakeholders. To study the authority of a
web site, we should verify the existence of
information such as author names, contact
information, copyright texts, and so on.
As mentioned by Zacklad, the identification of
such indicators is necessary to document
understanding, interpretation and exploitation
(Zacklad, 2007).
The authority notion is related to the “reliability
notion” defined as “the degree to which the user can
trust the information”. Within the web context,
many researchers tried to provide methods, metrics
and tools for reliability assessment. In fact, it is
recognized that authority is the most important
dimension of information reliability. Besides on, it is
an important relevance criterion (Naumann &
Rolker, 2000), (Knight & Burn, 2005) and (Rieh,
2002).
Document authority study is related not only to
the identification of intrinsic elements but also to the
analysis of extrinsic informations that can help in
document evaluation (e.g. biographies of authors).
Thus we should identify indicators and then metrics
allowing authority evaluation.
2.2 Social Segmentation
For Zacklad, a document is a semiotic production
that should have attributes that facilitate practices
related to its exploitation. These attributes permit
document circulation through communities of
interpretation. Zacklad (2007) distinguishes two
types of practices and the corresponding attributes
types:
 Practices related to the external exploitation of
the document when it is stored (and studied)
among a collection of documents: For these
practices, links between different documents
should be identified.
 Practices related to the internal exploitation of
the document: to consider these practices, a
document should be segmented into coherent
parts. Links between fragments should be
identified to allow semantic navigation and
orientation through the document.
KMIS 2009 - International Conference on Knowledge Management and Information Sharing
96

In other hand, we consider that the segmentation
of a document depends on the level of granularity
chosen. We believe that the best level can be defined
if we conduct a social study for identifying users’
practices. Thus, we argue that the same collection of
documents can be modelled and fragmented
differently according to the intended usages or the
social organization of the users
From the point of view of the socio-semantic
Web, a document is a result of a macro
communicational transaction composed of many
micro-transactions involving many actors.
Transactions can be decomposed until a fine level
corresponding to the elementary language acts.
According to Zacklad (2007), documents are
more fragmented when many actors participate in
their production. Besides, micro-transactions are
articulated together by elements related to the logical
structure such as the document titles and sub-titles,
titles numbers and other attributes specifying
fragment status or links with other fragments.
We can conclude that the logical structure of a
document corresponds to the social process that
leaded to its production. That's why the approach
that we propose is based on logical structure
extraction.
3 THE PROPOSED APPROACH
As conclusion of the above sections, we can claim
that reliability insurance is mainly based on
authority study which consists in identifying actors
who have produced or transmitted the document and
their roles. In addition, document segmentation
should be based on a social study consisting in
identifying users’ practices or document usages.
This segmentation should be based on the logical
structure of the document because this logical
structure reflects the communicational transactions
that leaded to the production of the document.
Finally, document structure study should allow both
the internal and external exploitation of the
document. Thus, we propose an approach composed
of the following steps: social study, document
modelling, logical structure extraction and social
indexing.
3.1 Social Study
According to the document type and/or the domain,
we should, first, study the social process of
document production and transmission to identify
actors’ roles typology in document production and
transmission. This study incorporates the
identification of any other authority indicators. The
second goal of this study is to identify external
users’ typology and the practices corresponding to
each user type (or community). By external users,
we mean end users and mediators.
3.2 Document Modelling
The goal of this step is to identify the different
document fragment types and the different types of
internal and external links. First, we identify
fragments types corresponding to internal actors’
names and roles. Second, we identify, for each
community of end users, fragment and links types
needed for their practice. The result of this step is an
Xml DTD representing the document model.
3.3 Logical Structure Extraction
The goal of this step is to structure documents
according to the document model constructed in the
previous step. We fulfil this goal in two steps:
macro-logical structure extraction and micro-logical
structure extraction. The goal of the first step is to
identify high level logical fragments such as
chapters or sections. In the second step, we identify
the logical entities of each macro-logical fragment
by labelling its sub-blocks. The idea behind
separating the two levels is to develop a generic and
reusable macro-logical analyzer and many micro-
logical analyzers each specialized in a analyzing a
macro-fragment type.
