OLAP AGGREGATION FUNCTION FOR TEXTUAL DATA
WAREHOUSE
Franck Ravat
(2)
, Olivier Teste
(1)
and Ronan Tournier
(1)
IRIT, team SIG-ED: Systèmes d’Informations Généralisés – Entrepôts de Données
(1)
IRIT,Université Toulouse 3, 118 rte de Narbonne
F-31062 Toulouse Cedex9, France
(2)
IRIT,Université Toulouse 1, 2 rue du doyen G. Marty
F-31042 Toulouse Cedex 9, France
Keywords: OLAP, Data Warehouse, Aggregation Function, Document Warehouse, Non-Additive Measure.
Abstract: For more than a decade, OLAP and multidimensional analysis have generated methodologies, tools and
resource management systems for the analysis of numeric data. With the growing availability of semi-
structured data there is a need for incorporating text-rich document data in a data warehouse and providing
adapted multidimensional analysis. This paper presents a new aggregation function for keywords allowing
the aggregation of textual data in OLAP environments as traditional arithmetic functions would do on nu-
meric data. The AVG_KW function uses an ontology to join keywords into a more common keyword.
1 INTRODUCTION
OLAP (On-Line Analytical Processing) systems
allow analysts to improve decision-making process
by analysing aggregated historical business data.
These analyses are based on a centralized data re-
pository, called a data warehouse (Kimball, 1996).
Within data warehouses, the use of Multidimen-
sional DataBases (MDB) enables decision-makers to
gain insight into an enterprise performance.
1.1 Context and Motivations
Multidimensional OLAP analysis displays analysis
subject data according to various levels of detail
(data granularity). The process aggregates the data
according to the level of detail with functions such
as sum, average, maximum, minimum… Drilling
operations are the most common OLAP operations.
They consist in allowing the analyst to change the
displayed data granularity, thus the analysed data is
aggregated according to a new granularity level. In
Figure 1, a decision-maker analyses the number of
keywords monthly used by authors. In order to get a
more global view on the data, he changes the display
by years (he “rolls-up”). As a consequence, the
monthly values are aggregated into a value for each
year.
Figure 1: Multidimensional analysis of keyword counts
displayed by authors and by months and rolled-up to
years.
According to (Tseng and Chou, 2006) 20% of
corporate information system data is transactional,
i.e. numeric. This may easily be processed because
multidimensional analysis is robust and it is a mas-
tered technique on numeric-centric data warehouses
(Sullivan, 2001). The remaining 80%, namely tradi-
tional “paperwork,” stays out of reach of OLAP
processes due to the lack of tools and resource man-
agement for non-numeric textual data such as text-
rich documents. OLAP provides powerful tools and
methods but within a rigid framework. Unstructured
documents do not fit in this framework. Recently,
XML technology has provided a wide framework
for sharing, spreading and working with documents
within corporate networks or over the web. Thus,
storing documents and semi-structured data was
integrated within data warehouses and repositories.
151
Ravat F., Teste O. and Tournier R. (2007).
OLAP AGGREGATION FUNCTION FOR TEXTUAL DATA WAREHOUSE.
In Proceedings of the Ninth International Conference on Enterprise Information Systems - DISI, pages 151-156
DOI: 10.5220/0002364401510156
Copyright
c
SciTePress
Document warehousing slowly emerged as solutions
were created (Sullivan, 2001), e.g. Xyleme
1
.
We argue that, to provide more exhaustive mul-
tidimensional analyses, OLAP decision support
systems should provide the use of a 100% of corpo-
rate information system data. But, up to now, the
OLAP framework lack the ability to cope with the
analysis of semi-structured text-rich document data.
As a consequence, there is a need for adapted con-
ceptual models and textual aggregation processing.
1.2 Related Works
Related works may be divided according to two ma-
jor categories. Firstly is the integration of XML data
with 1) physical integration of XML data into a data
warehouse. (Pokorný, 2001) builds a star schema on
a logical XML structure; (Niemi et al., 2002) assem-
bles “on the fly” XML data cubes from user queries;
(Zhang et al., 2003) deals with building data ware-
houses on top of XML data and (Vrdoljak et al.,
2003) creates a data warehouse multidimensional
schema from XML schemas; and 2) the association
of XML data with a data warehouse (logical integra-
tion). In (Yin and Pedersen, 2004), the authors
federate XML data and traditional multidimensional
data into an OLAP system. Although all these works
consider textual data through the use of XML docu-
ments, they are all based on numeric-centric analysis
and lack support for text-rich document-centric data
analysis.
