KNOWLEDGE MANAGEMENT FOR ADAPTED INFORMATION

RETRIEVAL IN UBIQUITOUS ENVIRONMENTS

Angela Carrillo-Ramos, Marlène Villanova-Oliver, Jérome Gensel and Hervé Martin

LSR Laboratory, SIGMA Team, 681 rue de la Passerelle, 38402 Saint Martin D’Hères, France

Keywords: Agents, Knowledge Management, Query Routing, Adaptation.

Abstract: PUMAS is a framework based on agents which provides nomadic users with relevant and adapte

information. Using PUMAS, information delivered to nomadic users is adapted according to, on the one

hand, their preferences and history in the system and, on the other hand, the limited capacities of thei

Mobile Devices (MDs). This framework is composed of four Multi-Agent Systems (MAS, Connection MAS,

Communication MAS, Information MAS and Adaptation MAS) for handling adaptation. In this paper, we

illustrate how the PUMAS agents augment user queries with information about her/his characteristics an

those of her/his MD and, how the Router Agent (which belongs to the Information MAS) redirects the use

queries towards the different Web based Information System (WIS) which contain all or part of the

information for answering them and which execute on servers or MDs.

1 INTRODUCTION

Ubiquitous Computing is defined by the W3C

(http://www.w3.org/TR/webont-req/) as an emerging

paradigm of personal computing characterized

mainly by the use of Mobile Devices (MDs). The

term MD refers generally to small, handheld and

wireless computing devices, used to access Web

based Information System (WIS). WIS are systems

which enable collecting, structuring, storing,

managing and diffusing information, like traditional

Information Systems (IS) do, but over a Web

infrastructure. WIS provide users with complex

functionalities which are activated through a Web

browser in a hypermedia interface. WIS designers

must be provided with mechanisms and architectures

that cope with the reduced capabilities of the MDs,

in order to efficiently retrieve and deliver data using

these devices. The WIS must provide users with

useful information retrieved from an intelligent

search and presented in a suitable way. We believe

that the agent paradigm is an interesting approach

for this purpose. The Multi-Agent System (MAS)

approach is defined in (El Fallah-Seghrouchni et al.,

2004) as a credible paradigm to design distributed

and cooperative systems based on the agent

technology.

The interest of MAS, when the Internet is used to

access and exchange information through MDs (that

they call “smart devices”), is shown in (Ramparany

et al., 2003). In this case, agents can be useful to

represent user characteristics inside the system and

the MDs can work like “cooperative devices”. The

W3C defines an agent as “a concrete piece of

software or hardware that sends and receives

messages”. In our context, these messages can be

used to access a WIS and to exchange information.

The MD applications require network architectures

able to support automatic and ad hoc configuration

which consider features of the ubi

quitous computing

environment such as heterogeneity, mobility,

autonomy, high distribution, etc. Such environment

is defined in (Pirker et al., 2004) as a dynamic

distributed network of embedded devices and

systems that can interact with humans to satisfy their

requirements and provide a variety of information,

communication, and collaboration services.

In order to provide nomadic users only with the most

relevant information (i.e. “the right information in

the right place at the right time"), a MD application

must embed mechanisms for propagating the user

queries towards the “right” information sources

(stored in one or several devices) which can answer

these queries considering user preferences, features

of her/his MDs, her/his location, etc. This is the

Carrillo-Ramos A., Villanova-Oliver M., Gensel J. and Martin H. (2006).

KNOWLEDGE MANAGEMENT FOR ADAPTED INFORMATION RETRIEVAL IN UBIQUITOUS ENVIRONMENTS.

In Proceedings of WEBIST 2006 - Second International Conference on Web Information Systems and Technologies - Internet Technology / Web

Interface and Applications, pages 21-29

DOI: 10.5220/0001237900210029

 SciTePress

main purpose of the Query Routing process. (Xu et

al., 1999) define this process as the general problem

of, on the one hand, evaluating the query using the

most relevant data sources and, on the other hand,

integrating results returned from data sources. In

order to optimize the Query Routing process,

(Agostini et al., 2004) and (Park et al., 2004)

propose to use some metrics related to the

trustworthiness of the information sources, their

capability to satisfy user information needs and their

timeliness of information delivery.