We should notice that our approach is based on
the labelling method widely used in logical structure
analysis approaches. At several steps of our analysis
process, we define a list of labels that are attributed
to blocks using rules. Thus, we inspired from the
rule based approach widely used in this filed (Mao,
Rosenfeld and Kanungo, 2003).
We use a pre-treatment step where the different
types of physical fragments are identified according
to the document type and format. We identify three
steps: physical analysis, macro-logical structure
extraction and micro-logical structure extraction.
3.3.1 Physical Analysis
In this step, the analysed document is restructured
according to a physical model shown by figure 1. In
fact, we consider that a document is defined by its
blocks vector (BV) and a styles vector (SV).
Our model regroups all the blocks types that can
be found in a document. Each block type is
A SOCIAL APPROACH FOR DOCUMENT ANALYSIS
97

characterized by several attributes. For example, we
compute the number of words of text blocks.
Physical analysis consists, also, in identifying the
different styles used in the document. Styles found
in the document are added to the SV vector and
linked to the corresponding blocks.
D = (BV , SV)
BV = [b1, b2,...,bn]bi  { Text, link, List, Table, Img}
SV = [s1, s2,...,sm]
sj= (Size, Blod, Italic, Uderlined, Color, Align)
Figure 1: The proposed physical model.
3.3.2 Macro-logical Structure Extraction
To recognize the macro-logical structure of a
document, we start by identifying macro-logical
fragments’ titles. To do so, we combine three types
of analysis: content analysis, context analysis and
style analysis. In fact, we define constraints about
the fragment’s attributes, the attributes of its
predecessor and its successor.
We compute the level of each style by combining
three criteria: the style attributes, the style regularity
and the style frequency.
When macro-logical titles are identified, we
regroup document blocks in a hierarchical manner.
We consider that a document is a tree where the top-
level elements are chapters or sections according to
its type. The leaves of this tree are the physical
blocks identified in the Physical analysis step.
This step is performed by using regrouping rules.
For example, the following rule regroups a caption
with an image into a block labelled "Figure" if the
caption is followed by a vertical separator:
Img + Caption_Figure (AfterSep > 0)
==> Figure
3.3.3 Micro-logical Structure Extraction
The goal of this step is to identify elementary logical
entities required for the different social activities of
the intended users. Thus conceiving rules for
labelling and regrouping blocks should be based on
the social study performed in the first step. For
example, according to users’ practices, we can
consider that a bibliographic reference is an
elementary block that should not be segmented. In
other cases, we can opt for analyzing each reference
to identify its sub-elements. Besides, micro-logical
analysis should allow identifying the names and the
roles of actors.
Typically a micro-logical analyzer is a semi-
structured text parser. In fact, our tool implements
an approach based on context free grammars (CFG)
which was recently recommended by many
researchers. Among the existent solutions, using
CFG is the most flexible approach. Besides on,
systems based on CFG are able to recognize the text
structure especially when it contains complex
relationships and constraints between labels (Viola
and Narasimhand, 2005).
The grammar used by a micro-logical analyzer is
constructed by semi-automatic learning. We made
this choice because training data is not available and
because we want to help the expert constructing the
grammar. Thus, we propose a semi-automatic
approach where an expert labels and groups
manually the text blocks. The set of labels and rules
are inferred from the expert actions to construct the
grammar. The expert intervention allows taking into
account the social constraints already fixed.
As output is concerned, each analyzer generates
an Xml flow that represents the logical structure of
the macro-logical fragment. Xml flows generated by
micro-logical analyzers are regrouped according to
the document model already conceived.
3.4 Social Indexing and Authority
Evaluation
By social indexing we do not mean the use of
folksonomies. In our sense, social indexing means
relating each document or fragment to actors who
produced, transmitted or published them. Our goal is
to develop tools able to identify names of these
actors and their respective roles in existent
documents. The document analysis process should
also allow identifying any indicator that can used to
study the document authority (contact information,
copyright texts…etc).
