Toward User Profile Representation in Adapted Mediation Systems

Sara Ouaftouh, Ahmed Zellou and Ali Idri

Mohammed V University In Rabat, Rabat, Morocco

Keywords: User Modeling, Mediation Systems, User Model Representation.

Abstract: The amount of information offered by different software systems is growing exponentially and the need of

personalized approaches for information access increases. This personalization aims to offer the user the

pertinent information corresponding to his needs basing on his profile. For the same purpose, mediation

systems have to identify user preferences in order to offer him the most relevant information .In this work

we discuss different representations of user profile models designed for providing personalized information

access in order to make a comparison and identify the most appropriate for our context in mediation systems.

1 INTRODUCTION

Today, the amount of information available in the

different information systems is increasing expo-

nentially. This information can be in heterogeneous

sources: relational or object sources, flat files,

structured data, applications, web services, etc.

In order to integrate these heterogeneous

information sources and offer an added value to its

services, information systems in different domain

are using data integration technologies. We basically

distinguish between two principle integration

methods: physical integration (J. Widom, 1995) and

virtual integration. The virtual integration also called

mediation allows combining a set of different

information sources by allowing a real-time access

to the sources while concealing their particularities

(G. Wiederhold, 1992). The mediation system must

be able to intercept user queries on the information

system and return appropriate responses.

However, any system that doesn't know who is

asking for information and for what purpose, will

never be able to provide more than general answers.

Therefore, we need a mechanism to adapt the

behavior of information systems to user's preferen-

ces. When a system integrates this mechanism of

adaptation, it is called an «Adaptive system» (K.

Cheverst et al., 2002). These mechanisms must be

able to provide the system with some "context

awareness" by extracting from the user context, the

information needed to identify his preferences. The

system will then provide the user with personalized

services. The personalization or adaptation in

information system is based on the concept of user

profile.

The user profile is defined as a set of information

describing the user and simulating his preferences.

The user profile is considered as a set of structured

data describing the interaction environment between

a user and a system (Y. Elallioui and O. El Beqqali,

2012). In the domain of Internet search engines (S.

Calegari and G. Pasi, 2010), the user profile is used

in order to have structured representation of user’s

interests.

The implementation of a user profile requires the

creation of a user model (R. Guha et al., 2015).

Adaptation via user modeling has started by the end

of the 1970s before the introduction of the Web (A.

Kobsa, 2001), recently it has become a main compo-

nent of many web applications and information

systems in general.

In this work we present a comparison between

the different representations of a user profile in order

to deduce the most appropriate to adapt in the

context of mediation system personalization.

The remaining parts of this paper can be

summarized as following: in section two, we present

mediation systems. Section three describes the

different factors that made the personalization of

mediation system a necessity to satisfy user’s

expectations while section four is dedicated to

discuss the different representations of user models

that will be compared and discussed in section Five.

We conclude then this paper and shares some future

works.

Ouaftouh, S., Zellou, A. and Idri, A.

Toward User Proﬁle Representation in Adapted Mediation Systems.

DOI: 10.5220/0006043000810087

In Proceedings of the 8th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2016) - Volume 2: KEOD, pages 81-87

ISBN: 978-989-758-203-5

2 MEDIATION SYSTEMS

Integration systems permit to link together

information coming from sources that are often

heterogeneous and distributed in order to provide a

global view of information to users.

In our work, we focus on the mediator approach

(G. Wiederhold, 1995) which consists of a mediation

layer between the user and information sources in

order to provide the user with a centralized and

uniform view of information by hiding specific

characteristics of their location, access methods and

formats. An information source can be in different

formats.

Figure 1: Mediation architecture.

In a general manner, the user interacts with the

system by querying a global schema (M. Lenzerini,

2002) which is a virtual representation of the data on

the sources. In the one hand, the mediation system

will perform the sources processing task to retrieve

information satisfying the user query. On the other

hand, the mediation system will hold an internal

representation of information sources, called source

or local schema, so that relations exists between

global entities and the local schema: the mapping

represents these relationships. Some software

modules called wrappers hide source characteristics

to facilitate the interaction between the mediation

system and the sources.

