Fuzzy Ontology-based Spatial Data Warehouse for Context-aware

Search and Recommendation

Hajer Baazaoui-Zghal

Riadi laboratory, University of Manouba, Tunis, Tunisia

Keywords:

Fuzzy Ontology, Spatial Data Warehouse, Context-aware Search.

Abstract:

The need to build a spatial data warehouse over all structured and unstructured data is becoming necessary in

many ﬁelds. In addition, contextualized results are considered as key challenges in data warehouses, which

are frequently related to only one context. However, in real life applications, decisional data are shared by

several users having different proﬁles, which makes contextual awareness essential for decision support. In this

paper, we propose a fuzzy Ontology-based Spatial Data Warehouse for contextual search and recommendation,

composed of 3 main layers: (1) Data layer composed of structred, unstructred and users data, (2) Knowledge

and context-aware layer and (3) Online application layer. The originality of the work described here consists

on integration of uncertain data at different levels of the knowledge layer and having in the same decisional

architecture contextual search and recommendation.

1 INTRODUCTION

Decision-makers become more and more exigent

face to the exponential growth of storage and large

amounts of structured and unstructured data. In this

context, data warehouses have been introduced to

present solutions for storage, mining and exploration.

Indeed, the decision-makers need a knowledge-based

warehouse taking into account the context. Thus,

integrated approaches based on semantics have been

proposed aiming exploration of users knowledge and

data sources. Nevertheless, several limits have been

observed mainly, related to the lack of contextual-

ization and the integration of uncertain data. A lack

of contextual intelligence in case of search in Data

Warehouse (DW) is remarked and a need for genera-

tion of personalized results is driving the usage of it

in context-aware applications. Context consists of all

aspects linked to the user and domain that may affect

the decision process. Actually data warehouse-based

information systems are faced to several challenges,

like considering users contexts and preferences,

and also considering uncertain data and needs. In

fact, vagueness, uncertainty inherently existing in

spatial data and imprecision of feature values are not

supported by such method. Indeed, fuzziﬁcation is

integrated into the case-based reasoning process to

handle such imprecision and uncertainty.

In this paper, we propose an architecture aiming

to assist users during their search for pertinent results

and presenting contextual recommendations. Further-

more, our proposal supports integration of imprecise

knowledge.

Our motivation is that considering input in the pres-

ence of context factors may improve performance and

efﬁciency of the decision process, which is not eas-

ily detectable with classic data warehousing meth-

ods. Indeed, due to nature of data which is both dy-

namic and uncertain, data should be interpreted dif-

ferently depending on current situation (context). Our

second motivation relies on the linguistic ambiguity

problems, which make crisp ontologies less sufﬁcient

when dealing with uncertain knowledge. So, we pro-

pose to integrate fuzzy logic into ontologies and con-

textualization. We bring three main contributions, the

proposal (1) integrates different data sources, (2) con-

siders contextual and uncertain data, (3) presents con-

textual search and recommendations.

The remaining of this paper is organized as follows.

Section 2 describes our proposal. Section 3 presents

an overview of works related and position our pro-

posal. Finally, Section 3 concludes and proposes di-

rections for future research.

Baazaoui-Zghal, H.

Fuzzy Ontology-based Spatial Data Warehouse for Context-aware Search and Recommendation.

DOI: 10.5220/0006010501610166

In Proceedings of the 11th International Joint Conference on Software Technologies (ICSOFT 2016) - Volume 2: ICSOFT-PT, pages 161-166

ISBN: 978-989-758-194-6

161

Figure 1: General architecture.

2 FUZZY ONTOLOGY-BASED

SPATIAL DATA WAREHOUSE

FOR CONTEXT-AWARE

SEARCH AND

RECOMMENDATION

The objective of the work described in this paper is

to propose an architecture that supports both context-

aware search and recommendation based on the ap-

proach of the data warehouses and data marts which

store the relevant data for the decision-makers. We

recall that the main steps of data warehouse design

starting from heterogeneous sources are: conceptual

design, logical design and physical design. The whole

process includes the storage, the analysis and the ex-

ploration within spatial data, with an aim of improv-

ing the process of decision-making. The general ar-

chitecture of our proposal is given by Figure 1, which

shows the workﬂow for returning answers queries and

generating recommendations. It is composed of three

main layers: (1) Data layer composed of structred, un-

structred and users data, (2) Knowledge and context-

aware layer and (3) Online application layer.