We should remember that authority study is not
limited to identifying names and roles of these actors
in documents but incorporates the precise
identification and the study of the biography of each
actor. The goal of this study is to allow authority
evaluation by supplying all possible informations
about actors. The nature and the size of these
informations depend on the domain and/or the
intended applications.
The last step consists in analyzing authority
indicators. Fundamentally, we want to provide a
clear idea about document authority. In better cases
can we define metrics for authority evaluation. In the
worst cases, should we identify and analyze intrinsic
and extrinsic authority indicators and highlight
sources of unreliability.
In the following two sections, we present and
discuss two examples that illustrate our approach.
KMIS 2009 - International Conference on Knowledge Management and Information Sharing
98

4 FIRST APPLICATION:
SCIENTIFIC PAPERS
The goal of this project is to create a virtual library
of scientific papers. The purpose is to provide
adaptable services to the intended users of the
library.
4.1 Social Study
Scientific paper production is a typical social
phenomenon. Firstly, the scientific progress is
realized by accumulating the efforts of researchers
around the world each contributing by his effort
published in a paper. Secondly, scientific paper
production involves many actors having different
profiles, responsibilities and roles (the writer, the
research head, the team or unit chair…etc). Then,
the publication of the paper requires the participation
of other types of actors (conference organizers,
reviewers, editors, and so on.). When it is published
a scientific paper is accessible to a more large
community of users what extends its usage.
By analysing the activities of the different actors,
we can distinguish two different practices. The first
practice is related to papers retrieval, discovery and
relevance evaluation. The first criterion for
relevance evaluation is the paper’s theme. Then, the
user would evaluate the paper according to the idea
or in general the knowledge it contains. A common
reading scenario is based on the paper’s logical
structure. Thus the user starts by checking the
paper’s title, then the abstract, after that the
introduction and the conclusion and finally the
whole content. In other cases, relevance judgment is
based on key-words or key-concepts.
To allow documents retrieval and discovery, the
virtual library defines two types of research. In the
first one, documents are clustered by theme that
allows the users to access to the library according to
their interests. For the second type of research,
documents are indexed by the concepts of a domain
ontology which allows finest research. To allow
personalised navigation, documents are fragmented
and each fragment is related to the concepts of the
domain ontology.
The second practice concerns scientific impact
evaluation and technological watch. The main
element used in scientometry is bibliographic
references. The impact of a paper depends on the
number of other papers it is cited in. Technological
watch means discovering new progresses in a
scientific field. The main elements that can be used
are the document bibliographic informations, the
conclusion and bibliographic references (Dou and
al., 1990). Bibliographic informations contains
paper’s title, authors’ names and affiliations what
allows discovering new fields and researchers
involved in. The conclusion contains, generally, the
perspectives that allow identifying new fields of
research.
4.2 Document Modelling
From the social study, we can identify the elements
that should be involved in a scientific paper model.
First, the macro-logical analysis should allow
identifying the paper’s sections because each section
is useful for a different practice. Sections of the
paper’s body should be identified and organized in
hierarchical manner what allows generating the
paper’s table of contents and thus facilitates
navigation in the paper. Because we need to perform
conceptual indexing, we should identify more fine
logical entities in the body such as paragraphs and
captions.
Second, bibliographic informations, references
and notes should be analyzed to allow linking
different papers and permit bibliometry
measurement and technological watch. This analysis
allows also identifying actors having participated in
the production of the paper (or that are
acknowledged) and their respective roles.
The main elements of the obtained document
model are shown by Figure 2. It is composed of
three parts. The first part called the front is
composed of the bibliographic information of the
paper, the abstract and the key-words. Thus it
summarizes the main elements of the paper. We
should note that the “Author” element regroups
author names, roles and affiliations. In the case of
papers containing a “biography” section, this
element regroups also the content of this section.