Mediation systems are very useful in the

presence of heterogeneous data sources, because

they make the user feel like using a homogeneous

system. Among the different categories of mediation

systems applications, we can mention the

information retrieval applications; online decision

support systems and more generally, knowledge

manage-ment applications.

However, mediation systems represent some

limitations, including the fact that query rewriting in

terms of data sources views requires the knowledge

of the sources. There is also the risk of non-

availability of all information sources needed to

build a response at the time of issuing the request by

the user. Furthermore, it is important to address the

problem of adaptability of the mediation system to

users’ needs due to the large number of data sources,

which may contain redundant information and

varying quality. In our work, we are interested in

how to personalize mediation systems in order to

resolve this last problem.

3 NECESSITY TO PERSONALIZE

MEDIATION SYSTEMS

As we presented in the previous section, despite the

benefits of mediation systems, they suffer from

some inconvenient. The one we are interested in our

work is caused by the big number of information

sources, which may contain redundant information

in addition to vocabulary problems like polysemy

and synonymy. In general, the evaluation of a user

query is independent of the context and the needs of

the user who issued it. Therefore, the same query

submitted by two different users, produces the same

results even if these users have different

expectations which could make the search results of

mediation systems not beneficial to the user. The

user is then faced to a large number of information

that doesn’t correspond to his expectations when he

submitted the request. However in personalized

systems, the user preferences are included. Applying

personalisation in an e-commerce context for

example, if two different users send the same query

to look for shoes, they could get different results.

Considering that the first user is a girl who lives in

Brazil; she will get as propositions, teenager sandals

style corresponding to the warm weather in Brazil.

The second user who is a 60 years old man living in

Germany will get shoes corresponding to old people

generation. In addition to adapted results, e-

commerce web site implement the concept of

recommender systems that permit the possibility to

give the user suggestion about recommended content

that could interest him according to his profile.

In order to offer the user the personalised answer

adapted to his expectations, we need first to identify

his needs and preferences. Many works in the

literature discusses how information systems are

being adapted to user’s expectations. Adapted

systems take into account the different

characteristics of the user and his situations in

different contexts basing on the concept of user

profile.

KEOD 2016 - 8th International Conference on Knowledge Engineering and Ontology Development

We usually distinguish two phases in the

modelling of the user profile: initialization and

update. In fact, the profile is initialized on the first

use of the system. The first elementary step to build

the user profile is the collect of user’s information (S.

Schiaffino and A. Amandi, 2009). We separate here

between three methods: Explicit, implicit and hybrid

information gathering. To build a user profile

different dimensions are considered: user’s

knowledge, preferences, habits, physical abilities,

intentions, psychological states and geographical

location (S. Ouaftouh et al., 2015). This list is not

exhaustive, it could include, for example, personal

data about the user, his professional or social role,

etc. These dimensions can represent relatively stable

characteristics over time or changing ones that are

therefore updated over the time. In general, the

nature of the information contained in the user

profile strongly depends on the application and

purpose of the system that implements it. The user

profile is then evaluated, modelled and represented

in a particular form. In this context and in a

perspective of user profile integration in the

conception of mediation systems, what is the most

appropriate user profile representation?

4 USER MODEL

REPRESENTATIONS

Adaptive systems are based on user model in order

to have different behaviors for different users. The

User Model is a representation of the information

about a specific user. In literature, we find different

categorizations for user models basing on different

characteristics.

A. Keyword-based User Modeling:

Keyword-based user modeling was initiated in the

domain of information retrieval and filtering (P.

Brusilovsky and C. Tasso, 2004), where the content

of a document is represented as a vector of terms

called keywords, extracted from the text. The

adaptive information retrieval and filtering applica-

tions combined for example a history of user’s

queries, accessed documents, e-mails, chat, etc. in a

form of a keyword vector and use this vector for

adapting a future retrieval or filtering process.

Many adaptive systems model users’ information

interests or needs as vectors of keywords extracted

from the documents that the users have browsed or

requested. Figure 2 is an example of a keyword

vector representation that models user interest in a

particular domain of interest, each keyword «K

»,

corresponds to a term found in the content of

document consulted by the user. Each keyword «K

is associated with a numerical weight «W

representing its importance in the profile.

)

Domain of interest

… W

… K

Figure 2: Keyword-based user model representation.