PUMAS (Carrillo et al., 2005a) is a framework for

retrieving information distributed between several

WIS and/or different types of MDs. The architecture

of PUMAS is composed of four MAS (Connection

MAS, Communication MAS, Information MAS and

Adaptation MAS), each one encompassing several

ubiquitous agents which cooperate to achieve the

different tasks handled by PUMAS (e.g., MD

connection/disconnection, communications between

agents, information exchange, storage and retrieval,

etc.). In this paper, we present the activities of

representation and data exchange of the PUMAS

agents (activities based on XML files). Through

PUMAS, the final objective is to build and propose a

framework which additionally to the management of

accesses to WIS performed through MDs, is also in

charge of performing an adaptation of information

according to user profiles (which refers to their

needs, preferences, histories in the system, current

location, etc.) and, the technical capabilities of

her/his MD. This paper focuses on the representation

of knowledge managed by PUMAS agents (to

achieve the adaptation tasks and support the Query

Routing process executed by the Router Agent) in

order to redirect queries formulated by users towards

the different WIS. We show here how the Knowledge

Bases (KBs) managed by PUMAS agents are used by

this process. We also explain and illustrate each

activity of the Query Routing process using as

example an airport WIS and a scenario we briefly

describe:

A passenger equipped with her/his MD must take a

plane. Let us suppose that she/he must arrive three

hours before for checking in and that she/he also

must buy some gifts at the duty free shops. Let us

assume that, at the airport, each airline and shop

has a WIS which provides customers with

information about their services (e.g., departure and

arrival schedule) and their products (e.g., sales, new

products). The passenger wants to know the closest

duty free shops to the departure gate of her/his flight

which sell each article of her/his gift list (at the

lowest price). Let us suppose that several shops sell

the same products (e.g., souvenirs, books, post

cards, liquors) which correspond to what the user

would like to buy.

The paper is organized as follows. We present in

section 2, the goal and the architecture of the

PUMAS framework. We describe more particularly

the data representation and data exchange of the

agents and their managed information. In section 3,

we present Knowledge Management, especially that

performed by agents which belong to the

Information and Adaptation MAS for adaptation

purposes. In section 4, through the example scenario

described above, we explain the Query Routing

process performed by the Router Agent. Finally, we

present some related works before we conclude in

section 6.

2 THE PUMAS FRAMEWORK

The architecture of PUMAS is composed of four

Multi-Agent Systems (MAS) which will be explained

in the following subsections (see Figure 1).

Figure 1: The PUMAS Architecture.

WEBIST 2006 - INTERNET TECHNOLOGY

The PUMAS framework has been extended

compared to the architecture presented in (Carrillo et

al., 2005a). We have introduced in (Carrillo et al.,

2005b) a new MAS, called the Adaptation MAS, in

the architecture of PUMAS. The agents belonging to

the Adaptation MAS have as responsibilities to

manage specific XML files which contain

information about the user and her/his MD. Its

knowledge allows selection and filtering of

information for users. This paper focuses, on the one

hand, on the managed knowledge and the exchanged

information between agents of PUMAS, especially,

those belonging to the Information and the

Adaptation MAS for adaptation purposes, and on the

other hand, on the strategies followed by the Router

Agent in order to perform the Query Routing

process. The following subsections give a short

description of each MAS, focusing on the

information exchanged between PUMAS agents. A

detailed description of the Connection MAS, the

Communication MAS and the Information MAS can

be found in (Carrillo et al., 2005a) while the

Adaptation MAS is presented in detail in (Carrillo et

al., 2005b).

2.1 The Connection MAS

This MAS includes several Mobile Device Agents

and one Connection Controller Agent.

