Towards Data Warehouse Schema Design from Social Networks

Dynamic Discovery of Multidimensional Concepts

Rania Yangui

, Ahlem Nabli

and Faiez Gargouri

Institute of Computer Science and Multimedia, Sfax University, BP 1030, Sfax, Tunisia

Faculty of Sciences, Sfax University, BP 1171, Sfax, Tunisia

Keywords:

Data Warehouse, Social Network, Ontology, Flexibility, Scalability, Dynamicity, Clustering.

Abstract:

This research work is conducting as part of the project BWEC (Business for Women in Women of Emerging

Country) that aims to improve the socio-economic situation of handicraft women by providing true techno-

logical means. In fact, since few years, the Web has been transformed into an exchange platform where users

have become the main suppliers of information through social media. User-generated data are usually rich

and thus need to be analyzed to enhance decision. The storage and the centralization of these data in a data

warehouse (DW) are highly required. Nevertheless, the growing complexity and volumes of the data to be

analyzed impose new requirements on DW.

In order to address these issues, in this paper, we propose four stages methodology to deﬁne a DW schema

from social networks. Firstly we design the initial DW schema based on the existing approaches. Secondly,

we apply a set of transformation rules to prepare the creation of the NOSQL(Not Only SQL) data warehouse.

Then, based on user’s requirement, clustering of social networks proﬁling data will be performed which allows

the dynamic discovery of multidimensional concepts. Finally, the enrichment of the NoSQL DW schema by

the discovered MC will be realized to ensure the DW schema evolution.

1 INTRODUCTION

Because of the rapid development of the Internet,

the availability of various types of data has increased

tremendously. In fact, the creation of many sites (such

as Facebook, LinkedIn, Twitter, etc.) and forums

has made the users perceive the Web as a place in

which they exchange ideas, opinions as well as con-

tents. However, if these tools make the sharing and

collaboration between the users easy, they may cause

new challenges concerning the relevant exploitation

of the produced data. Analyzing, understanding, as

well as managing the huge volumes of complex data

produced from the social networks (SN) broach a

paramount importance and draw the attention of many

researchers. In fact, the companies expect to acquire

important information from this data so as to improve

their marketing. This is the case of handicraft women

in the BWEC

project. This latter aims at improving

the socio-economic situation of these women by pro-

viding true technological means in concordance with

Towards a new Manner to use Affordable Technologies

and Social Networks to Improve Business for Women in

Emerging Countries (http://projetat.cerist.dz/)

the women habits and the technical context of their

countries. The stages proposed to accomplish the pur-

pose of this project are summarized in (Figure 1).

The use of online SN can play a very important

role in the social and economic development of this

population. For instance, SN can be used not only

to promote their products but also to enhance the

brand value. It can also be used to strengthen con-

sumer relations as well as to improve the quality of

the services and products through receiving feedback

from the market itself. Thus, the establishment of a

decision-making process has proved to be necessary.

Originally developed for the needs of support de-

cision, data warehouses (DW) have proven to be an

adequate solution to a variety of applications and

ﬁelds. DW contains all the information integrated

from heterogeneous sources into multidimensional

schema to enhance data access for both analysis and

decision making.

Many methodologies can be used to create a

DW (Nabli, 2013) which are demand-driven method-

ologies, data-driven methodologies and Mixed (de-

mand/data driven) methodologies. These warehous-

ing methodologies have shown their efﬁciency when

338

Yangui R., Nabli A. and Gargouri F..

Towards Data Warehouse Schema Design from Social Networks - Dynamic Discovery of Multidimensional Concepts.

DOI: 10.5220/0005383903380345

In Proceedings of the 17th International Conference on Enterprise Information Systems (ICEIS-2015), pages 338-345

ISBN: 978-989-758-096-3

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

Figure 1: Project Steps.

dealing with structured data. Its require measures

and dimensions of a DW to be known at the design

stage (N. U. Rehman and Scholl, 2012). However,

the growing complexity and volumes of the data to be

analyzed impose new requirements on data warehous-

ing systems.