The second category concerns multidimensional
analysis of documents within an OLAP framework.
In (Pérez et al., 2005) the authors combine tradi-
tional numeric analysis and information retrieval
techniques to assist multidimensional analysis by
providing relevant documents to the ongoing analy-
sis context. In (McCabe et al., 2000) and (Mothe et
al., 2003), the authors propose the use of traditional
OLAP framework to count documents according to
keywords or topics in order to query more precisely
a document collection. Similarly in (Chakrabarti et
al., 1998) and (Agrawal et al., 2000), the authors
offer tools and methods to efficiently build a hierar-
chical classification of documents based on typical
keywords. In (Tseng and Chou, 2006) and (Keith et
al., 2005), the authors suggest to build a specific
keyword dimension to allow multidimensional
analysis of documents. Nowadays, industrial solu-
tions start to appear such as Text OLAP
2
. In (Khrouf
et al., 2004) the authors describe a document ware-
house where documents are grouped by similar
structures; multidimensional analysis may be per-
formed but still with the use of numeric analysis.
These advanced propositions show the follow-
ing limitations: 1) textual data is difficult to analyse
as systems use numeric measures to get round the
analysis of non-numeric data; 2) the most advanced
systems are limited to counting keywords in docu-
ment sets; and 3) non numeric indicators may not
be processed. Finally, in (Park et al., 2005), the au-
thors introduce the concept of multidimensional
document analysis within an XML framework. Un-
fortunately, all aggregation functions using text
mining techniques are not detailed.
1.3 Aims and Contributions
The next step of decision making is to leap ahead of
numeric indicators and to allow the powerful OLAP
framework to operate on non-numeric data. Con-
trarily to previously stated works, we wish to focus
the analysis on text. Our approach has the advan-
tage of combining qualitative analysis with
quantitative analysis, e.g. the analysis of the key-
words of a specific publication, in order to provide
an overview of publication contents. To allow mul-
tidimensional OLAP analysis of documents, we
provide an aggregation function for textual OLAP
analysis. This function is based on a conceptual
model that provides: 1) adapted concepts to support
non-numeric textual measures; and 2) a new con-
cept to drive OLAP textual aggregation processing
with the use of a domain ontology.
The rest of this paper is organised as follows:
section 2 defines the conceptual model and section
3 describes the aggregation function AVG_KW.
2 CONCEPTUAL MODEL
In this section we define an extension a traditional
multidimensional model to handle textual data
analysis. We provide the addition of specific textual
measures as well as a hierarchical representation of
the analysed concepts with the use of an ontology.
2.1 Multidimensional Model
Multidimensional models have been used for over a
decade. See (Torlone, 2003) for recent survey. Most
use facts and dimensions to model multidimen-
sional structures.
Dimensions model analysis axes and are com-
posed of a set of parameters which are organised
into one or more hierarchies. Each hierarchy repre-
sents an analysis perspective along the axis. The
1
Xyleme server from http://www.xyleme.com
2
http://www.megaputer.com/products/pa/
ICEIS 2007 - International Conference on Enterprise Information Systems
152
parameters represent different levels according to
which analysis data may be observed.
The subject of analysis, namely a fact, is a con-
ceptual grouping of measures which are numeric
indicators. These measures are traditionally numeric
and may be additive, semi-additive or non-additive
(Kimball, 1996), (Horner et al., 2004). Here, analy-
sis of textual data requires textual measures that fall
into non-numeric and non-additive categories.
Definition 1. A textual measure is a measure that
holds textual data, i.e. non-numeric and non-
additive data.
A textual measure represents words, strings,
paragraphs or even whole documents. Within these
measures, we define the following categories:
Definition 2. A raw textual measure is a textual
measure that corresponds to the full text of a docu-
ment or to a fragment of that document.
Definition 3. A keyword measure is an elaborated
textual measure, where each measure instance x
i
is
represented by x
i
= (kw
i
, d
i
) such that kw
i
is a key-
word and d
i
a distance.
Raw textual measures are provided for flexibil-
ity, allowing the user to consult document contents.
