UNIFICATION OF XML DOCUMENT STRUCTURES FOR
DOCUMENT WAREHOUSE (DocW)
Ines Ben Messaoud, Jamel Feki, Kais Khrouf
Laboratory Mir@cl, University of Sfax, Airport Road Km 4, Sfax, Tunisia
Gilles Zurfluh
IRIT, Institut de Recherche en Informatique de Toulouse, Université Toulouse 1, Toulouse, France
Keywords: Document warehouse, DTD unification, XML document, Unified tree.
Abstract: Data warehouses and OLAP (On Line Analytical Processing) technologies analyse huge amounts of
structured data that companies store as conventional databases. Recent works underline the importance of
textual data for the decision making process and, therefore, lead to build document warehouses. In fact,
documents help decision makers to better understand the evolution of their business activities. In general,
these documents exist in XML format, are geographically distributed and described by multiple and
different structures. This paper deals with a method to build a distributed document warehouse. This method
consists of two steps: i) unification of XML document structures in order to set a global and generic
perception/view of the distributed document warehouse, and ii) multidimensional modeling of unified
documents for decisional purposes. More specifically, this paper focuses on the unification step.
1 INTRODUCTION
Data warehouses (DW) are widely used in many
organizations and their benefits are no longer to be
proven. They allow storage and analysis of huge
amounts of structured business data (Perez and al.,
2008). These data are generally numeric values
tracing the whole activities of the organization and,
therefore, are used as the basis for
evaluating/analyzing the business performances as
well as for well-founding decisional processes
managed by decision makers (i.e., analysts).
Recently, enterprises have become aware that DW is
solving a small part of their real integration and
analysis needs (Pedersen, 2009). Indeed, 20%
information is extracted from data warehouses as
numeric data and the other 80% information is
hidden in non-numeric data (i.e., in documents)
(Tseng and Chou, 2006). Consequently, decisional
data (i.e., data useful for the decisional process) can
be found not only as numeric values directly
recorded in database objects (i.e., tables or classes)
but may be spread in textual data as in reports. This
leads to enrich the Decision Support System (DSS)
by including documents. Those documents can be
structured, semi-structured or non-structured. As
example of documents, we can mention business
journals, credit reports and industry newsletters.
These documents help decision makers to better
understand the evolution of the corporate data over
time. However, they represent an important volume
to be integrated into the DSS.
On the other hand, the number of documents is
permanently growing through time. As a result,
decision makers pay out much of their working time
to explore documents looking for data that help them
in their decisional processes. Obviously, without
appropriate design and powerful software tools,
decisional analysts can not easily and rapidly
explore documents. As a consequence, in many
cases, when making important decisions, some
relevant documents may be ingnored whereas some
irrelevant ones may be considered by intuition. The
result may be imperfect because the decision making
process is based on incomplete information, or even
noisy (Tseng and Chou, 2006). In order to alleviate
this problem, documents must be warehoused
(McCabe and al., 2000).
In addition, documents are usually stored in
several distant geographical sites and may be
85
Ben Messaoud I., Feki J., Khrouf K. and Zurfluh G..
UNIFICATION OF XML DOCUMENT STRUCTURES FOR DOCUMENT WAREHOUSE (DocW).
DOI: 10.5220/0003502100850094
In Proceedings of the 13th International Conference on Enterprise Information Systems (ICEIS-2011), pages 85-94
ISBN: 978-989-8425-53-9
Copyright
c
 2011 SCITEPRESS (Science and Technology Publications, Lda.)

presented in many different formats. Among these
formats, the most popular one in use is XML
«Extensible Markup Language». So, we focus on
XML documents since XML is the most used format
for data representation and exchange. In general,
those documents are described by several structures.
Consequently, a step to unify the structures of XML
documents is compulsory in order to produce a
global view describing a large document set. This
step should be followed by a multidimensional
modeling of documents; it enables multidimensional
analysis of the documents content. In this paper, we
present a method to build a document warehouse.
This method is composed by two steps namely:
unification of XML document structures, and
multidimensional modeling of these documents.
More specifically, we focus on the unification step
that produces a global view for the distributed
document warehouse (DocW).
This paper is organized as follows. In section 2,
we describe the most popular works that treat DocW
and unification of structures of XML documents.