2.1 The Data Layer

The data layer allows manipulation of unstructured

data, structured data as well as user’s data which is

Figure 2: Ontology-based Star Schema.

also considered as input in our proposal as. In a pre-

vious work we have proposed a tool for Spatial Data

Marts Design and Generation (Zghal et al., 2003), the

idea is to create a spatial data warehouse by assem-

bling Spatial Data Marts (SDM). In the current work,

from the global sources which can be represented

through traditional models dedicated to the manage-

ment of the spatial data, we propose at ﬁrst to build

the ontology-based Star Schema for SDM building.

We recall that ontology is knowledge structure that

represents the concepts, relationships, instances and

properties for a given domain, which has proven their

usefulness to model information systems based on the

semantic and knowledge level. In the currect work,

we deﬁne an ontology for the extraction and integra-

tion of data (cf. ﬁgure 2). An ontology provides a

conceptual representation of the application domain

(a shared vocabulary). Then, a selection of attributes

from the determined ontological representation and

ICSOFT-PT 2016 - 11th International Conference on Software Paradigm Trends

162

their correspondence to attributes of the deﬁned star

schema in order to populate a speciﬁc instance. The

meta-ontology (cf. Figure 3) allows modeling and

keeping trace about the way the ontology was built;

(3) An ontology SDM is characterized by its multidi-

mensional structure formed by facts, dimensions and

measurements. Dimensions and measurements can be

spatial or nonspatial. We propose to model the hier-

archy dependences as a class called Hierarchy. We

insist here on the importance to keep a history of the

different levels of the hierarchy for dimensions and

measurements given the nature of the spatial data.

The fact corresponds to the topic or subject of analy-

sis and is represented by a class. A measurement can

be numerical when it contains only numerical data

such as the returned monthly one of an area. A mea-

surement can also be spatial: a collection of links to

spatial objects. For example the case of a generaliza-

tion of areas having temperature and precipitations in

the same cell. Thus, measurement forms a collection

of links to the corresponding areas. So, the global

ontology consists of the union of the application on-

tologies, and a set of axioms that deﬁne properties re-

quirements. Our proposal uses ontology as a tool to

solve the semantic heterogeneity problem instead of

using metadata only.

Four types of ontologies dimensions can be iden-

tiﬁed: non-geometric spatial ontology dimension, ge-

ometric to nongeometric spatial ontology dimension,

entirely geometric ontology dimension and temporal

ontology dimension.

In general the Spatial dimension (cf. Figure 3)

describes the representation of the territory surface,

it could comprise speciﬁc members, but that would

restrict the cartographic representation of the data at

one moment given according to only valid territorial

cutting to this moment. For the thematic dimensions,

several logical models has been be deﬁned. The non-

geometric spatial dimension contains nongeometric

data, like nominal data making it possible to locate

a phenomenon in space. Such a dimension can start

with the names of the municipalities and their gen-

eralization can, also be non-geometric. The entirely

geometric dimension is a dimension of which all the

levels and even close generalization are and geomet-

ric. Finally, temporel dimension is an unavoidable el-

ement in any information systems. Temporal dimen-

sion also constitutes strategic information to predict a

future behavior or to explain the causes of the current

state of the things.

Ontologies allows ﬁrstly to deﬁne the overall in-

tegration schema (global ontology) and the different

sources to be integrated. The sources are mapped to

a local ontology which will describe the data sources

Figure 3: MetaOntology.

Figure 4: Spatial object description.

semantics. Following are the rules to map a database

to ontology:

• The database table is mapped to an ontology class.

• If a database table is related to another, then the

two tables are mapped to classes with parents-

child relationship. If a database table is related

to two tables, then the table is divided into two

transferred classes.

• The primary key is mapped to a data type prop-

erty of the ontology. The foreign key would be

mapped to an object property of the ontology.

• The attributes of a table are mapped to properties

of the equivalent class.

Our architecture integrates a reusable metaontol-

ogy which aims to explicitly specify knowledge about

the concepts, relationships, instances and axioms ex-

traction, the learned patterns and frames, and the se-

mantic distance, which is generic and could be used

in other domains.