Each keyword can represent a topic of interest.

Keywords can be grouped in categories to reflect a

more standard representation of users’ interests.

Some systems model users’ interests as networks of

keywords instead of plain lists, where nodes

represent keywords and arcs connect keywords co-

occurring in the content.

Keyword-based modeling support only simple

content data. To remedy problems like homonymy

and synonymy, Natural Language (NL) technologies

are required. Pure keyword based modeling is not

able to represent the true meaning of the content. It

relies on statistical regularities within the text and

provides a framework for retrieving statistically

close documents.

B. Overlay User Modeling:

This approach also called concept user modeling,

was employed for the first time in 1988. It’s about

graining the domain knowledge into elementary

components and using them to evaluate user’s

knowledge. The domain knowledge components

have been named differently by different authors:

topics, knowledge elements and – the most used –

concepts.

Figure 3: Overlay user model.

A concept represents an atomic piece of

declarative domain knowledge, coherent and

semantically complete. An aggregate of concepts

forms the domain model. The overlay user model

Toward User Proﬁle Representation in Adapted Mediation Systems

consists on a set of concept-value pairs, where the

value represents an assessment of a particular

concept. The user is characterized in terms of user’s

knowledge about these concepts in relation to the

top level knowledge.

As shown in figure 3, «Ci», correspond to the

different concepts from a domain knowledge, the

user model represent user knowledge about these

concepts in relation to the ideal knowledge level.

The benefit of the overlay user model is its precision

and flexibility. An overlay model is capable to

dynamically and precisely reflect the evolution of

users’ characteristics.

C. Stereotype User Modeling:

The main goal of adaptive systems is to adjust its

behavior to each user’s needs. However, for some

contexts it is possible to identify typical categories

of users that use the system the same way, expect

from it similar reactions and can be described by

similar characteristics. These categories are called

stereotypes. Stereotype user modeling is one of the

oldest approaches to user modeling. It was

developed in the works of Elaine Rich (E. Rich,

1997). An adaptive system using stereotype-based

modeling does not update every single feature of the

user model directly; it uses a stock of preset

stereotype profiles. Hence, the application can make

expectations about a user even though there could be

no data about that specific area, because studies have

shown that other users in this stereotype have the

same characteristics. Such categories are called

stereotypes.

Figure 4: stereotype user model.

As shown in figure 4, an adaptive system in a

specific context can identify a set of stereotypes Si

and each user «U

» of the system is assigned to a

stereotype. A stereotype can correspond to one or

many users.

As an example of stereotype-based user modeling is

linear set of categories for typical levels of user

proficiency: novice, beginner, intermediate, and

expert.

Stereotype-based user modeling is advantageous

when from a little evidence about a user the system

should infer a great deal of modeling information.

However, for modeling fine-grained characteristics

about users as is the case in knowledge level of a

particular concept, the overlay models should be

employed.

D. Constraint-based user Modeling:

Constraint-Based model is a way to represent the

domain knowledge as a set of constraints. It is

mostly used for modeling users in Intelligent

Tutoring Systems (A. Mitrovic, 2012). Constraint-

based tutors are Intelligent Tutoring Systems that

use Constraint-Based modeling to represent the user

model as a set of information about abilities,

knowledge and needs of the user. Constraint-based

tutors are problem-solving environments; in order to

provide personalized instruction; they diagnose users’

actions, and maintain user models. These models are

then used to provide appropriate examples and offer

hints and help where the user is most likely to need

them. In this approach, every constraint represents

an acceptable set of equivalent problem states and a

violated constraint indicates an error.

Each constraint consists of an ordered pair (Cr, Cs),

where Cr is the relevance condition and Cs is the

satisfaction condition. The relevance condition

checks whether the constraint is applicable to the

user solution by testing the features of the solution.

The satisfaction condition specifies additional test

that must be met by correct solutions. If the

relevance condition is met, but the satisfaction

condition is not, then the user’s solution is incorrect.

Therefore, the general form of a constraint as

presented in figure 5 is:

(Cr ,Cs)

If <

Cr > is true,

Then <

Cs > had better also be true.

Figure 5: Constraint-based user modelling.