Secondly, in section 3, we propose our method to
build a DocW. Then, we present the process to unify
the sructures of XML documents set. After that, we
present an example of unification in section 5.
Finally, in section 6, we conclude the paper and
overview the next steps of our current works.
2 RELATED WORK
In (Inmon, 1994), the author underlines the
importance of external contextual information to
understand the results of the historical analysis
operations. In fact, the external contextual
information is usually available in documents (e.g.,
on-line news and company reports). Those
documents can be presented as XML formats.
Usually, the structure of XML documents may be
different for documents belonging to the same
domain (e.g., DTDs of research papers are different).
In this section, we first define the concept of DocW
and then, we describe the most popular works
related to document warehousing. Finally, we depict
the relevant works that treat unification of XML
document structures.
A DocW is designed to store documents issued
from internal and external data sources. In fact,
within a DocW, documents are organized for
effective analysis, or feature extraction to enable
distilled and fruitful business intelligence (Tseng
and Chou, 2006). Moreover, a DocW can be seen as
a special case of a Content Warehouse. In fact, a
Content Warehouse archives Web data composed of
texts. This warehouse is derived into Text
Warehouses when little structure was available and
Document Warehouses when structure was available
(Ravat and al., 2010). In the literature, there are few
works regarding DocW; significant ones are (Tseng
and Chou, 2006), (Perez, 2007) and (Perez and al.,
2008).
In (Tseng and Chou, 2006), the authors present
architecture for a DocW where documents are
supplied by internal and external sources. Those
documents undergo a pre-processing treatment, such
as text summarization and text feature extraction
(front-end component). After that, the extracted
elements are stored as a metadata. The warehouse
administrator performs all the operations associated
with the management of the documents in the
warehouse, such as create dimensions and document
indexes for constructing document cubes and
archive documents. In addition, the proposed
architecture is composed of a back-end component
that performs all the necessary operations for the
queries management (document access tools, multi-
dimensional document query interface, etc.).
However, only quantitative reports can be made
because indicators are numerical. Furthermore, the
authors consider only the textual parts of documents
(excluding graphics, multimedia data, etc.).
Another work, (Perez, 2007) and (Perez and al.,
2008) propose an architecture that integrates
corporate warehouse with a DocW (text-rich XML
documents), resulting in a contextualized warehouse.
“A contextualized warehouse is a DSS that allows
users to combine all their sources of structured and
unstructured data and to analyze the integrated data
under different contexts”. In order to build a
contextualized OLAP cube, the analyst specifies, by
a sequence of keywords, the context of analysis. The
resulting cube is called R-cube (R stands for
Relevance). This cube is characterized with two
specific dimensions namely: relevance and context.
The relevance dimension depicts the importance of
each fact of the cube in the selected context. But, the
context dimension represents the documents of the
warehouse. Also, each fact in the R-cube is linked to
the set of relevant documents that describe its
context. However, the proposed approach provides
additional information for decisional
analysis. Indeed, it returns to the decision maker the
relevant documents to an analysis. Furthermore, the
process responsible for detecting the facts described
in the documents does not consider fact without
corresponding corporate facts.
ICEIS 2011 - 13th International Conference on Enterprise Information Systems
86

Other works, as (Júnior and Mello, 2008) and
(Yoo and al., 2005), treat the unification of the
structures of XML documents.
In (Júnior and Mello, 2008), the authors propose
an approach for semantic integration of XML
instances. This approach is composed of two
processes: Similarity definition and Unification. The
similarity definition compares each pair of instances
of document and stipulates a similarity scores for
them. It generates several sets of XML instances.
Those sets contain XML instances semantically
similar. The second process (i.e., unification)
generates a unified XML representation for each set.
It uses domain ontology and a dictionary in order to
name and to structure the resulting XML instance.
Nevertheless, the comparison of documents may be
time consuming because some of these documents
can have the same structure. The processing time
can be reduced by raising the comparison to the
level of documents structures.
(Yoo and al., 2005) propose an algorithm for the
unification of DTDs for XML documents having
similar structures and belong to the same
domain. This algorithm accepts a set of similar
DTDs (Document Type Definition) and generates a
unified DTD. In fact, a unified DTD plays the role
of a global conceptual schema for subjects common
to a given domain. The algorithm uses finite
automata and tree structure to represent
documents. It includes four steps: Pre-processing,
DTD representation, Uniform DTD generation and
Post processing. First, the pre-processing resolves
names ambiguities of the elements of DTDs. It
unifies the names of elements having the same
meaning using an Element Name Resolution
Table. In the second step, DTDs are represented as
trees and finite automata. Then, trees and finite
automata are merged together to create a unified
DTD. Finally, the final DTD is verified using a
DTD parser. However, the quality of result relies on
the completeness of the Element Name Resolution
Table.