The metaontology contributes to:

• Ameliorate the ontology building process by spec-

ifying the knowledge that could help the designer

to add concepts and relationships,

• Keep trace of the knowledge that led to the in-

sertion of each element in the ontology (concepts,

relationships, axioms and instances),

• Maintain the clarity and the coherence of the do-

main ontology that is dynamically enriched,

• Reuse the knowledge associated to the construc-

tion of each ontology (domain, structure and ser-

vices), It contains knowledge about the represen-

tation of each ontology of our architecture (do-

main, structure and services) and about the ax-

ioms.

Fuzzy Ontology-based Spatial Data Warehouse for Context-aware Search and Recommendation

163

The output of the data layer is an ontology-based

Data Warehouse which stores semantics, annotations

along with the mechanisms that allow the execution

of analysis operations over the stored data, so we ob-

tain a new semi-structured repository with all sources

integrated.

2.2 The Knowledge and Context-aware

Layer

This layer is mainly composed of a fuzzy ontology-

based Case-Base Reasoning (CBR) component and

Contextualization component.

Fuzzy Ontology-based CBR Component: We

point out that CBR is a problem-solving method

based on the concept of ”case”, which is the de-

scription of a problem and its solution. The main

idea under CBR consists in storing experiences as

cases and problem-solving processes as instances of

cases. When a new problem is found, the system

uses the relevant past stored cases to interpret or to

solve it. The combination of ontologies (as semantic

background) and CBR mechanism (to enrich the

ontology from search feedback) can improve the

performance of Semantic Web search.

In a previous work we proposed to manage and store

the cases in a crisp ontology (Elloumi-Chaabene

et al., 2011). But crisp ontologies are not able

to support uncertain information. One interesting

solution is to integrate fuzzy logic into ontology to

handle vague and imprecise information. Indeed, this

component converts our crisp case base into a fuzzy

case-base ontology. We apply the methodology for

converting a crisp ontology to a fuzzy ontology pro-

posed (Zghal and Gh

ezala, 2014). The most critical

steps in CBR process are the case representation and

case retrieval. We concentrate on these two main

steps to improve the performance of the knowledge

and contextual process.

A set of all the cases is modelled throw the fuzzy

representation. For each fuzzy case, we calculate the

weighted fuzzy similarity degree between the current

case and all the elements of the fuzzy case set, and

select the most similar ones based on maximum simi-

larity degree. Case attributes could be Fuzzy Property

Attributes, Fuzzy Valued Property Attributes and

Fuzzy Relation Attributes. So, when attributes have

uncertain values we need fuzzy ontology-based CBR.

The case is retrieved in the fuzzy ontology according

to the new problem. the types of records features

including numerical, fuzzy, ordinal, lexical, and

semantic types. The fuzzy types are represented by a

fuzzy ontology, and the semantic types are based on

the SDM ontology. The system converts the query

case crisp concept into a fuzzy semantic ontology,

which passes to the retrieval engine to ﬁnd the most

similar cases. Cases are stored in a fuzzy ontology as

concept instances.

During retrieval the fuzzy similarity of a case can is

calculated based on a fuzzy membership function for

each feature that speciﬁes the desired similarity for

any possible difference in the feature values.

Contextualization Component: The context is

deﬁned as a set of ontological concepts present in

the items recently selected by the user. A contextual

fuzziﬁcation for the personalized spatial ontology

is applied. It begins by extracting the context of

the users query from his proﬁle. Then, a contextual

fuzziﬁcation using Babelnet is performed in order

to assign membership values based on the users

interests. Context extraction: We consider the context

to be a concept extracted from the user proﬁle

which is semantically related to the current concept.

In order to determine this concept from a given

proﬁle, a stop word removal, a lemmatization, POS

(Part-Of-Speech) tagging and a matching process

with ontology are ﬁrstly executed. The output of this

process is a set of concepts extracted from the user

proﬁle. The most close concept to the concept is

considered as the context. In order for the context

to be the most representative of the concept, we

measure the degree of co-occurrence between these

two concepts, used as an estimation of similarity be-

tween them. Our component computes the similarity

between two concepts in unsupervised manner using

the number of results returned by a specialized search

engine. The research of similar vocabulary based on

text statistics adopts the hypothesis that the context of

a word can provide enough information for the word

deﬁnition. Therefore, the candidate concept which

has the higher hit counts with the current concept is

considered as the context. The particularity of this

extraction method lies in the exploitation of concepts

that appear the most in a given corpus which helps

extracting relevant concepts the users query and

his proﬁle. The contextual fuzziﬁcation is mainly

based on the fuzziﬁcation process of an existing crisp

ontology is usually performed using a fuzziﬁcation

function. The proposed fuzziﬁcation function relies

on the context ctx in order to favor related concepts to

the proﬁle and the concept. The membership value of

the relation between a given concept and the current

concept, namely contextualized concept.