E. Collaborative Filtering:

Being different from user modeling technologies

cited before, this approach relies on modeling the

user in terms of his relationships with other users. A

typical collaborative user model is based on a vector

of ratings that the user provided for particular items.

The original implementation of this approach is

recommender systems (N. Tintarev and J. Masthoff,,

2011), they recommend to the active user the items

that other users with similar preferences liked in the

past. The similarity in preference of two users is

calculated based on the similarity in the rating

history of the different users. The central hypothesis

KEOD 2016 - 8th International Conference on Knowledge Engineering and Ontology Development

behind this method is that other users’ opinions can

be selected and combined in such a way to provide a

reasonable prediction of the active user’s preference.

Intuitively, we assume that, if users agree about the

quality or relevance of some items, then they will

likely agree also about other items.

The information domain of a collaborative filtering

system consists of users which have expressed

preferences for various items. A preference

expressed by a user for an item is called a rating and

is frequently represented as a (User, Item, Rating)

triple. These ratings can take many forms, depending

on the system. Some systems use real or integer

valued rating scales such as 0–5 stars, while others

use binary (like/dislike) scales (J. B. Schafer et al.,

2007). As represented in the example in table 1, the

set of all rating triples forms a matrix referred to as

the ratings matrix. The values of R

correspond to

the rating each user U

gave to a specific item I

(User, Item) pairs where the user has not expressed a

preference for the item are unknown values in this

matrix and are marked with ‘?’ to indicate unknown

values i.e. the user has not rated that item.

)

… I

? R

… R

? R

… ?

… … … … … …

… ?

Figure 6: Collaborative filtering using ratings matrix.

Given a user and an item, what is the user’s likely

preference for the item? If the ratings matrix is

viewed as a sampling of values from a complete

user–item preference matrix, than the predict task

for a recommender is equivalent to the matrix

missing values problem.

Recommender technology, often based on

collaborative filtering, has been integrated into many

e-commerce and online systems. An important

motivation for doing this is to increase sales volume;

customers will likely buy an item if it is suggested to

them but may not otherwise.

F. Bayesian Networks:

Bayesian Networks are probabilistic graphical models

that consist of a qualitative and a quantitative part.

The qualitative part is the structure of the network: a

directed acyclic graph where nodes correspond to

variables and arcs representing influences between

variables. The quantitative part provides the condi-

tional probability tables that make up the network

settings.

More precisely, a Bayesian Network is a set

consisting of a directed acyclic graph and n random

variables

, X

, .. , X

) such that there is a bijection between

the set of vertices graph and the set of random

variables and that:

P(X1, X2, .. , Xn) =

∏

P(Xi|pa(Xi))





where pa(Xi)

is the set of parents of X

in the graph.

Multiple systems used Bayesian Networks to model

the relations between different components or

dimensions of a user model, such as emotions, goals

and knowledge (X. Zhou and C. Conati, 2003).

Other systems used them to implement an overlay

user model with internal inference capabilities, where

every node represents a domain concept and links

represents the concept relations (F. De Rosis, et

al.,1992).

In the e-commerce applications, it is often useful

to model a customer without developing any explicit

modeling rules about him, but only by identifying

certain statistical predictabilities that can be used for

constructing an effective selling strategy. A user

model in this case can contain a set of transactions

matched against an association rule of items bought

together or satisfying some conditions of buyers’

behavior, or belonging to a cluster of similar buyers

(S. Sosnovsky, 2010).

5 COMPARISON AND

DISCUSSION

A. Comparison:

After having presented the different categories of

user models and the different representations, we

constructed a table summarizing each type of user

model representation to help us make a comparison.

Table I represent for each category, the definition,

representation, explication, principle domain of use,

advantages and disadvantages.

After having presented the different categories of

user models and the different representations, we

constructed a table summarizing each type of user

model representation to help us make a comparison.

The keyword based modeling approach can be

considered lightly similar to overlay user modeling

because it uses elements of domain representation as

a reference to express user characteristics. Overlay

user modeling is most used to model student

knowledge in e-learning systems while keyword

based models are used to model user interests for

example in the domain of information retrieval and

filtering. Constraint-based models are also used in

Toward User Proﬁle Representation in Adapted Mediation Systems

Table 1: Comparison table between different user model representations.