After this short overview of related works, we
focus on the work related to DocW instead of
heterogeneity and distribution issues for XML
documents. Also, we emphasize that usually the
designer does not participate in the process of
unification of XML documents. The remaining of
this paper presents our method to build a DocW.
3 BUILDING A DocW
Documents help decision makers to better
understand the evolution of the corporate data over
time. However, they represent an important volume
to be integrated into the DSS. In (Tseng and Chou,
2006), the authors claim that 20% information
extracted from data warehouses is numeric data and
the other 80% information is hidden in documents.
In fact, there are several formats of documents, such
as XML. XML documents allow the exchange of a
wide variety of data on the Web and elsewhere
(http://www.w3.org/XML/).
In general, documents are distributed on various
sites and their structures differ from one site to
another. So, to query several sites of documents, we
need a common structure which plays the role of the
global schema in a distributed database. Therefore,
we propose a method to build a DocW composed of
distributed XML documents. This method consists
of two steps (Ben Messaoud and al., 2010):
 Unification of XML document structures.
 Multidimensional modeling of documents.
The first step defines a common structure to
describe XML documents located at different sites.
We can assume that all documents stored at the same
site have a unique and common structure.
The second step aims to design multidimensional
schemas that highlight the OLAP analysis
components: facts, measures, axes (i.e., dimensions),
perspectives (i.e., hierarchical levels). It is currently
studied as a complementary task for this work (Ben
Messaoud and al., 2011).
In the remainder of this paper, we restrict
ourselves to detail our proposal for the unification of
XML document structures.
4 UNIFICATION OF XML
DOCUMENT STRUCTURES
The content of an XML document is encapsulated
within elements that are defined by tags. Those
elements are hierarchically organized as a tree.
Syntactically, two formalisms may be used to
describe the structure of XML documents namely:
DTD and XSchema. Usually, XML document
structures may differ among documents. As a result,
when a decision maker needs to retrieve information
from several sites where documents are stored, he
has to consider this structural heterogeneity. This
may lead him/her to write multiple queries; i.e., as
many queries as the number of different structures
for the document set and, therefore,
manually/programmatically build the final result
using the returned sub-results. To overcome this
UNIFICATION OF XML DOCUMENT STRUCTURES FOR DOCUMENT WAREHOUSE (DocW)
87

Tre e
representation
Validation of
unified tree(s)
Wo r d n e t
Structures of
XML documents
.
.
Site 1
Site 2
Site n
Generation of
unified trees
Tre es
Unified trees
Unified trees
validated
Repository
Trees
Figure 1: Unification of XML document structures.
problem, we expect to define a unified structure; i.e.,
a structure convenient to describe multiple
heterogeneous XML documents.
Thus, this unified structure will be useful to write
a single query that operates on several documents
simultaneously.
In order to build this unified structure for
structurally heterogeneous XML documents, we
propose a method that exploits the XML tree
structure. Our unification method consists of the
three following steps:
 Tree representation.
 Generation of unified trees.
 Validation of unified tree(s).
Figure 1 shows the sequencing of these steps.
The first step translates the XML document structure
into the formalism of trees as done in (Yoo and al.,
2005) and (Lee and al., 2002). Such a result tree
characterizes the site from which it is issued its
nodes represent the elements of an XML structure
with their cardinalities, and its arcs denote
relationships between XML elements. We choose
the tree formalism because it is easy to be
understood by end-users (i.e., decision-makers). In
fact, the usage of trees motivates decision-makers to
participate in the next steps, mainly during the
unification process. Figure 2 depicts an example of
DTD and its corresponding tree; it is for research
papers belonging to the site called Laboratory1.
In the remainder of this paper, we note a tree T as
follows: T (E, r, N, S) where E represents the set of
all nodes of T, r is its root node (r
∈ E), N is the set
of arcs of T and S is the site of XML structure
characterized by its DTD. Also, each node e
i
∈ E is
noted as e
i
(n
i
, c
i
) where n
i
is the node name and c
i
represents its cardinality, if any.