The second element - the body - constitutes a
detailed view of the paper by regrouping the
introduction, the conclusion and the other sections.
The last element –the back- regroups elements used
for scientometry and technological watch such as the
bibliographic references and the notes. Each
reference or note is decomposed to the finest
granularity level to allow authority study and papers
linking.
A SOCIAL APPROACH FOR DOCUMENT ANALYSIS
99

<!ELEMENT ScientificPaper (Front, Main, Back?)>
<!ELEMENT Front (title1, title2?,Author+,Abstract*,
Keywords*, Edition?, CopyrightText?)>
<!ELEMENT Main(Introduction?, Section+,
Conclusion?)>
<!ELEMENT Section(title?, (Section | Paragraph | List |
Figure | Table)+)>
<!ATTLIST Section level CDATA >
<!ELEMENT Back (Bibliography?,Notes?,
acknowledgements?)>
<!ELEMENT Bibliography (reference+) >
<!ELEMENT Notes (Note)+>
…
Figure 2: Scientific paper model.
4.3 Logical Structure Extraction
Our goal in this step is to prepare the document to be
injected into a virtual library where document
fragments, themes, concepts and actors are linked
together. We performed macro-logical and micro-
logical structure extraction according to the above
presented approach.
To evaluate our approach, we selected 25
scientific papers in the HTML format from different
conferences (and thus having different styles). The
prototype was able to analyze a scientific paper in
22.5 seconds as average. At the macro-logical level,
89.09% of the sections were correctly recognized. At
the micro-logical level the prototype realized a
global success rate equal to 92.85%. These success
rates are encouraging if we consider that we used
papers having different styles. In fact we recorded
better results than those of Rangoni and Belaïd
(2006) who recorded a success rate equal to 91.7%
for documents sharing the same style.
4.4 Social Indexing and Authority
Evaluation
For authority evaluation we consider two criteria.
Our prototype permits recognizing bibliographic
informations at the finest level of granularity. These
informations are linked, for some papers, to
biographic informations found in the “biography”
section. We are planning to construct a database of
conferences and journals. At this stage, we plan to
integrate information available on the Web and
allowing conferences and journals evaluating and
ranking. Finally, we plan to develop a tool allowing
papers linking based on bibliographic references
what will help in evaluating the impact of each
paper.
5 SECOND APPLICATION:
BOOKS OF ARABIC STORIES
The purpose of this project is to help in Arabic story
study. An Arabic story reports historical events or
speeches assigned to a person. In the Arabic history,
stories were reported among generations by persons
called reporters or narrators. Because these stories
report important historical events, Arabic scholars
instituted strict rules for story reporting and
transmission. Each narrator has an obligation to cite
the list of narrators from which he got the story.
Thus, the story is preceded by a chain of narrators.
Besides on, when a narrator (called the sheikh)
communicates a story to his follower (the disciple),
he uses verbs that indicate how he got the story from
his predecessor (his sheikh) that is called manner of
transmission.
Chain of narrators study is a fundamental step
that should be performed before studying the story’s
content. Compared to other types of texts such as
scientific papers, authority study gets a greatest
importance for Arabic stories. Thus, a story without
a chain of narrators (or with an incomplete chain) is
rejected whatever is its content.
Stories are actually regrouped in historical books
organized by theme. These books were written by
specialized scholars who tried to identify reliable
stories. Thus, in some cases, the story is followed by
comments about the credibility of its narrators
and/or a global reliability judgment. The story can
also be followed by indications about other versions
of the same story. These indications include
references about other books where the same story
exists. In other cases, these indicators highlight the
difference between different versions.
5.1 Social Study
Arabic storytelling is a socio-historical phenomenon
characterized by the intervention of many actors
having different roles. We can classify these actors
as follows:
 Actors participating in the story,
 Actors having reported the story,
 Scholars having collected stories,
 Scholars having judged narrators or stories.
We can identify two types of practices related to
arabic study. In one hand, arabic scholars defined
strict rules for story acceptance. To be accepted a
story should be reported by credible narrators.