Keyword measures require a certain amount of
pre-processing in order to be created. The domain of
all keywords is dom(kw). Notice that x
i
X with
X=dom(kw) ×
and all distances d
i
=0, this value
will be used during the aggregation process.
For example, to get a view of the subjects of a
collection of scientific articles, a decision-maker
analyses keywords used by authors. The fact Articles
has a numeric measure: Acceptance, corresponding
to the acceptance rate of each article; and two textual
measures: the raw textual measure representing the
complete article (Text) and the elaborated textual
measure (Keywords) which holds keywords ex-
tracted from article bodies. The resulting
multidimensional schema is displayed in Figure 2.
Graphic notations are inspired by (Golfarelli et al.,
1998).
Figure 2: Example of a multidimensional conceptual
schema for textual analysis.
2.2 Ontology and Operations
In order to allow analysis of textual measures, we
use a hierarchical representation of domain con-
cepts. These concepts are modelled through a
“light” or “informal is-a” ontology (Lassila and
McGuinness, 2001). It corresponds to a hierarchy of
domain concepts where each node represent a con-
cept (a keyword) and each link between nodes
models a more complex relation than an “is-a” rela-
tion.
Definition 4. Given an ontology O, the domain
of O, noted dom(O), represents all the keywords
of O.
For example in Figure 3, OLAPdom(O_IS).
Definition 5. We call depth of an ontology the
maximum number of nodes between the root
node and lowest nodes, i.e. the leaves.
In our example the depth(O_IS) = 8.
Figure 3: Example of a simple domain ontology on in-
formation systems named O_IS.
To allow the model to operate with the ontol-
ogy, we provide two operations that take two
nodes—keywords—as input: n
1
and n
2
.
Definition 6. The Least Common Ancestor:
(
)
(
)()
()
LCA
nnn
OdomOdomlca
a
21
2
,
:
is a function returning the least common ancestor
(n
LCA
) within O between n
1
and n
2
.
Definition 7. The Distance between two nodes:
(
)
(
)
() ()()()()()
21221121
2
nnlcandnnlcandnn
NOdomd
,,,,,max,
:
a
is a function that returns the number of nodes be-
tween the least common ancestor (LCA) and the
lowest node.
In O_IS, lca(ROLAP, Document Warehouse)=
Storage. The distance between these two keywords
is 4: d(ROLAP, Document Warehouse) = max
(d(ROLAP, Storage), d(Document Warehouse, Stor-
age)) = max (4, 1) = 4.
OLAP AGGREGATION FUNCTION FOR TEXTUAL DATA WAREHOUSE
153
3 AGGREGATION FUNCTION
Multidimensional OLAP analysis on non-additive
measures is very limited because actual systems pro-
vide only two aggregation functions: COUNT and
LIST (Kimball, 1996). We redefine the LIST func-
tion in order to operate on a keyword measure.
Definition 8. LIST aggregation function:
(
)
(
)
()( )
nn
n
kwkwxx
OdomXLIST
n
,...,,...,
:
11
a
where
X = dom(kw) ×
generates the list of keywords without performing
any aggregation and removes the keyword dis-
tance.
In this section we define the aggregation function
for domain keyword measures.
3.1 Keyword Aggregation Function
The aggregation function AVG_KW is designed to
aggregate sets of keywords. Given a set of keywords
as input, the function generates a new set of aggre-
gated keywords. The aggregation process uses the
domain ontology defined in the conceptual model
(ontology and document sources are supposed to be
from the same domain). For each pair of keywords,
the function finds the corresponding least common
ancestor (LCA). But, when aggregating very distant
keywords, no matter how deep the ontology is, there
is a high probability of systematically returning the
root keyword of the ontology. To avoid this, a limit
within the aggregation process must be specified.
Indeed, the further keywords are from one another,
the more sense is lost during aggregation process. In
order to overcome this problem, the function uses a
maximum authorized distance when aggregating
keywords: D
MAX
. So far heuristics suggest a distance
of 3 or 4 nodes and a domain ontology as deep as
possible. So far, the ontology research field has not
solved this problem.
To display results, we use a bi-dimensional table
displaying a fact and two dimensions (Gyssens and
Lakshmanan, 1997), (Ravat et al., 2006). For each
combination of analysis axis values, the table con-
tains a cell. AVG_KW takes as input the content of
these cells (sets of keywords) and produces a new
set as output. The new set is composed of aggre-
gated keywords and/or keywords from the original
cell if aggregation failed due to excessive distances
between the keywords.