According to this notation, the tree of Figure 2
which belongs to the site S {Laboratory 1} is made
up of a set of eight nodes E {Article, Title, (Author,
+), (Section, +), Name, Affiliation, (Title, ?), (Para,
+)} from which one is the root r {Article} and a set
of seven arcs N {Article – Title, Article – Author,
Article – Section, etc.}.
Figure 2: An example of DTD and its tree.
4.1 Generation of Unified Trees
This generation compares all trees resulted from the
previous step (i.e., tree representation step) and then
produces a unified tree set; i.e., trees not yet
validated by the decision-maker. It is composed of
the three following sub steps:
 Semantic processing of the tree nodes.
 Similarity calculation.
 Unified trees production.
ICEIS 2011 - 13th International Conference on Enterprise Information Systems
88

The semantic processing resolves the semantic
ambiguities of the node names, belonging to several
trees, by using the Wordnet ontology. To do so,
nodes having the same meaning are detected and
their names are substituted with a unique/standard
name. Secondly, for the obtained trees, we define a
similarity factor that helps to determine which trees
are analogous and, therefore, have to be merged.
Finally, we produce a unified tree for each set of
similar trees according to the similarity factor.
4.1.1 Similarity Calculation
Once all XML document structures are translated
into trees, we need to build a unified tree by fusion;
that is a global structure for the initial document set.
To realize this fusion, we define a similarity factor
(noted Sim) that measures the pertinence of trees to
be integrated; its calculation is based on the number
of the common nodes between two trees.
The similarity factor of two trees T
i
(E
i
, r
i
, N
i
, S
i
)
and T
j
(E
j
, r
j
, N
j
, S
j
) noted Sim(T
i
, T
j
) is calculated
according to formula (1).
()
⎪
⎩
⎪
⎨
⎧
<=
=
otherwise
q
c
nnandcnIf
TTSim
ji
jijii
ji
,
,
75.0
,
(1)
Where:
n
i
= | E
i
|, n
j
= | E
j
|, c
i,j
= | E
i
∩ E
j
| et q = n
i
+ n
j
- c
i,j
In this formula, the similarity factor calculation
is based on n
i
, n
j
and c
ij
representing, respectively,
the number of nodes in trees T
i
, T
j
and the number of
their common nodes.
Now, given a set of n trees, we need to
determine which trees have to be merged first; i.e.,
which trees are more closely in structure than others.
For this purpose, we define the Similarity Matrix
(noted SM). It is inspired from the matrix presented
in (Feki, 2004) used for multidimensional schema
integration. Our SM matrix is a triangular matrix
with n trees in rows and in columns. Each cell
SM(i,j) contains the similarity factor Sim(i,j) of tree
in row i and tree in column j. This matrix facilitates
the search of the trees having the highest Sim. We
note that the comparison of two trees T
i
and T
j
is
meaningful when Sim(T
i
, T
j
) is greater than a given
threshold determined by experimentation.
4.1.2 Generation Principle
The unified trees production step merges each two
trees identified in the previous step.
In the literature, (Golfarelli and al., 1998) and
(Golfarelli and Rizzi, 1999) propose algorithms to
treat the attribute tree so as to generate an XML
schema from an XML cube. The attribute tree may
be pruned and grafted in order to eliminate the
unnecessary levels of detail. Likewise, (Boussaid
and al., 2006) use the merging operators: pruning
and grafting in order to compare XML document
trees and the tree of the multidimensional conceptual
model.
In fact, pruning is carried out by removing
portions of the trees. Figure 3 (a) shows an example
of pruning the two trees T
1
and T
2
. The result tree T
3
represents their merged tree. It contains only the
common sub-trees to T
1
and T
2
. The uncommon
nodes are dropped with their sub-trees, if any.
Grafting is used when sub-trees have not the
same structure. Figure 3 (b) depicts an example of
grafting for T
1
and T
2
. In these trees, the sub-trees b-
(e, f) of T
1
and b-(e, x, f) of T
2
don’t have the same
structure (also c-(y, g) and c-(g)). In the result tree
(i.e., T
3
), the common nodes and their arcs are
maintained while the other nodes are dropped.