A Mobile Device Agent is executed on the user’s

MD. This agent manages the Device Profile XML

file, located on the user’s MD, which describes the

characteristics of the MD, using OWL (Ontology

Web Language, http://www.w3.org/2004/OWL/) in

order to define a common ontology for all agents

which share this information (e.g., the

DisplayFilterAgent which belongs to the Adaptation

MAS, see section 2.4). This file contains some

information about hardware and software

requirements, network status, type of hypermedia

files supported by the MD, conditions for

disconnecting (i.e. finishing sessions), etc. A Mobile

Device Agent also manages the Session XML file

which describes characteristics of the user sessions:

who is the user connected, when the session begun

and what MD is connected. This file will be

exchanged with the UserAgent (belonging to the

Adaptation MAS).

The Connection Controller Agent executes on the

central platform of PUMAS. This agent checks

connections established by users and agent status

(e.g., connected, disconnected, killed, etc.). It also

gets the user’s location and the MD type (e.g., PDA)

from the User Location XML file (which contains

physical and logical location features) and from the

Device Profile XML file (which contains MD

features). Both files are provided by the Mobile

Device Agents and locally managed by the

Connection Controller Agent.

The XML files (i.e., User Location, Session and

Device Profile XML files) managed by the Mobile

Device Agent and the Connection Controller Agent

have been defined using

OWL and the extensions

introduced by (Indulska et al., 2003) to CC/PP.

These extensions include some user characteristics

like location (physical and logical location),

requirements of available applications (hardware,

software, browser and WAP requirements),

characteristics of sessions (user, device, application)

and user profiles (general user’s requirements,

preferences).

2.2 The Communication MAS

This MAS is composed of several Proxy Agents, one

MDProfile Agent and one Coordinator Agent. These

agents are located in the central platform of PUMAS.

There is one Proxy Agent for each connection from

a Mobile Device Agent. The main task of this agent

is to represent a Mobile Device Agent within the

system and has the same properties and behaviour as

the Mobile Device Agent except those concerning

the connection.

The MDProfile Agent has to check the user profiles

according to her/his MD. This agent shares

information about specific MD features for user

session with the DisplayFilterAgent (which belongs

to the Adaptation MAS).

The Coordinator Agent is in permanent

communication with the Connection Controller

Agent in order to verify connection status of the

agent which searches for information. This agent

knows all agents connected in the system using a

yellow pages mechanism. If there are some

problems with the Connection Controller Agent

(e.g., if the Connection Controller Agent fails or, if

there is a lot of connections), the Coordinator Agent

can play the role of the Connection Controller Agent

up until problems are fixed. At that moment, the

Connection Controller Agent and the Coordinator

Agent must synchronize information about the

connected agent and check current connections.

KNOWLEDGE MANAGEMENT FOR ADAPTED INFORMATION RETRIEVAL IN UBIQUITOUS

ENVIRONMENTS

2.3 The Information MAS

The Information MAS is composed of one or several

Receptor/Provider Agents, one or several Router

Agents and one or several ISAgents.

The Receptor/Provider Agents which are located in

the central platform of PUMAS own a general view

of the whole system. They receive requests that are

transmitted from the Communication MAS and

redirect them to the Router Agents. Once a query has

been processed by the ISAgents, a Receptor/Provider

Agent checks whether the query results consider the

user profile according to her/his preferences, history

in the system, etc.

In order to redirect a query to the “right” WIS, the

Router Agent applies a strategy which depends on

one or several criteria (see section 4). This agent

also compiles results returned by the ISAgents and

analyzes them (according to defined criteria in the

user preferences) to decide whether the whole set of

results or only a part of it has to be sent to the

Receptor/Provider Agents.

An ISAgent associated with a WIS (and which

executes on the same device that the WIS) receives

user queries from the Router Agent and is in charge

of searching for information. Once a result for a

query is obtained, the ISAgent returns it to the

Router Agent. An ISAgent can execute a query by

itself or delegate this task to the adequate WIS

component.

2.4 The Adaptation MAS

This MAS is composed of one or several

UserAgents, one DisplayFilterAgent and one

ContentFilterAgent. They are located in the central

platform of PUMAS.

Each UserAgent manages a User Profile XML file

(defined using OWL) which contains personal

characteristics of a user (e.g., user ID, location, etc.)

and her/his preferences (e.g., the user wants only

video files). This file is obtained by means of the

Mobile Device Agent (which executes on the user’s

MD). There is only one UserAgent which represents

a user at the same time and centralizes all the

characteristics of the same user who can have

several sessions opened. The UserAgent

communicates with the ContentFilterAgent to send

the User Profile XML file in order to update user

preferences.