Definition 9. We define the aggregation func-
tion:
()( )
n, m,...,yy,...,xx
XAVG_KW: X
mn
mn
11
a
X=
dom(kw) ×
Input: (x
1
,…,x
n
)X
n
is an ordered set of key-
words such that x
i
X, x
j
X | i<j, d(x
i
,
x
ROOT
) d(x
j
, x
ROOT
) (i.e. the furthest nodes from
the root are first) and x
i
=(kw
i
,d
i
) with
kw
i
dom(O) and d
i
D
MAX
.
Output: (y
1
,…,y
m
)X
m
is a set of aggregated
keywords.
Output is generated using the following func-
tion:
Definition 10.
()
()
(
)
(
)
(
)
=
MAXjiLCALCA
ji
ji
D,xxlxxlkwx
,xx
,xx
21
if
otherwise
,,
a
where:
(
)
(
)
()
jiLCA
jijiji
kwkwLCAkw
ddkwkwd,xxl
,
,
=
+
+
=
If x
i
and x
j
are aggregated into x
LCA
then x
i
and x
j
are removed from the input set X and x
LCA
is
added to X. The aggregation process is iterated
on X until no more aggregation may be per-
formed:
(x
i
, x
j
)
X
2
, x
LCA
| l(x
i
,x
j
)
D
MAX
Notice that for a given y
k
of X, if d
k
=0, then the
corresponding keyword kw
k
was not aggregated
during the process and x
i
X | x
i
=y
k
. Notice also
that if x
i
, x
j
X, l(x
i
, x
j
)>D
MAX
, then there is no
aggregation possible and (y
1
,…,y
m
) = (x
1
,…,x
n
) with
m=n.
3.2 Algorithm
The algorithm takes as input a list of keywords to
be aggregated: KW_LIST={kw
1
, kw
2
,…, kw
n
} and
an ontology O. It produces as output an aggregated
keyword list: Output_List. d(keyword
1
,
keyword
2
) is function that computes the distance
between both keywords. Order_List is a func-
tion that orders a list of keywords such that d(kw
i
,
kw
ROOT
) d(kw
j
, kw
ROOT
). That is, keywords are or-
dered by the level they may be found in O, starting
by the lowest levels, i.e. the keywords furthest from
the root. LCA is a function finding the least com-
mon ancestor of a pair of nodes in a tree. See (Harel
and Tarjan, 1984) and more recently (Bender and
Farach-Colton, 2000) for discussion and implemen-
tation of the LCA problem.
{KW_List = OrderList(KW_List,O);
ICEIS 2007 - International Conference on Enterprise Information Systems
154
For each KW
i
of KW_LIST Do
l
i
= 0;
For each KW
j
of KW_List, (j>i) Do
KW
LCA
=LCA(KW
i
,KW
j
) ;
l
LCA
=MAX(d(KW
i
,KW
LCA
),d(KW
j
,KW
LCA
))+l
i
If ( l
LCA
D
MAX
) Then
KW_List=KW_List-{KW
i
, KW
j
};
KW
i
=KW
LCA
; l
i
=l
LCA
;
end_If;
end_For;
Add KW
i
to Output_List;
end_For;}
3.3 AVG_KW Example
The use of drilling operations makes intensive use of
aggregations. Thus, for this example, we shall use
the Roll-Up operation presented in introduction and
the conceptual schema displayed in Figure 2.
Table 1
presents a sample dataset of three documents. Two
keywords have been extracted from each document.
Table 1: A sample dataset of three documents.
In Figure 4, the positions of the different key-
words of the previous table are pointed out by
rectangles in the ontology. Arrows show possible
aggregation process (with distances between nodes
specified). Here, D
MAX
=3.
Figure 4: The position of the different keywords in the
domain ontology O_IS (only partly represented).
In Figure 5-(a), the decision-maker analyses the
publications of author Au_1 during 2004 displaying
results by months. The keywords of the two publica-
tions in September are aggregated: Fact Table,
Conceptual Model and Logical are aggregated into
Design with a distance of 3. Data Warehouse is too
far away from the lowest keyword of the set that
generated Design, thus it is not aggregated:
d(Fact Table, Data Warehouse) = 4 > D
MAX
.