As argued by many works (N. U. Rehman and

Scholl, 2012) (E. Gallinucci and Rizzi, 2013), exist-

ing warehousing methodologies cannot be success-

fully applied to handle the above-mentioned chal-

lenges. So, a remarkable effort must be made to inte-

grate the huge amount of complex data from SN and

make them accessible to OLAP (On-Line Analytical

Processing) and analyzing tools.

Our challenges can be summarized by the follow-

ing questions: how to determine multidimensional

concepts (MC) from social data? How to adjust the

DW to take into account these new concepts? How to

overcome the problems caused by the heterogeneity

of the data? How to take into account the dynamicity

of user-generated data as well as the users needs in an

existing DW?

This paper tends to answer the mentioned ques-

tions by proposing a methodology to design a DW

schema via content-based dynamic discovery of MC

from social networks generated data. The proposed

methodology will be applied on real case study. This

case study looks at handicraft women social networks.

Our contributions focus on the DW schema evolv-

ing to manage data about handicraft women from SN.

In this context, DW schema modeling is a complex

task which involves knowledge of SN structure and

familiarity with DW technologies. What makes this

task even more challenging is the fact that social data

continue to grow rapidly and analytical requirements

change over time. Given these challenges, the semi-

automatic modeling of the warehouse schema is re-

quired.

This paper is organized as follows. Section 2 re-

views some related works concerning the DW schema

creation from SN and the semi-automatic generation

of DW schema. Section 3 overviews our method-

ology. Sections 4 describes the initial DW schema

creation. Sections 5 details the dynamic discovery of

MC. Section 6 concludes the paper and draws future

research directions.

2 RELATED WORKS

Data generated from SN are usually rich and need to

be analyzed to support decision. The storage and the

centralization of these data in a DW are highly re-

quired. However as mentioned above, warehousing

methods have shown some limitations when dealing

with SN data. Challenged by these limitations, DW

researchers put tremendous efforts into extending its.

2.1 Multidimensional Modeling from

Social Networks

In the literature, many researchers have proposed ap-

proaches for semi-automatic modeling of DW.

For example, a data warehousing architecture for

analyzing large data sets at Facebook, used for friend

recommendation, is described by (A. Thusoo and Liu,

2010). The authors describe the challenges of im-

plementing a DW for data intensive Facebook appli-

cations and present a number of contributed open-

source technologies for warehousing petabytes of

data. These include Scribe

, Hadoop

and Hive

which together form the cornerstones of the log col-

lection, storage and analytic infrastructure at Face-

book.

The paper mainly focuses on ﬂexibility and scal-

ability issues. However no insight on the underlying

models is given and no dynamic discovery of MC is

mentioned.

(P. Kazienko and Brdka, 2011) focus on develop-

ing a conceptual generic model for multidimensional

SN that allows capturing information about different

types of activities and interactions between users. It

also represents the dynamics of users behavior. The

proposed model encompasses information about the

different relations and groups that exist within a given

relation layer and in a speciﬁc time window. The pro-

posed model covers three main dimensions: relation

layers, time windows and groups. Social groups are

extracted by means of clustering methods. However,

http://wiki.github.com/facebook/scribe

http://wiki.apache.org/hadoop

http://wiki.apache.org/hadoop/Hive

TowardsDataWarehouseSchemaDesignfromSocialNetworks-DynamicDiscoveryofMultidimensionalConcepts

339

the proposed model doesnt manage the ﬂexibility and

the scalability of social networks data.

(N. U. Rehman and Scholl, 2012) provide a DW

solution for hosting the public data stream of Twit-

ter messaging. The authors enrich the multidimen-

sional analysis of such data via content-driven discov-

ery of dimensions and classifying hierarchies. In the

ﬁrst step, data mining algorithms are applied to clus-

ter dimensional data. In the second step, the acquired

classiﬁcation is added as a new aggregation path to

the respective dimension, leading to the third step of

enabling this new aggregation path in OLAP queries.