However, the application of the pruning and grafting
operators leads to the intersection of trees.
a
b c
ye
(b)Grafting
f g
T
1
T
2
T
3
a
b d
ke
a
b d
kf
a
b d
kf
(a)Pruning
c
f l
m
l l
T
1
T
2
T
3
a
b c
ge
x
f
a
b c
ge
f
Figure 3: Example of merging trees through pruning (a)
and grafting (b).
Let us recall that in our work we want to build a
common structure that acts as the global schema in
distributed databases. This structure is obtained by
unification of XML tree structures issued from
several distributed sites. For this unification we need
a set of operators: We retain only the merging
operator grafting as defined in (Golfarelli and al.,
1998), (Golfarelli and Rizzi, 1999) and reused in
UNIFICATION OF XML DOCUMENT STRUCTURES FOR DOCUMENT WAREHOUSE (DocW)
89

(Boussaid and al., 2006). Note that the pruning
eliminates nodes/sub-trees and therefore is not
useful because it leads to eliminate sub-trees. In
addition, we propose the following three operators:
fusion by inclusion, fusion by union of sub-trees and
fusion by merging nodes.
Definition 1. The fusion by inclusion F-Inclusion
operator of two trees T
1
and T
2
produces a tree T
3
such as T
3
=T
2
if T
1
⊆ T
2
and T
3
= T
1
if T
2
⊆ T
1
. Its
syntax is as follows:
F-Inclusion (T
1
, T
2
) = T
3
Input:
- T
1
(E
1
, r
1
, N
1
, S
1
).
- T
2
(E
2
, r
2
, N
2
, S
2
).
Conditions:
- T
1
⊆ T
2
or T
2
⊆ T
1
.
Output:
- T
3
= T
2
if T
1
⊆ T
2
.
- T
3
= T
1
if T
2
⊆ T
1
.
Figure 4 (a) gives an example for such a fusion.
Definition 2. The fusion by union operator F-Sub-
Trees of two sub-trees produces a tree T
3
composed
of sub-trees belonging simultaneously to T
1
and to
T
2
.
F-Sub-Trees (T
1
, T
2
) = T
3
Input:
- T
1
(E
1
, r
1
, N
1
, S
1
).
- T
2
(E
2
, r
2
, N
2
, S
2
).
Conditions:
- E
1
∩ E
2
≠ ∅.
- ∃ e
i
∈ E
1
and e
j
∈ E
2
∀ e
i
= e
j
, Parent(e
i
) =
Parent(e
j
) and Child(e
i
) ≠ Child(e
j
).
/* Parent (e
i
) returns the parent of node e
i
*/
/* Child (e
i
) determines sub-trees of node e
i
*/
Output:
- T
3
(E
3
, r
3
, N
3
, S
3
) with
- E
3
= E
1
∪ E
2
- N
3
= N
1
∪ N
2
- r
3
= r
1
= r
2
- S
3
= {S
1
, S
2
}
Figure 4 (b) illustrates this operator.
Definition 3. The fusion by merging nodes F-
Merging-Nodes produces a tree T
3
composed by a
specific node.
F-Merging-Nodes (T
1
, T
2
) = T
3
Input:
- T
1
(E
1
, r
1
, N
1
, S
1
).
- T
2
(E
2
, r
2
, N
2
, S
2
).
Conditions:
- E
1
∩ E
2
≠ ∅
- ∃ e
i
∈ E
1
and e
j
∈ E
2
∀ e
i
≠ e
j
, Child(e
i
) = Child(e
j
)
and Parent(e
i
) = Parent(e
j
).
Output:
- T
3
(E
3
, r
3
, N
3
, S
3
) with
- E
3
= (E
1
– {e
i
}) ∪ (E
2
– {e
j
}) ∪ (e
i
| e
j
)
/* (e
i
| e
j
) is a specific node or */
- N
3
= N
1
∪ N
2
- r
3
= r
1
= r
2
- S
3
= {S
1
, S
2
}
Figure 4 (c) shows an example where the two
sub-trees b-(e, f) and d-(e, f) become the sub-tree
b|d-(e, f) linked to their original common node a in
T
3
. Note that b|d signifies a node named b or d.