In the document from November, the keyword Al-
gebra is also too far from Document Warehouse,
thus they are not aggregated either.
In Figure 5-(b), to get a more general view, the
analyst “rolls-up” the analysis to a more general
level of detail. Instead of observing results by
months he will analyse them by year. Thus the sys-
tem aggregates the two sets of keywords of the
table (a) into a unique set in table (b). The keyword
Data Warehouse has a distance of 1 with Document
Warehouse and thus will be aggregated into Storage
but Design has already a distance of 3 (D
MAX
), thus,
as Algebra these keywords are too far and are not
aggregated. The resulting cell in the mTable is:
AVG_KW((Data Warehouse, 0), (Design, 3),
(Document Warehouse, 0), (Algebra, 0)) = ((Stor-
age, 1), (Design, 3), (Algebra, 0)).
Figure 5: Analysis of keywords by months (a) and Roll-
Up operation from TIME.Month to TIME.Year (b).
4 CONCLUSION
Up to now, OLAP systems are based on quantita-
tive analysis with the use of numeric measures. As a
first step towards multidimensional OLAP textual
analysis, we presented in this paper a framework for
the use of textual measures. In order to focus the
analysis on textual data, textual measures were
added to traditional multidimensional modelling.
These measures allow the specification of elabo-
rated textual measures such as keyword. We
provide an aggregation function used during opera-
tions of the analysis process (such as drilling
operations). This aggregation function aggregates
keywords into more general ones with the use of a
light domain ontology.
We are currently implementing our approach on
top of an existing OLAP analysis tool: Graphic
OlapSQL. This tool is based on a ROLAP data
warehouse held in an Oracle 10g RDBMS. The tool
is a Java 5 client composed of a hundred classes.
We intend to continue our researches on several
fields. The use of a light ontology (hierarchy of
OLAP AGGREGATION FUNCTION FOR TEXTUAL DATA WAREHOUSE
155
concepts) as a domain ontology is simplistic. The
idea would be to use an ontology with greater ex-
pressive power to be closer to domain semantics and
concepts. Thus further studies should be conducted
on the desirable ontology characteristics. Most end-
user reporting tools display results with a tabular
display such as the one used in this paper. This
graphic interface is far from being adapted to display
loads of keywords or textual data. Future efforts
should also be oriented on a new display with a
greater expressive power. Finally, keyword meas-
ures are part of a greater family of textual measures:
elaborated textual measures, we intend to focus on a
more general framework for all types of textual
measures.
REFERENCES
Agrawal R., Bayardo R.J., Srikant R., 2000. “Athena:
Mining-based Interactive Management of Text Data-
bases”, 7
th
Int. Conf. on Extending Database
Technology (EDBT 2000), LNCS 1777, Springer, pp.
365-379.
Bender M.A., Farach-Colton M., 2000. “The LCA Prob-
lem Revisited”, 4
th
Latin American Symposium on
Theoretical Informatics (LATIN 2000), LNCS 1776,
Springer-Verlag, pp. 88-94.
Chakrabarti S., Dom B., Agrawal R., Raghavan P., 1998.
“Scalable Feature Selection, Classification and Signa-
ture Generation for Organizing Large Text Databases
into Hierarchical Topic Taxonomies”, in The VLDB
Journal, vol.7(3), pp. 163-178.
Golfarelli M., Maio D., Rizzi S., 1998. “The Dimensional
Fact Model: a Conceptual Model for Data Ware-
houses”, in Int. Journal of Cooperative Information
Systems, vol. 7, n. 2&3.
Gyssen M., Lakshmanan L.V.S., 1997. “A Foundation for
Multi-Dimensional Databases”, in 23
rd
Int. Conf. on
Very Large Data Bases (VLDB 1997), pp. 106–115.
Harel D., Tarjan R.E., 1984. “Fast algorithms for finding
nearest common ancestors”, in SIAM Journal on
Computing Archive, vol.13(2), pp. 338-355.
Horner J., Song I-Y., Chen P.P., 2004. “An analysis of
additivity in OLAP systems”, ACM 7
th
Int. Workshop
on Data Warehousing and OLAP (DOLAP 2004),
ACM, pp. 83-91, 2004.