Nevertheless, this work is limited to the granularity

level addition and ignores the other MC such as facts

and dimensions. Moreover, the proposed model is in-

ﬂexible and no scalable.

(E. Gallinucci and Rizzi, 2013) propose a method-

ology called meta-stars to model topic hierarchies

in ROLAP systems. Its basic idea is to use meta-

modeling coupled with navigation table and with tra-

ditional dimension tables. The navigation tables sup-

port hierarchy instances with different lengths and

with non-leaf facts, and allow different roll-up seman-

tics to be explicitly annotated. The meta-modeling

enables hierarchy heterogeneity and dynamics to be

accommodated. However, this work is based on a re-

lational approach which presents limitations regard-

ing to schema scalability.

(Moalla and Nabli, 2014) present a method to

multidimensional schema construction from unstruc-

tured data extracted from SN. This construction is car-

ried out from Facebook page in order to analyze the

customers opinions. A real case study has been devel-

oped to illustrate the proposed method and to conﬁrm

that the SN analysis can predict the success prospects

of the products. Nevertheless, the dynamic discovery

of MC is not supported. The proposed model is not

ﬂexible and not adaptable to the huge amount of so-

cial data.

Based on the previous study, most of the works

show no indication of the dynamic determination of

MC seen the velocity of SN data. Also the DW

schemas are generally ﬁxed at design stage.

2.2 Dynamic Discovery of

Multidimensional Concepts

Nowadays, we are experiencing a rapid growth of so-

cial structures supported by communication technolo-

gies and various Web-based services. Due to scale,

complexity and dynamicity, user-generated data from

SN are very difﬁcult to store and analyze in terms of

traditional data warehousing methods (N. U. Rehman

and Scholl, 2012). To overcome these problems,

many authors have worked on dynamic discovery of

MC and have used data mining to build a DW.

In this context, (Usman and Pears, 2011) provide

a methodology to design semi-automatically DWs

schema with hierarchical clustering. This latter is

used to perform a pre-processing on the data. After

that, the system identiﬁes both facts and dimensions

into the clustered data.

Rehman proposes a system to dynamically build

hierarchies based on data from Twitter (N. U. Rehman

and Scholl, 2012). This paper has two Interests: a)

The cube is built on original data which are the mes-

sages of users on a SN. b) Data mining is used to

dynamically build hierarchies. Thanks to data min-

ing, the categories of network users described in hi-

erarchies are updated automatically. On the other

hand, Ceci uses a hierarchical clustering to integrate

continuous variables as dimensions in a DW schema

(M. Ceci and Malerba, 2011). It discretizes a continu-

ous dimension so that the user can perform operations

on existing querying a cube: Roll-up and DrillDown.

As for the current work, (L. Sautot and Molin,

2014) propose using hierarchical agglomerative clus-

tering with a metric that comes from ecological stud-

ies to build semi-automatically hierarchical dimen-

sions in an OLAP cube. The authors perform a hierar-

chical clustering on heterogeneous data sets that con-

tains qualitative and quantitative variables. They offer

a prototypical automatic system which builds dimen-

sion for an OLAP cube and measure the performances

of this system according to the number of clustered

individuals and according to the number of variables

used for clustering.

Table 1 highlights a summary of the literature re-

view which is based on seven criteria (Concept M.:

Conceptual Model, D. MC: multidimensional con-

cepts, Methodology, SN: Social Network, Ontology,

Flexibility, Scalability).