While merging trees, cardinalities are treated in
respect to rules defined in (Hachaichi and al., 2010).
a
b c
ed
a
b c
ed
x
a
b c
ed
x
(a)Fusionbyinclusion
a
b c
k
a
d c
e
a
b d
mk
(b)Fusionbyunionof sub‐trees
l m
c
l e
a
b c
e
a
d c
e
(c)Fusionbymergingnodes
f f
a
b|d c
e
f
T
1
T
2
T
3
T
1
T
2
T
3
T
1
T
2
T
3
Figure 4: Merging trees through fusion by inclusion (a),
fusion by union of sub-trees (b) and fusion by merging
nodes (c).
4.1.3 Algorithm to Merge Trees
In order to produce unified tree(s), we propose a
Tree-Merge algorithm. It uses the similarity matrix
to identify trees to be merged, merge them using the
operators defined in section 4.1.2 and then produces
unified trees. The input and output trees are stored
according to the meta-model of Figure 5 where we
associate to each node of unified-trees its origin; i.e.,
one or several sites from which it issues. This
association is fundamental because it prepares for
querying the distributed DocW.
ICEIS 2011 - 13th International Conference on Enterprise Information Systems
90

Algorithm Tree-Merge
Input
T = {T
1
, T
2
…, T
i
…, T
n
}: a non empty
set of n trees where
),,,(
iiiii
SNrET =
Threshold: a real value in [0.5..1[
Output
R = {R
1
, R
2
, R
m
}: m(m ≤ n) merged trees.
Begin
R := T
Calculate_SM (T) /* calculates the
similarity matrix SM [n, n] */
Size_SM := n
For each i ∈ [1..(Size_SM – 1)] do
Max := Determine-Max(SM)
If (Max ≥ Threshold) Then
For each j ∈ ]i..Size_SM] do
If (SM[i,j] = max) Then
Mark row i and column j
For each K (i<k<j) do
If (SM[k,j] = max) Then
Mark row k
End If
End For
End If
End For
R = R – {trees corresponding to rows
and columns marked}
/* Merge trees corresponding to rows
and columns marked */
If (T
i
⊆ T
j
or T
j
⊆ T
i
) Then
R := R ∪ F-Inclusion (T
i
, T
j
)
Else
If (E
i
∩ E
j
≠ ∅ and (
∃
e
i
∈ E
i
and e
j
∈ E
j
and e
i
= e
j
and
Parent(e
i
) = Parent(e
j
) and
Child(e
i
) ≠ Child(e
j
) )) Then
R := R ∪ F-Union-Sub-trees(T
i
,T
j
)
Else
If (Ei ∩ Ej ≠ ∅ and (
∃
ei ∈ Ei
and ej ∈ Ej and Child(ei) =
Child(ej) and Parent(ei) =
Parent(ej) and ei ≠ ej)) Then
R := R ∪ F-Merging-Nodes(T
i
,T
j
)
End If
End If
End If
Delete rows and columns marked
Calculated-SM (R)
Size_SM := |R|
Else
Stop
End If
End For
End.
4.2 Validation of Unified Tree(s)
In this third and last step (cf. figure 1), the decision-
maker/designer should validate the unified tree(s) by
adjusting them to his/her analytical requirements:
delete and/or rename nodes. The result tree(s) are
saved in a specific repository (cf. figure 5). This
repository is later used in querying the distributed
DocW.
5 EXAMPLE
In this section, we apply our method on an example
of four DTDs (cf. Figure 6) issued from four
distributed sites.
Applying the first step, each DTD is translated
into a tree (cf. Figure 7). Secondly, semantic
ambiguities are resolved by using the Wordnet
ontology and produces trees shown in Figure 8.
Thus, in tree T1 Writer is replaced with its synonym
Author much more present in the other trees. Also,
the node Section (in T
3
) is changed into Paragraph.
Similarly, Symposium (in T
4
) is standardised as
Conference. After that, the similarity matrix is
calculated in order to find trees to compare.
In our running example, we set the threshold
value to 0.5.
The Tree-Merge algorithm iterations are:
First iteration:
Input: T = {T
1
, T
2
, T
3
, T
4
}
n = 4 ; Size_SM = 4
T
1
1
T
2
2
T
3
3
T
4
4
T
1
1
0.75
0.12 0.12
T
2
2
0.16 0.27
T
3
3
0.66
T
4
4
Since SM[1,2]= 0.75 (is greater than the threshold
0.5) then T
1
will be merged with T
2
: T
1
is included
in T
2
thus, the algorithm performs the fusion by
inclusion operator that produces T’ (which is similar
to T
2
). The result of this step is the set R = {T
3
, T
4
,
T’} (cf. figure 9).