Fuzzy Ontology Proﬁle: We adopted in this

work the ontological structure for the user proﬁle

ICSOFT-PT 2016 - 11th International Conference on Software Paradigm Trends

164

representation. The idea is to deﬁne individual

ontologies which could be composed later into the

same fuzzy ontology.

Deﬁnition:

The formal Fuzzy Proﬁle Ontology structure is

deﬁned as follows:

f uz

= {C, R, A}, where C is a set of fuzzy concepts,

R is the set of fuzzy relations and A is a set of Axioms

expressed in a logical language.

Let us consider an ontology set C={ O1, O2,... On},

where O1,O2,...,On are fuzzy proﬁle ontologies.

It is important to note that these fuzzy ontologies are

proposed to generate more contextual results.

The proposed fuzzy ontology allows the manage-

ment of the :

• The historic of user manipulations is represented

and the associated dates are also stored in the pro-

posed fuzzy ontology.

• the users’ interactions with the system in order to

implicitly extract information about his interests

and preferences. It is composed of the requested

concepts in the made searches and the relations

between them. For each concept, the last search’s

date is also recorded.

– The positive preferences, which represents the

desired concepts and the relations between

them, are stored in the same ontology. The user

interests are represented by the concepts which

are considered relevant in previous search and

the relations between them.

– The negative preferences, which represents the

undesired concepts and the relations between

them are also stored in the same ontology. Stor-

ing the concepts which are considered irrele-

vant in previous search and the relations be-

tween them, allow the IR system to avoid re-

turning results containing these concepts.

Contextual User Proﬁle Building: The contextual

user proﬁle building step is designed to extract con-

cepts and relations and assign to them membership

values in order to enrich the proposed proﬁle ontol-

ogy with fuzzy concepts and relations. A knowledge

extraction step is ﬁrstly done by analyzing results in

order to extract concepts and relations (both taxo-

nomic and semantic) from resulted documents (both

relevant and irrelevant document groups). Then, a

contextual fuzziﬁcation process is applied in order

to assign membership values to each concept and

relation. Our contextualized concept fuzziﬁcation

method favors the most appearing concepts. It also

favors the concepts that appear with the right context

since it relies on the sum of weights of the concept

and its context. Therefore the memberships values

of contextualized concepts are higher than the ones

without their context.

Recommendation Component: this component

uses the records of previous similar experiences to

generate suggested or create new items when no

existing ones meet the needs or preferences of the

user. In a previous work a comparative study (Haddad

et al., 2015) concerning the proposed approaches of

recommendation has shown that Content-Based Fil-

tering (CBF) have the better results when considering

a quality measure. In our architecture the CBF engine

is mainly used to adapt the recommendations to the

preferences of the users and ensure a degree of diver-

sity and novelty in the suggested recommendations.

The user preferences are represented as vectors, the

intensity of the interest of user for a given concept (a

class or an instance) in a the ontology is measured

taking into account its positive or negative status

described in the previous section.

2.3 The Online Application Layer

The application layer provides results of the previ-

ous components to the user during decision support

process. The search process begins when the user

submits a query or when recommendation is gener-

ated by the knowledge layer. In order to show that

our proposal can have a great interest for contextual

search and recommendation, the architecture has been

implemented. From more technical point of view

the frameworks that have been employed in the im-

plementation of the fuzzy ontology-based CBR are

mainly PostgreSQL, PostGIS and Java-based proto-

type using Fuzzy Owl (Bobillo et al., 2013)(for man-

aging fuzzy ontologies).

3 RELATED WORKS AND

DISCUSSION

Different proposals have been made regarding how

to represent a conceptual multidimensional schema to

model data warehouses and data marts. Different cat-

egorization of data models and comparisons of several

multidimensional data models are proposed in the lit-

erature, where the main identiﬁed levels were: con-

ceptual, logical, physical and formal.

On the one hand, extensions were made to make ap-

propriate the analysis and the algorithms to speciﬁci-

ties of the handled spatial data . On the other hand,

several context models have been proposed to support

Fuzzy Ontology-based Spatial Data Warehouse for Context-aware Search and Recommendation

165

users context (Khouri et al., 2013),...