Name Overlay Keyword Stereotype Constraint-based

Collaborative

Filtering

Bayesian

Network

Principe

Measure how well

a user knows a

concept.

Measure user’s

interests on a

specific keyword

Define typical

categories of users.

Define a set of

constraints

representing

domain

knowledge.

Model user in

terms of his

relationships with

other users.

Model the

relations between

the components of

a user model

Representation

Vector of Concept-

value pairs.

Vector of

keywords-value.

Set of preset

profiles.

Set of constraints. Vector of ratings :

(User, Item,

Rating)

Directed acyclic

graph

Explication

Concepts are

subset of domain

knowledge.

Keyword extracted

from the text

consulted by the

user.

Users that belong

to the same

stereotype

described by

similar

characteristics.

Diagnose of user’s

actions to provide

personalized

instructions and

decisions.

Predict user

interest basing

similar user’s

feedback.

Nodes of the graph

represent variables

and arcs represent

influences between

variables.

Principle

Domain of use

E-learning:

Modeling user’s

knowledge.

Information

retrieval and

filtering.

Model user

interests.

Modelling groups

of users.

Modeling user’s

knowledge in

Intelligent

Tutoring Systems

Recommender

system: predicting

user's interests and

preferences.

Modeling

emotions, goals

and knowledge.

Advantages

Automatic

modeling of

content.

Faster results Make expectations

about a user even

though there is no

data about him

Encodes

correct domain

knowledge

No need to

additional

information about

user except

ratings.

Model a user only

by identifying

certain statistical

predictabilities

Disadvantages

Support only

simple content.

Support only

simple content

data.

Lack of semantics

and Polysemy.

Use of preset

profiles, no update

to users’ features

Overspecificity: a

highly detailed

model of user’s

knowledge

Cold-start

problem : no rating

available in the

start of the system

Difficult reaching

agreement on the

Bayesian network

structure with

experts

the domain of e-learning especially for intelligent

tutoring systems to model students’ knowledge

basing on a set of constraint.

For its part stereotype user modelling can be used

to model groups of users in the case the categories of

system’s user are predefined.

Collaborative filtering modeling is mainly used in

recommender systems to predict user’s interests and

preferences. This method utilizes only ratings and do

not require any additional information about users or

items. The principal disadvantage of Collaborative

filtering systems is the Cold-Start problem which

cannot produce recommendations if there are no

ratings available.

To model emotions, goals and also knowledge,

the Bayesian network approach can be used with the

advantage of modeling the user only by identifying

certain statistical predictabilities.

B. Discussion:

Roughly, the user dimension considered in a

particular system depends on the intended field of

application. For example, in the domain of e-

learning we are mainly interested to model user’s

knowledge, skills and interests. In e-commerce

context, it’s more interesting to know user’s

preferences. Among the deductions of our

comparison between the different user models

representation, each representation is satisfying the

particularities of a certain application domain.

In order to select the suitable user model

representation for our context, which is mediation

systems, we have to specify our system’s

characteristics. To personalize a mediation system,

we are interested to model the most of user’s

dimensions. Mediation systems are characterized

with a set of exchanges between the mediator and

sources (couple of requests and responses). We can

then recommend the use of keyword based user

modeling or collaborative filtering as the most

appropriate to be applied in a mediation system

context.

KEOD 2016 - 8th International Conference on Knowledge Engineering and Ontology Development

6 CONCLUSIONS AND

PERSPECTIVES

A mediation system is a powerful tool allowing easy

access to different information collected from

distributed data sources that can be heterogeneous. It

must integrate diverse information in order to

provide the user with a centralized and uniform view

of data by masking the specific characteristics of

their location, access methods and formats. In a

perspective of mediation system improvement, it has

been necessary to adapt system’s responses to user’s

expectations, represented by his profile, via the

implementation of a user model.

The user profile corresponds to a set of

information describing the user. It contains data that

represent user preferences. The implementation of a

user profile requires the creation of a user model. To

identify the most suitable user model representation

in a mediation system, we presented a study about

the different approaches found in the literature.

Many authors have classified user modeling

approaches basing on a variety of criteria.