The DisplayFilterAgent manages a Knowledge Base

which contains general information about the

characteristics of different types of MDs (e.g.,

format files supported) and knowledge acquired

from previous connections (e.g., problems and

capabilities of networks according to data

transmissions).

The ContentFilterAgent manages a Knowledge Base

which contains preferences and characteristics of the

users. It communicates with the UserAgent, asking

for user preferences defined for a specific session

(e.g., the current session).

3 KNOWLEDGE MANAGEMENT

IN PUMAS

In this section, we describe the knowledge managed

by agents of the Information and the Adaptation

MAS of PUMAS to achieve their adaptation tasks

and support the Query Routing process. This

knowledge is stored in Knowledge Bases (KBs) in

the shape of pieces of knowledge called “facts” and

defined using JESS (which is a rule engine and

scripting environment for building Java applications

which has the capacity to "reason" using knowledge

supplied in the form of declarative rules.

http://herzberg.ca.sandia.gov/jess/). We declare

these facts as instances of JESS Templates in order

to represent user preferences, features of the MD, the

WIS, etc. as described in the following subsections.

3.1 Knowledge of the Information

MAS

The Router Agent stores in its KB a fact for each

WIS. This agent exploits these facts to redirect user

queries. A fact which represents a WIS describes

characteristics of the WIS like its name, managed

information, the type of device where it is executed

(e.g., server, MD) and the ISAgent associated with

the WIS (i.e., the ISAgent which execute on the WIS

and can be asked for information and consequently

answers queries). The following template defines a

WIS:

(deftemplate WIS

(slot name)

(slot agentID)

(slot device)

(multislot info_items)) ; fact (1)

The following fact (instance of the template defined

above) represents the WIS of a store. The WIS is

called StoreWIS and executes on a server. The

ISAgent which executes on this WIS is called

StoreISA. The StoreWIS contains information (a list

of info_items) about the articles, sales, and new

products which are sold in the store:

WEBIST 2006 - INTERNET TECHNOLOGY

(assert (WIS

(name StoreWIS)(agentID StoreISA)

(device server)

(info_items “articles” “sales” “new

products”)))

3.2 Knowledge of the Adaptation

MAS

The ContentFilterAgent (CFA) manages a KB which

contains user preferences. These preferences are

represented as facts defined as follows:

(deftemplate User_Preference

(slot userID)

(slot required_info)

(multislot complementary_info)

(multislot actiontodo)

(slot problem)

(multislot actionforrecovering))) ; fact

(2)

The User_Preference fact is composed of a userID

(which identifies the owner of this preference),

required information (required_info) and

complementary_info which is added to the

User_Preference definition by the CFA and is

inferred from queries of previous sessions (i.e.

information frequently asked simultaneously with

the required_info). This fact is also composed of

information about what and how the user would like

the system to present results (list of actionstodo for

displaying information to her/him) and in the case of

problems, what the system has to do

(actionsforrecovering).

We consider that queries also depend on several

criteria (criteria managed by the CFA): user location,

her/his history in the system, activities developed

during a time period, etc. Such Criterion is defined

as:

(deftemplate Criterion

(slot userID)(multislot criteria)

(multislot attributes)) ; fact (3)

Here is an example of Criterion which expresses

that all of John Smith’s queries depend on his

location, especially if he is in the airport:

(assert (Criterion

(userID “John Smith”)

(criteria location)

(attributes “airport” )))

In the next section, we describe the Query Routing

process which is performed by the Router Agent

exploiting the knowledge we have described in this

section.