Keith S., Kaser O., Lemire D., 2005. “Analyzing Large
Collections of Electronic Text Using OLAP”, 29
th
Conf. Atlantic Provinces Council on the Sciences
(APICS 2005), Wolville, Canada.
Khrouf K., Soulé-Dupuy C., 2004. “A Textual Warehouse
Approach: A Web Data Repository”, in Intelligent
Agents for Data Mining and Information Retrieval,
Masoud Mohammadian (Ed.), Idea Publishing Group,
pp. 101-124.
Kimball R., 1996. “The data warehouse toolkit”, Ed. John
Wiley and Sons, 2
nd
ed. 2003.
Lassila O., McGuinness D.L., 2001. “The Role of Frame-
Based Representation on the Semantic Web”, Knowl-
edge Systems Laboratory Report KSL-01-02, Stanford
University. (Also appeared in Computer and Informa-
tion Science, Vol.6(5), Linköping University, 2001).
McCabe C., Lee J., Chowdhury A., Grossman D. A.,
Frieder O., 2000. “On the design and evaluation of a
multi-dimensional approach to information retrieval”,
23
rd
Annual Int. ACM Conf. on Research and Devel-
opment in Information Retrieval (SIGIR 2000), ACM,
pp. 363-365.
Mothe J., Chrisment C., Dousset B., Alau J., 2003. “Doc-
Cube: Multi-dimensional visualisation and
exploration of large document sets”, in Journal of the
American Society for Information Science and Tech-
nology (JASIST), vol.54(7), pp. 650-659.
Niemi T., Niinimäki M., Nummenmaa J., Thanisch P.,
2002. “Constructing an OLAP cube from distributed
XML data”, 5
th
ACM Int. Workshop on Data Ware-
housing and OLAP (DOLAP 2002), ACM, pp.22-27.
Park B-K., Han H., Song I-Y., 2005. “XML-OLAP: A
Multidimensional Analysis Framework for XML
Warehouses”, 6
th
Int. Conf. on Data Warehousing
and Knowledge Discovery (DaWaK 2005), LNCS
3589, Springer, pp.32-42.
Pérez J.M., Llavori R.B., Aramburu M.J., Pedersen T.B.,
2005. “A relevance-extended multi-dimensional
model for a data warehouse contextualized with
documents”, 8
th
ACM Int. Workshop on Data Ware-
housing and OLAP (DOLAP 2005), ACM, pp.19-28.
Pokorný J., 2001. “Modelling Stars Using XML”, in
Proc. 4
th
ACM Int. Workshop on Data Warehousing
and OLAP (DOLAP 2001), pp.24-31.
Ravat F., Teste O., Zurfluh G., 2006. “Constraint-Based
Multi-Dimensional Databases”, Chapter XI in Data-
base Modeling for Industrial Data Management,
Zongmin Ma (ed.), IDEA Group, pp.323-368.
Sullivan D., 2001. Document Warehousing and Text Min-
ing, Wiley John & Sons.
Torlone R., 2003. “Conceptual Multidimensional Mod-
els”, Chapter III in Multidimensional Databases:
Problems and Solutions, M. Rafanelli (ed.), Idea
Group, pp.69-90.
Tseng F.S.C., Chou A.Y.H, 2006. “The concept of docu-
ment warehousing for multi-dimensional modeling of
textual-based business intelligence”, in journal of De-
cision Support Systems (DSS), vol.42(2), Elsevier, pp.
727-744.
Vrdoljak B., Banek M., Rizzi S., 2003. “Designing Web
Warehouses from XML Schemas”, 5
th
Int. Conf. on
Data Warehousing and Knowledge Discovery
(DaWaK 2003), LNCS 2737, Springer, pp.89-98.
Yin X., Pedersen T.B., 2004. “Evaluating XML-extended
OLAP queries based on a physical algebra”, 7
th
ACM
Int. Workshop on Data Warehousing and OLAP
(DOLAP 2004), ACM, pp.73-82.
Zhang J., Ling T.W., Bruckner R.M., Tjoa A.M., 2003.
“Building XML Data Warehouse Based on Frequent
Patterns in User Queries”, 5
th
Int. Conf. on Data
Warehousing and Knowledge Discovery (DaWaK
2003), LNCS 2737, Springer, pp.99-108.
ICEIS 2007 - International Conference on Enterprise Information Systems
156