All the mentioned works present several interest-

ing mining. It has been recognized that mining tech-

niques such as Clustering can help in designing DW

schema. That is why we adopt this orientation for

the dynamic discovery of MC. However, no work has

ever dealt with the semantic heterogeneity. More-

over, no work has ever followed a mixed approach

(data/demand driven approach). Furthermore, it is

worth noting that just one work has provided the scal-

ability of the schema. At the same time, the hetero-

geneity and the growth of the social data need to be

considered in order to properly retrieve needed data.

The frequent arrival of new needs requires that the

system should be adaptable to changes.

Based on the above discussion, there is a strong

need of a signiﬁcant methodology that allows a dy-

ICEIS2015-17thInternationalConferenceonEnterpriseInformationSystems

340

Table 1: Summary of the literature review.

Work Concept M. D. MC Methodology SN Ontology Flexibility Scalability

(Thusoo, 2010) - No Data driven Yes No Yes Yes

(Ceci, 2011) - Yes Demand driven No No No No

(Usman, 2011) - Yes Data driven No No No No

(Kazienko, 2011) - Yes Data driven Yes No No No

(Rehman, 2012) x-DFM Yes Data driven Yes No No No

(Gallinucci, 2013) Meta-Star Yes Data driven Yes No No No

(Moalla, 2014) Star Schema No Mixed Yes No No No

(Sautot, 2014) Constellation Yes Data driven No No No No

Our Proposition x-DFM Yes Mixed Yes Yes Yes Yes

namic discovery of MC based on ontology. The mod-

eling of a ﬂexible and scalable DW schema is also

required to deal with data from SN.

3 OVERVIEW OF THE

PROPOSED METHODOLOGY

In this paper, we propose four stages to deﬁne data

warehouse methodology from social networks (Fig-

ure 2). We will begin with the creation of the ini-

tial DW schema from structured and heterogeneous

sources following a classical approach. The second

stage will relate to the transformation of the DW

schema into a NoSQL Data Base. After that, we will

dynamically determine the MC, their types as well

as their locations. These MC are used to enrich the

NoSQL DW schema.

Our methodology takes advantages of the matu-

rity of existing design approaches, the scalability of

NOSQL Data Base and the capability of the dynamic

discovery of multidimensional concepts through clus-

tering techniques. Figure 2 depicts our proposed

methodology.

1. Initial DW schema creation: involves the creation

of the initial DW schema following a classical

mixed approach;

2. NOSQL DW schema creation: it consists on the

generation of NOSQL data base for the initial data

warehouse schema based on a rules set. These

rules allow the transformation of a DW schema

concepts to speciﬁc concepts of NOSQL data

base;

3. Discovery of multidimensional concepts: consists

on deﬁning the features set that meet the users

needs, generating clusters from social networks

based on the deﬁned features and then determin-

ing the MC, their types as well as their locations;

4. NoSQL DW schema evolving: using the discov-

ered MC to enrich the NoSQL DW schema.

Figure 2: Overview of the proposed methodology.

In the following we expose the initial data ware-

house creation, then we detail the discovery of multi-

dimensional concepts using clustering technique.

4 INITIAL DW SCHEMA

CREATION: A DW FOR CRAFT

PRODUCTION ANALYSIS

In our research project, we need to create a DW to an-

alyze the impact of using IT solutions on the situation

of handicraft women. To accomplish this stage, we

have data sources (both internal and external) avail-

able in the project. They are the following:

• S1: Data Base of the National Ofﬁce for Tunisian

Handicrafts;

• S2: Ontology deﬁned from interviews;

• S3: Tunisians and Algerian postcodes data bases.

In our case study, ontology (S2) represents an

internal source which is an essential component to

TowardsDataWarehouseSchemaDesignfromSocialNetworks-DynamicDiscoveryofMultidimensionalConcepts

341

evolve a DW schema. It contains both data and meta-

data. This source is a consistent support in the dis-

covery of dynamic multidimensional concepts. The

used ontology presents knowledge about the proﬁle

of handicraft women. Similarly, the data Base of the

National Ofﬁce for Tunisian Handicrafts (ONAT) is

an internal source. Otherwise, the Tunisian and Alge-

rian postcodes data bases s an external source which

is important to enrich the craft DW. The craft DW is

built from these three sources (S1, S2 and S3) using

the ETL (Extract Transform Load) process based on

data driven approach using Talend Open Source

. An

overall view of the initial craft DW schema is depicted

in Figure 3.