4 QUERY ROUTING IN PUMAS

The Query Routing (QR) process in PUMAS is

achieved by the Router Agent (RA) which receives

queries together with user characteristics and those

of their MDs. In order to redirect a query to the

“right” WIS, the strategy chosen by the RA depends

on several criteria: user location, peer similarity,

time constraints, preferences, etc. The strategy can

lead to sending the query to a specific WIS, or to

sending the query through broadcast, or to splitting

the query in sub-queries, each one being sent to one

or several ISAgents (ISAs, agents which belong to

the Information MAS and execute on the WIS). The

RA is also in charge of compiling results returned by

the ISAs and of analyzing them (according to the

defined criteria for the queries, see section 3.2) to

decide whether the whole set of results or only a part

of it will be sent to the user.

In PUMAS, the QR process consists of three

activities, based on the work of (Xu et al., 1999)

which are described and illustrated in the next

subsections, using the airport scenario presented in

introduction.

4.1 Analyzing the Query

This activity is related to the possible split of a query

into sub-queries. The RA analyzes the complexity of

a query. A query is considered as simple if it can be

answered by only one agent and complex if several

agents are required. This analysis is more precisely

based on facts, stored in the KB of the RA, about the

WIS (which notably contains knowledge about

information managed by this WIS). The RA also

analyzes criteria of a query (e.g., location, user’s

activities, etc.), knowledge of the query receivers

(e.g., if the query is directed to specific known

receivers), etc. After this analysis, the RA decides

whether it has to divide a query in sub-queries or

not.

For the scenario, the RA must split the query (“all

the shops which sell the articles of my gifts list”) in

several sub-queries (“all the shops which sell each

article of my gifts list”). The number of sub-queries

depends of the number of articles. If there is only

one article, the query is simple (only one agent will

answer). Otherwise, the query is complex. The RA

must also consider two criteria: proximity of the

departure gate and price of the article in the shop.

For that, the RA asks the CFA for the user

preferences and criteria of the query (i.e., fact (2,3)

and its instances; see section 3.2). The RA could

receive from CFA facts as the following which

KNOWLEDGE MANAGEMENT FOR ADAPTED INFORMATION RETRIEVAL IN UBIQUITOUS

ENVIRONMENTS

expresses that when the passenger “John Smith”

consults the “closest shops”, he also wants those

which sell their products at the “lowest prices”:

(assert (User_Preference

(userID “John Smith”)

(required_info “closest shops”)

(complementary_info “lowest prices”)

(actiontodo show)

(problem “empty list of shops”)

(actionforrecovering cancel)))

4.2 Selecting the Information

Sources

A query could be directed to a specific agent or to a

group of agents; if the query receivers are known,

the selection is simple (the potential information

sources are the specific agents). Otherwise, the RA

selects information sources and computes the

network of neighbours, based on ideas of (Yang et

al., 2004). These authors propose an efficient QR

approach for information retrieval in unstructured

P2P networks. The QR policies are utilized to

determine to how many nodes and to which nodes,

the query should be forwarded in the network. This

work introduces the Routing Guide (RG) concept

which is based on results returned for queries

previously processed, and is used to determine

routing directions for queries. In the information

retrieval process in P2P systems, each node owns a

collection of data that is shared with other nodes.

When a user submits a query, her/his node becomes

the source of the query (requestor). Such node may

send query messages to any number of its

neighbours. Then any neighbour receiving the query

message firstly processes the query over its local

information. If the node finds some results, it sends

back its address to the requestor node so that it

fetches data directly.

In our proposal, a peer is neighbour of some others

if it satisfies a set of characteristics (criteria defined

in user preferences of an application). For example,

close location, same activities, same role, similar

knowledge, colleagues who work in group. The

characteristics are not restricted to proximity criteria.

We can consider several cases for composing a

network of neighbours in which each node is an

information source:

First case, there could be one or several agents

which answer the same query. The simplest way of

composing this network is to group all these agents.

This gathering is useful, for example, when the RA

does not have information about the sources or when

it is the first time that the RA works with the

neighbours. In order to avoid unnecessary,

redundant or useless communications and select the

most relevant neighbours, the RA applies criteria of

dependency of the query. For instance, if the

criterion is location, the network is composed of the

nearest neighbours; if user queries depend on her/his

previous queries, the RA must redirect them to the

most trusted neighbours; if the criterion is similarity,

the network could be composed of the neighbours

with a similar profile, tasks, etc. If no criteria are

established, the RA analyzes the trust level of these

neighbours. The RA associates a trust level to each

neighbour from answers to previous queries, based

on the work of (Agostini et al., 2004). In these

authors’ work, when a peer receives a query, it

analyzes the result of its queries and increases the

trust of those peers who reply with more appropriate

semantic contents. This work explains the process

for sending queries from a peer to other ones. After

query formulation, a peer named “seeker” checks

what peers are connected (“active”) and chooses,

among those connected, which peers send the query.