During this iteration, rows 1 and column 2 are
marked in order to determine which trees to merge.
Second iteration:
Input: {T
3
, T
4
, T’}
T
3
1
T
4
2
T’
3
T
3
1
0.66
0.16
T
4
2 0.27
T’ 3
As SM[1,2]= 0.66 (> 0.5), then T
3
will be
merged with T
4
by applying the fusion by merging
nodes operator. Indeed, the node (Paper) and
(Article) have the same sub-trees in T
3
and T
4
. In the
UNIFICATION OF XML DOCUMENT STRUCTURES FOR DOCUMENT WAREHOUSE (DocW)
91

Tree
Tree_Name
Node
Node_ Name
Node_Synonym
Node_Cardinality
1..n1..n 1..n1..n
1
+Child
+Parent 1
Tree_Node
Node_Level
Site
Site_Name
Localisation
Structure
Unified Tree
Non_Unified Tree
1..n
1
1..n
1
1
1..n
1
1..n
Figure 5: Class diagram for the storage of unified trees.
DTD 1
< ! ENTITY Affiliation (University | Industry) >
< ! ELMENT Writer (Name, %Affiliation, Style) >
< ! ELMENT Name (#PCDATA) >
< ! ELMENT University (#PCDATA) >
< ! ELMENT Industry (#PCDATA) >
< ! ELMENT Style (#PCDATA) >
DTD 2
< ! ELMENT Publication (Title, Author + , Article) >
< ! ELMENT Title (#PCDATA) >
< ! ELMENT Article (Abstract, Body) >
< ! ELMENT Abstract (#PCDATA) >
< ! ELMENT Body (#PCDATA) >
< ! ELMENT Author (Name, (University | Industry), Style) >
< ! ELMENT Name (#PCDATA) >
< ! ELMENT Industry (#PCDATA) >
< ! ELMENT Style (#PCDATA) >
< ! ELMENT University (#PCDATA) >
DTD 3
< ! ELMENT Paper ( Title, Section +, Author +, Conference)
>
< ! ELMENT Title (#PCDATA) >
< ! ELMENT Section (#PCDATA) >
< ! ELMENT Author (#PCDATA) >
< ! ELMENT Conference (#PCDATA) >
DTD 4
< ! ELMENT Article ( Title, Author +, Paragraph +,
Symposium) >
< ! ELMENT Title (#PCDATA) >
< ! ELMENT Author (#PCDATA) >
< ! ELMENT Paragraph (#PCDATA) >
< ! ELMENT Symposium (#PCDATA) >
Figure 6: Example of four DTDs.
Figure 7: Generated trees for the four DTDs (of Figure 6).
ICEIS 2011 - 13th International Conference on Enterprise Information Systems
92

Figure 8: Trees after semantic processing of their nodes.
Figure 9: The result trees T
3
, T
4
and T’.
Figure 10: The result trees T’ and T".
result tree T", the node (Paper|Article) appears. The
set R is composed of the two trees T’ and T" (cf.
figure 10).
Third iteration:
Input: {T’, T”}
T’
1
T”
2
T’
1 0.16
T” 2
Since SM[1, 2] = 0.16 (< 0.5), then there are no
trees to merge. As a consequence, T’ and T"
represent the final set of unified trees. As a final step
of our method, these trees are presented to the
decision-maker/designer in order to adjust them to
their analytical requirements.
6 CONCLUSIONS
In this paper, our main concern was to enrich the
data warehouse with textual data in order to improve
the quality of decisional analysis. More specifically,
we have proposed a method to build a document
warehouse. This method is composed of two main
steps: unification of XML document structures, and
multidimensional modeling of documents. We have
focused on the unification step which unifies
heterogeneous XML document structures issued
from several sites. To do so, we have translated
XML structures into trees, solved semantic
ambiguities of their nodes using Wordnet ontology,
and then merged those trees producing unified tree.
Finally, the unified trees are presented to the user in
order to validate them according to his analytical
requirements. We note that the unified trees
represent a common structure that will be used later
in querying documents.