As our goal is to evaluate the user profile, we

were based on user model representation and

distinguish between overlay, keyword, stereotype,

constraint-based, collaborative filtering and

Bayesian network models. Each representation is

mainly used to model a set of user’s dimension and

is generally applied in a particular domain of use.

Practically, the user dimension considered in a

particular system depends on the envisioned field of

application. Referring to the particularity of

mediation system and as deduced from the

comparison table of user models that we constructed,

we recommend the use of keyword user model or

collaborative filtering approach.

In our future work, we will focus on applying

one of the suggested representations to personalize

mediation systems. This model will implement a set

of dimension that we qualified necessary in a

mediation system.

REFERENCES

J. Widom, “Research Problems in Data Warehousing,” 4th

Int. Conf. Inf. Knowl. Manag., pp. 25–30, 1995.

G. Wiederhold, “Mediators in the architecture of future

information systems,” Computer (Long. Beach. Calif).,

vol. 25, no. 3, pp. 38–49, 1992.

K. Cheverst, K. Mitchell, and N. Davies, “Adaptive

Hypermedia,” vol. 45, no. 5, pp. 47–51, 2002.

Y. Elallioui and O. El Beqqali, “User profile Ontology for

the Personalization approach,” Int. J. Comput. Appl.,

vol. 41, no. 4, pp. 31–40, 2012.

S. Calegari and G. Pasi, “Ontology-Based Information

Behaviour to Improve Web Search,” Futur. Internet,

vol. 2, no. 4, pp. 533–558, 2010.

R. Guha, V. Gupta, V. Raghunathan, and R. Srikant, “User

Modeling for a Personal Assistant,” Proc. Eighth ACM

Int. Conf. Web Search Data Min. - WSDM ’15, pp.

275–284, 2015.

A. Kobsa, “Generic User Modeling Systems,” User Model.

User-adapt. Interact., vol. 11, pp. 49–63, 2001.

G. Wiederhold, “Mediation in information systems,” ACM

Comput. Surv., vol. 27, no. 2, pp. 265–267, 1995.

M. Lenzerini, “Data Integration: A Theoretical

Perspective,” Proc. Twenty-first ACM SIGMOD-

SIGACT-SIGART Symp. Princ. Database Syst., pp.

233–246, 2002.

S. Schiaffino and A. Amandi, “Intelligent User Profiling,”

Ifip Int. Fed. Inf. Process., vol. LNAI 5460, pp. 193–

216, 2009.

S. Ouaftouh, A. Zellou, and A. Idri, “User profile model:

A user dimension based classification,” 10th Int. Conf.

Intell. Syst. Theor. Appl., pp. 1–5, 2015.

P. Brusilovsky and C. Tasso, “User modeling for web

information retrieval,” Pref. to Spec. issue User Model.

User Adapt. Interact., vol. 14, no. 2–3, pp. 147–1571,

2004.

E. Rich, “User Modeling via Stereotypes,” Cogn. Sci., vol.

3, no. 3597, pp. 329–354, 1979.

A. Mitrovic, “Fifteen years of constraint-based tutors:

What we have achieved and where we are going,”

User Model. User-adapt. Interact., vol. 22, no. 1–2, pp.

39–72, 2012.

N. Tintarev and J. Masthoff, Recommender Systems

Handbook, vol. 54. 2011.

J. B. Schafer, D. Frankowski, J. Herlocker, and S. Sen,

“Collaborative Filtering Recommender Systems,” Int.

J. Electron. Bus., vol. 4321, no. 1, pp. 291–324, 2007.

X. Zhou and C. Conati, “Inferring user goals from

personality and behavior in a causal model of user

affect,” Proc. 8th Int. Conf. Intell. user interfaces -

IUI ’03, p. 211, 2003.

F. De Rosis, S. Pizzutilo, A. Russo, D. C. Berry, and F. J.

Nicolau Moulina, “Modeling the user knowledge by

belief networks,” User Model. User-adapt. Interact.,

vol. 2, no. 4, pp. 367–388, 1992.

S. Sosnovsky, “Ontological Technologies for User

Modeling,” vol. 5, no. 1, pp. 32–71, 2010.

Toward User Proﬁle Representation in Adapted Mediation Systems