Figure 3: Data warehouse for craft production analysis.

The structure of the initial DW schema is a graph

centered at two facts. First, the Production fact: in-

cludes (Quantity, Pieces Number, Prod Duration)

as measures and is linked to (Artisan, Produc-

tion

Tool, Process, Raw Material, Product and

Date) as dimensions. Second, the Selling fact which

includes (Quantity, Amount) as measures and is

linked to (Artisan, Product and Date) as dimensions.

In this stage, we have followed a traditional approach

for DW schema design which is not suitable for the

latest data source (S4). Therefor, we propose to use

Clustering method.

5 DYNAMIC DISCOVERY OF

MULTIDIMENSIONAL

CONCEPTS

Classical decision support is used in analyzing sim-

http://www.talend.com/

ple data. However, these systems are not adapted

for SN analysis which highlights the need of creat-

ing new models. In fact, data are heterogeneous and

changeable over time. Thus, a comprehensive schema

for craft DW cannot be ﬁxed at the time of design

and must be dynamically modiﬁed. In a nutshell,

a DW schema can be extended by adding new ele-

ments of type measure, dimension, or hierarchy level.

These extracted values are to be fed into the semiauto-

matic schema evolving to dynamically model the DW

schema. This stage is divided into three main steps:

features deﬁnition, clusters generation and multidi-

mensional concepts determination.

5.1 Features Deﬁnition

In DW lifecycle, user requirements deﬁnition is one

of the most important tasks which ensure a successful

DW project. The main objective is to identify analyst

goals in order to reduce the risk of failure. Since the

expert has prior knowledge of the analysis goal, our

methodology allows to evolve a DW schema under

the designers guidance. At ﬁrst, the expert is required

to select the multidimensional concepts that should

be dynamically evolved according to his objectives.

He is then required to select one or a set of features

that meet his needs. These features are the basis for

grouping objects in the next step.

Example. To analyze the business of handicraft

women, we should cluster Products values based on

the Product Designation and then based on the used

Raw Material and the used Production Tool values.

In fact, if two products have the same designation,

they belong, therefore, to the same Group Activity.

Otherwise, if they use a set of Raw Material and Pro-

duction Tool in common, they probably belong to the

same Group Activity. Consequently, we derive three

features which are respectively Product Designation,

Raw Material and Production Tool.

5.2 Clusters Generation

The hierarchical clustering technique is applied to the

data set to generate clusters based on a similarity mea-

sure. As most of the clustering algorithms are un-

supervised, in this step, we target the semi super-

vised hierarchical clustering in order to get the op-

timal results that meet the analysts needs. To do

that, we have used the SHICARO (Semi-supervised

Hierarchical Clustering based on Ranking features

using Ontology) (R. Yangui and Gargouri, 2014a)

method with a proﬁling ontology. This method con-

sists of two important components. The ﬁrst one

consists in deﬁning effective ontology-based similar-

ICEIS2015-17thInternationalConferenceonEnterpriseInformationSystems

342

Figure 4: SHICARO method.

ity measures that combine both numerical and nom-

inal variables along different dimensions (instances,

attributes, and relation-ships) (R. Yangui and Gar-

gouri, 2014b), while the second consists in providing

a performable clustering algorithm based on ranking

features. SHICARO method aims to cluster objects

based on scheduled features.

Since the expert knows the goal behind which the

clustering is performed, SHICARO method performs

clustering under the users guidance. Thus, the user

is required to order the features from the highest to

the lowest ones. In each iteration, a set of features

F =

{

( f

, r

), . . . , ( f

, r

)

}

that have the same rank r

are applied to cluster objects. SHICARO steps are

depicted by Figure 4.