Narrators are evaluated by specialized scholars
based on their behaviours. Twelve classes of
KMIS 2009 - International Conference on Knowledge Management and Information Sharing
100

credibility are defined. In addition, a story should
have a continuous chain of narrators what means
that any geographic or temporal gap between two
successive narrators is considered as a source of
suspect. Thus relations between actors should be
studied. At this stage, many types of relations are
considered. Besides the kinship and the sheikh-
disciple relations, places of residence and dates of
birth and death are studied.
On the other hand, an arabic story constitutes a
rich source of knowledge. In fact, books of arabic
stories support activities of research in arabic history
and civilisation.
5.2 Document Modelling
Book of story modelling should allow identifying
elements used to story study. As far as authority is
concerned two elements should be modelled. In one
hand chains of narrators should be identified and
parsed to identify names of narrators and manners of
transmission. On the other hand, comments about
reliability that often follow the story should be
recognized and analyzed.
To allow studying stories at the semantic level,
the book themes should be recognized. Besides on,
story content should be separated from chains and
reliability comments to allow conceptual indexing.
As shown by Figure 3, we consider that a
historical book is composed of themes and stories.
Each story is defined by a chain of narrators, a
content and eventually reliability comments and/or
versions indicators. As actors are concerned, we
developed (by semi-automatic learning) a grammar
for full arabic names. This grammar is based on
arabic name components described in (Shaalan &
Raza, 2007). Besides on, it takes into account social
relations between actors (kinship and the sheikh-
disciple relation).
<!ELEMENT HistoryBook (Theme+) >
<!ELEMENT Theme (Theme | Stroy)+>
<!ELEMENT Story (Chain | Content | Comment |
Version)+>
<!ELEMENT Chain(Actor, Manner)+)>
<!ELEMENT Comment (ActorComment|
StroyComment)>
…
Figure 3: Scientific paper model for historical book.
5.3 Logical Structure Extraction
To test our approach we composed a base of 1600
stories selected from four different books. We
started by identifying the themes of each book to
recognize its macro-logical structure. At this step,
we did not record any error. Then, we developed a
grammar for parsing an Arabic story. The grammar
takes into account actor’s names, chains, versions
indicators and comments. We evaluated the success
rate of our micro-logical parser for four elements as
shown by Table 1.
Table 1: Experimentation results for the story parser.
Success rate
Actors
97.24%
Chains
95.66%
Versions indicators
93.01%
Reliability comments
85.51%
Our results in recognizing arabic person names
are better than the results reported by other
researchers. For example, the success rate of Zitouni
and al. (2005) is equal to 70.2%, Abuleil (2004)
reported 83% and Shaalan and Raza (2007) achieved
89%. Being the first to invoke the problem of Arabic
story parsing, we could not make a comparison for
the remaining elements.
5.4 Social Indexing and Authority
Evaluation
Having realized encouraging success rates in story
parsing, we are actually working in integrating an
existent database of narrators. This database contains
exhaustive informations about the narrators of the
main arabic historical books. It includes classes of
credibility attributed by different scholars to each
narrator, information about places and dates of birth
and death and information about social relations.
Our task is to link names of narrators found in
stories to the names of the database. It is a difficult
task because an arabic person can have many names
and many persons can share the same name(s).
Besides, thousands of persons had reported stories
which means that narrators’ names are very
ambiguous. When narrators are exactly identified,
we can use the information available in the database
to judge the story’s reliability. It will be also
possible to link different stories based on the
narrators they have in common. Then we can
perform version comparison. For example when the
same story is reported by two narrators, we can
compare the two versions and discover any existent
anomaly.
A SOCIAL APPROACH FOR DOCUMENT ANALYSIS
101

6 CONCLUSIONS
In this paper, we discussed the problem of document
analysis from a social point of view. We followed
the guidelines of Zacklad and other researchers
which argue that authority study is one of the main
relevance criteria (Zacklad, 2007) and (Rieh, 2002).