UNIFICATION OF XML DOCUMENT STRUCTURES FOR DOCUMENT WAREHOUSE (DocW)
93

Currently, we are developing a software
prototype to support our proposed unification
method. As a long term perspective for this work,
we will propose a querying language for the
distributed document warehouse.
REFERENCES
Ben Messaoud, I., Feki, J., Zurfluh, G., 2010. Unification
des structures des documents XML pour l’entreposage
de documents. In ASD’10, Cinquième Atelier sur les
Systèmes Décisionnels, pages 1-12, ISBN 9973-9900-
2-0, Sfax, Tunisie.
Ben Messaoud, I., Feki, J., Zurfluh, G., 2011.
Modélisation multidimensionnelle des documents
XML. In EDA’11, 7
ème
journée francophones sur les
Entrepôts de Données et d’Analyse en ligne, Clermont
Ferrand, France (To appear).
Boussaid, O., Ben Messaoud, R., Choquet, R., Anthoard,
S., 2006. Conception et construction d’entrepôts en
XML. In EDA’06, 2ème journée francophone sur les
Entrepôts de Données et l'Analyse en ligne, Versailles,
France.
Golfarelli, M., Maio, D., Rizzi, S., 1998. Conceptual
Design of Data Warehouses from E/R Schema. In
HICSS’98, Proceedings of the 31st Annual Hawaii
International Conference on System Sciences, IEEE
Computer Society, pages 334–343, Washington, DC,
USA.
Golfarelli, M., Rizzi, S., 1999. Designing the Data
Warehouse : Key Steps and Crucial Issues. In Journal
of Computer Science and Information Management
2(3), pages 88–100.
Feki, J., 2004. Vers une conception automatisée des
entrepôts de données : Modélisation des besoins
OLAP et génération de schémas multidimensionnels.
In MCSEAI’04, 8th Maghrebian Conference on
Software Engineering and Artificial Intelligence,
pages 473-485, ISBN 9973-37-193-3, Sousse, Tunisie.
Hachaichi, Y., Feki, J., Ben-Abdallah, H., 2010.
Modélisation multidimensionnelle de documents XML
centrés-données. In Journal of Decison Systems, vol
19/3, pages 313-345.
Inmon, W., H., 1994. Building the Data Warehouse. John
Wiley & Sons.
Júnior, C., A., S., Mello, R., S., 2008. An ontology-driven
process for unification of XML instances. In Brazilian
Symposium on Multimedia and the Web, pages 242-
249, Vila Velha, Brazil.
Lee, M., L., Yang, L., H., Hsu, W., Yang, X., 2002.
XClust: clustering XML schemas for effective
integration. In CIKM’02, Proc. of the ACM
International Conference on Information and
Knowledge Management, pages 292–299, McLean,
Virginia,
McCabe, M., C., Lee, J., Chowdhury, A., Grossman, D.,
Frieder, O., 2000. On the design and evaluation of a
multi-dimensional approach to information retrieval.
In Proceedings of the 23th Annual International ACM
SIGIR Conference, pages 363–365.
Pedersen, T., B., 2009. Warehousing The World: A vision
for Data Warehouse Research. In Kozielski S.,
Wrembel R. (Eds.): New Trends in Data Warehousing
and Data Analysis. Annals of Information Systems,
Vol.3.
Pérez, M., J., M., 2007. Contextualizing a data warehouse
with documents. Thèse de doctorat. Université Jaume
I, Spain.
Pérez, M., J., M., Berlanga, L., M., R., Aramburu, C., M.,
J., Pederson, T., B., 2008. Contextualizing data
warehouses with documents. In Decision Support
System (DSS), Elsevier, pages 77-94.
Ravat, F., Teste, O., Tournier, R., Zurluh, G., 2010.
Finding an application-appropriate model for XML
data warehouses. In Information Systems, volume 35,
issue 6, pages 662-687.
Tseng, F., S., C., Chou, A., Y., H., 2006. The concept of
document warehousing for multi-dimensional
modeling of textual-based business intelligence. In
Decision Support Systems (DSS), vol 42, Elsevier,
pages 727– 744.
Yoo, C., S., Woo, S., M., Kim, Y., S., 2005. Unification of
XML DTD for xml Documents with Similar Structure.
In Computational Science and its Applications –
ICCSA, LNCS 3482, pages 954-963.
ICEIS 2011 - 13th International Conference on Enterprise Information Systems
94