Example. By applying SHICARO method based

on the set of features F=(Product Designation,

1), (Raw Material, 2), (Production Tool, 2), we

obtained at the ﬁrst iteration (based on Prod-

uct Designation feature) ﬁves clusters and at the sec-

ond iteration (based on Raw Material and Produc-

tion Tool features) two clusters (Figure 5).

Figure 5: Clustering Product instances.

The extracted values, present in each of the gener-

ated clusters, become the input to the dynamic schema

evolving in the next step.

5.3 Multidimensional Concepts

Determination

This step is performed by the designer. It consists in

analyzing generated clusters, determining the type of

multidimensional concept, assigning names to clus-

ters and specifying the location of insertion in the

multidimensional schema.

Example1. Adding the Group Activity level to the

Product dimension hierarchy (Figure 6).

Figure 6: Adding the Group Activity level to the Product

dimension hierarchy.

In Figure 6.a, we have two clusters: the ﬁrst one

describes the Traditional Weaving and the second

one describes the Traditional Clothing. So we can

deﬁne the concept Group Activity as MC. This MC is

considered as parameter of the Product dimension as

depicted in Figure 6.b.

Example2. Adding the Customer, Supplier and Fan

dimensions (Figure 7) Artisan friends in SN can be

divided into groups based on their job, their link and

the exchanged clips. Feature set used to cluster wom-

ens friends is F= (link, 1), (job,2), Clip,3). Clustering

algorithm based on F determine three groups named

Customer, Supplier and Fan.

TowardsDataWarehouseSchemaDesignfromSocialNetworks-DynamicDiscoveryofMultidimensionalConcepts

343

Figure 8: DW schema enriched by discovered MC.

Figure 7: Adding the Customer, Supplier and Fan dimen-

sions.

Each group represents a CM, so we have three

new MC: Customer, Supplier and FAN. These MC

are considered as three new dimensions. Customer is

added as dimensions to the Selling fact. Supplier and

FAN are added as dimensions to the Production fact

as depicted in ﬁgure 7.

Figure 8 shows the DW schema enrichment for

storing various cumulative data about handicraft

women derived from SN in the the extended Dimen-

sional Fact Model (xDFM) (Mansmann, 2008). Di-

mensions are modeled as aggregation paths. All paths

of a dimension converge in an abstract Tnode, which

corresponds to the aggregated value all. A level node

in a dimension consists of at least one key attribute,

but may include further attributes represented as un-

derlined terminal nodes. As shown in Figure 8, three

new dimensions are added to the initial DW schemas

which are Customer, Supplier and Fan. These di-

mensions are discovered by performing the hierar-

chical clustering based on a set of features. Cus-

tomer, Supplier and Fan reﬂect the communities to

which handicraft women are connected at SN. How-

ever, an Artisan can be also a Supplier, a Customer

or a Fan. That is why a generalization link is deﬁned

Tperson. Similarly, Raw Mat Type, Activity Group

and Tool Type proprieties are added as levels succes-

sively in the Raw Material, Product and Fan dimen-

sion hierarchies. These properties are deducted us-

ing the hierarchical clustering based on the appro-

priate features. However, a Product can also be a

Raw Material. Hence, a generalization link is added

Tmaterial.

6 CONCLUSION

In this paper, we proposed four stage methodology

to deﬁne a data warehouse schema from social net-

works. Starting in a ﬁrst stage, by the design of the

initial data warehouse schema based on the existing

approaches. In a second stage, we have generated a

NOSQL Data warehouse schema by applying a set of

transformation rules. Then, based on users require-

ICEIS2015-17thInternationalConferenceonEnterpriseInformationSystems

344

ment, clustering of social networks proﬁling data is

performed which allows the dynamic discovery of

multidimensional concepts. Finally, the enrichment

of the NoSQL data warehouse schema by the discov-

ered MC is realized to ensure the DW schema evolu-

tion.