We were also based on the document usage type
notion introduced by Aussenac-Gilles and
Condamines (2004). Using these notions, we
proposed an approach to extract the logical structure
of documents. The first step in this approach is the
social study that allows identifying the actors
involved in the documentarisation process,
document usages and finally fragments and links
types to be identified for each document usage. This
will lead, in a second step, to a document model that
regroups all the required elements. The third step
consists in structuring the documents according to
this document model. In this step, we distinguished
two levels of logical structure: the macro-logical
level and the micro-logical level. This distinction
allowed us developing reusable and interoperable
components that can be used in different document
types and styles.
To illustrate our approach, we presented two
projects we are actually working on. We used
documents in different domains, different types
(papers and books) and different languages (Arabic
and French).
As future research, we plan to study knowledge
extraction and mapping. It includes conceptual
indexing and knowledge visualisation. Our goal is to
structure a document collection as a map linking
themes, fragments, concepts and actors. We think of
an intelligent map that guides the user navigation
according to his practice represented in his profile.
This navigation is also guided by the authority
evaluation process.
REFERENCES
Zacklad, M., 2007. Processus de documentarisation dans
les Documents pour l’Action (DopA). Babel - edit -,
Le numérique: impact sur le cycle de vie du document,
ENSSIB.
Naumann, F. & Rolker, C., 2000. Assessment Methods for
Information Quality Criteria, In. International
Conference on Information Quality (IQ),
Cambridge, MA..
Knight, S. & Burn, J., 2005. Developing a Framework for
Assessing Information Quality on the World Wide
Web, Informing Science Journal, vol. 8, pp. 59-73.
Rieh, S. Y., 2002. Judgment of Information Quality and
Cognitive Authority in the Web, Journal of the
American Society for Information Science and
Technology, vol. 53, no. 2, pp. 145-161.
Shaalan, K. & Raza, H., 2007. Person Name Entity
Recognition for Arabic. In ACL’07. Workshop on
Computational Approaches to Semitic Languages,
Prague, Czech Republic, pp.17-24.
Viola, P. & Narasimhand, M., 2005. Learning to Extract
Information from Semi-structured Text using a
Discriminative Context Free Grammar. In 28th annual
international ACM SIGIR conference on Research and
development in information retrieval, Salvador, Bahia,
Brazil, pp. 330-337.
Rangoni, Y. & Belaïd, A., 2006. Document Logical
Structure Analysis Based on Perceptive Cycles. 7th
IAPR Workshop on Document Analysis Systems - DAS
2006, Springer Verlag (Ed.), pp. 117-128.
Dou, H. & Hassanaly, P., Quoniam, L.; La Tela A., 1990.
Technological watch and information: on bibliometric
analysis in information services, Documentaliste, vol.
27, no. 3, pp. 132-141
Connan, J. & Omlin., C. W., 2000. .Bibliography
Extraction with Hidden Markov Models, technical
report US-CS-TR-00-6, computer science department,
University of Stellenbosch.
Aussenac-Gilles, N. & Condamines, A., 2004. Documents
électroniques et constitution de ressources
terminologiques ou ontologiques, Information-
Interaction-Intelligence, vol. 4, no. 1 pp. 75-94,
Wenger, E., 1998. Communities of Practice: Learning,
Meaning and Identity, Cambridge University Press.
Zitouni, I., Sorensen, J., Luo, X. & Florian R., 2005. The
Impact of Morphological Stemming on Arabic
Mention Detection and Coreference Resolution, In
ACL’05, workshop on Computational Approaches to
Semitic Languages, 43rd Annual Meeting of the
Association of Computational Linguistics. Ann Arbor,
Michigan, USA, pp. 63-70.
Abuleil, S., 2004. Extracting Names from Arabic Text for
Question-Answering Systems, In RIAO’2004,
Coupling approaches, coupling media and coupling
languages for information retrieval, Avignon, France.
pp. 638- 647.
KMIS 2009 - International Conference on Knowledge Management and Information Sharing
102