Recent works proposed to describe data sources by

using global and local ontologies, for more seman-

tic data warehouses (Bellatreche et al., 2013) (Cuz-

zocrea and Simitsis, 2012). Indeed, several limits

have been observed, mainly related to the lack of inte-

gration of: uncertain data for decision support, users

preferences, and their contexts and preferences. Crisp

ontologies are not capable to support uncertain in-

formation and integration of fuzzy logic into ontol-

ogy to handle vague and imprecise information has

proven its usefulness. Moreover, Case Based Rea-

soning is an important ﬁeld which has been applied

to various problems (Elloumi-Chaabene et al., 2011).

To the best of our knowledge among existing works,

this is the ﬁrst time that context, CBR and ontologies

are integrated together in a decision support system

which consider uncertain data at different levels (data

sources, semantics and cases)

4 CONCLUSION AND

PERSPECTIVES

In this paper, we proposed a Fuzzy Ontology-based

Spatial Data Warehouse for contextual search and rec-

ommendation. Our proposal takes place through three

main layers: Data layer, Knowledge and context-

aware layer and Online application layer. Unlike

many previous decision support approaches, the orig-

inality of this work is that it takes into account im-

precise data at two different levels: fuzzy proﬁle on-

tology and fuzzy ontology-based CBR. The architec-

ture has been implemented and an evaluation of the

retrieval tasks is currently conducted.

REFERENCES

Bellatreche, L., Khouri, S., and Berkani, N. (2013). Seman-

tic data warehouse design: From ETL to deployment

a la carte. In Database Systems for Advanced Applica-

tions, 18th International Conference, DASFAA 2013,

Wuhan, China, April 22-25, 2013. Proceedings, Part

II, pages 64–83.

Bobillo, F., Delgado, M., and G

omez-Romero, J. (2013).

Reasoning in fuzzy OWL 2 with delorean. In Un-

certainty Reasoning for the Semantic Web II, Inter-

national Workshops URSW 2008-2010 Held at ISWC

and UniDL 2010 Held at FLoC, Revised Selected Pa-

pers, pages 119–138.

Cuzzocrea, A. and Simitsis, A. (2012). Searching seman-

tic data warehouses: models, issues, architectures. In

Proceedings of the 2nd International Workshop on Se-

mantic Search over the Web, Istanbul, Turkey, August

27, 2012, page 6.

Elloumi-Chaabene, M., Mustapha, N. B., Zghal, H. B.,

Moreno, A., and S

anchez, D. (2011). Semantic-based

composition of modular ontologies applied to web

query reformulation. In ICSOFT 2011 - Proceedings

of the 6th International Conference on Software and

Data Technologies, Volume 1, Seville, Spain, 18-21

July, 2011, pages 305–308.

Haddad, M. R., Zghal, H. B., Ziou, D., and Gh

ezala, H.

H. B. (2015). Towards a generic architecture for rec-

ommenders benchmarking. In ICAART 2015 - Pro-

ceedings of the International Conference on Agents

and Artiﬁcial Intelligence, Volume 2, Lisbon, Portu-

gal, 10-12 January, 2015., pages 435–442.

Khouri, S., Saraj, L. E., Bellatreche, L., Espinasse, B.,

Berkani, N., Rodier, S., and Libourel, T. (2013). Cid-

house: Contextual semantic data warehouses. In

Database and Expert Systems Applications - 24th In-

ternational Conference, DEXA 2013, Prague, Czech

Republic, August 26-29, 2013. Proceedings, Part II,

pages 458–465.

Zghal, H. B., Fa

ız, S., and Gh

ezala, H. H. B. (2003).

CASME: A CASE tool for spatial data marts design

and generation. In Design and Management of Data

Warehouses 2003, Proceedings of the 5th Intl. Work-

shop DMDW’2003, Berlin, Germany, September 8,

2003.

Zghal, H. B. and Gh

ezala, H. B. (2014). A fuzzy-ontology-

driven method for a personalized query reformulation.

In IEEE International Conference on Fuzzy Systems,

FUZZ-IEEE 2014, Beijing, China, July 6-11, 2014,

pages 1640–1647.

ICSOFT-PT 2016 - 11th International Conference on Software Paradigm Trends

166