We have especially detailed the dynamic content-

based discovery of dimensions, hierarchies and mea-

sures using hierarchical clustering. This latter, is per-

formed using proﬁling ontology with adequate simi-

larity measures. The detailed stages are experimented

on the real case study of the BWEC project.

We are currently studying the NOSQL data bases,

we intend to deﬁne transformation rules from a con-

ceptual data warehouse schema to NOSQL database

and the evolution rules. Moreover, we think it would

be interesting to formally specifying transformation

rules to allow the automatic schema generation.

ACKNOWLEDGEMENT

We are very thankful to the Algerian Tunisian Project

dealing with the improvement of handicraft women

business in emerging countries through affordable

technologies and social networks. This project is ﬁ-

nanced by the Tunisian Ministry of Higher Education,

Scientiﬁc Research and Information and Communi-

cation Technologies Higher Education and Scientiﬁc

Research sector.

REFERENCES

A. Thusoo, Z. Shao, S. A. D. B. N. J. J. S. S. R. M. and Liu,

H. (2010). Data warehousing and analytics infrastruc-

ture at facebook. In International conference on Man-

agement of data SIGMOD’10, pages 1013–1020.

E. Gallinucci, M. Golfarelli, A. W. and Rizzi, S. (2013).

Meta-stars: Multidimensional modeling for social

business intelligence. In International Workshop On

Data Warehousing and OLAP DOLAP13, pages 11–

18.

L. Sautot, B. Faivre, L. J. and Molin, P. (2014). The hi-

erarchical agglomerative clustering with gower index:

A methodology for automatic design of olap cube in

ecological data processing context. In Ecological In-

formatics, pages 1–14.

M. Ceci, A. and Malerba, D. (2011). Olap over contin-

uous domains via densitybased hierarchical cluster-

ing. In International Conference on Knowledge-Based

and Intelligent Information and Engineering Systems

KES’11, pages 559–570.

Mansmann, S. (2008). Extending the olap technology

to handle nonconventional and complex data. In

Ph.D. dissertation, University Konstanz, Department

of Computer and Information Science Germany.

Moalla, I. and Nabli, A. (2014). Towards data mart build-

ing from social network for opinion analysis. In Inter-

national Conference on Intelligent Data Engineering

and Automated Learning IDEAL14, pages 295–302.

N. U. Rehman, S. Mansmann, A. W. and Scholl, M. H.

(2012). Building a data warehouse for twitter stream

exploration. In International Conference on Advances

in Social Networks Analysis and Mining ASONAM’12,

pages 1341–1348.

Nabli, A. (2013). Approche d’aide la conception au-

tomatise d’entrept de donnes: Guide de modlisation.

Presses Acadmiques Francophones.

P. Kazienko, K. Musial, E. K. T. K. and Brdka, P. (2011).

Multidimensional social network: Model and analy-

sis. In International Conference on Computational

Collective Intelligence Technologies and Applications

ICCCI’11, pages 378–387.

R. Yangui, A. N. and Gargouri, F. (2014a). Shicaro: Semi-

supervised hierarchical clustering based on ranking

features using ontology. In International Conference

on Management and Technology in Knowledge, Ser-

vice, Tourism & Hospitality SERVE’14, pages 233–

238.

R. Yangui, A. N. and Gargouri, F. (2014b). Soim: Simi-

larity measures on ontology instances based on mixed

features. In 4th International Conference on Model &

Data Engineering MEDI2014, pages 169–176.

Usman, M. and Pears, R. (2011). Multi level mining of

warehouse schema. In NDT, volume 136 of Communi-

cations in Computer and Information Science, pages

395–408.

TowardsDataWarehouseSchemaDesignfromSocialNetworks-DynamicDiscoveryofMultidimensionalConcepts

345