A query session is then opened to manage the

answer from each peer, or from peers that are

connected with the peer in charge of solving the

query. The strategy used by the seeker in order to

select a provider and referrals from providers to

solve the query is a Trust and Reputation Strategy.

The Trust and Reputation Strategy proposed by

(Agostini et al., 2004) consists of the following

process: the seeker faces the problem of selecting

which one among the peers is able to solve a query

Q with highest probability, or who makes the most

progress towards solving the query. To decide, the

seeker constructs and manages a list <p1, p2,…pk>

of trusted peers to which submit the query. The list

is conventionally ordered according to decreasing

trust level. The seeker strategy of query resolution is

the following: first, it asks p1, then p2, and

continues up to pk until it receives relevant answers

from previous peers in the list. It is important to note

that the list of trusted peers evolves with the system.

The Trust of a peer is acquired by its Reputation.

Second case, a query could be only answered by one

agent which is known. The RA uses its KB

(describing what are the WIS, their ISAs and their

managed information) to contact the WIS from

which it could obtain the answer to this query. This

is a specialization of the first case.

Third case, the query has been split in several sub-

queries in the analysis step. The RA analyzes which

agents can answer each one. The network of

neighbours is then composed by the agents which

could answer the sub-queries. The process applied in

WEBIST 2006 - INTERNET TECHNOLOGY

order to select information sources (ISAs) for each

sub-query is the same that the process defined in the

first case. Finally, the network of neighbours is

composed of the union of the different sub-networks

generated for each sub-query.

For the scenario, the RA could include in the

network of neighbours all ISAs executing on WIS of

the duty free shops which sell the products she/he

searches (based on fact (1) and its instances; see

section 3.1). The RA must also analyze the trust level

associated with these neighbours (e.g., the first shop

which answers). If it is the first time that the RA

executes this query or that works with these ISAs,

the RA sends the query to them through a broadcast

message. The RA must compose the network of

neighbours of the agents which could answer the

sub-queries of the query (“the closest shops to the

departure gate which sell the wanted article at the

lowest price”). In order to select the WIS of those

shops, the RA must apply criteria for the queries

(based on fact (3) and its instances; see section 3.2),

in this case, the proximity of the shops to the

departure gate and the lowest price for the articles.

For this case, the RA could store facts in its KB like

the ones presented below. These facts express that

all queries from passenger “John Smith” depends on

both his location, particularly if he is at the airport,

and, the proximity of the departure gate:

(assert (Criterion

(userID “John Smith”)

(criteria location)

(attributes “airport” )))

(assert (Criterion

(userID “John Smith”)

(criteria proximity)

(attributes “departure gate” )))

4.3 Redirecting the Query

Once the RA has identified potential information

sources (neighbours), it redirects the query, sending

a message which includes the query to its

neighbours. The RA can use an oriented message

(for specific receivers) or broadcast it to all

neighbours (e.g., waiting for the first one to reply,

obtaining all the answers and analyzing which are

the most trusted ones). If the RA has a trust schema

for the agents which compose the network of

neighbours, the RA could send the message in a

sequential way, starting with the most trusted one. If

it answers, the process is finished. Otherwise, the RA

has to continue sending messages until the least

trusted agent has been contacted, according to the

ideas of (Agostini et al., 2004).

For the scenario, the network is composed of ISAs

which execute on the WIS of the duty free shops. If

there is only one shop WIS, the RA sends it the

query. Otherwise, the RA sends the query to each

ISA, beginning with the most trusted one.

If the RA knows that neighbor1 can answer sub-

query1, neighbor2 can answer sub-query2 and so on,

it sends the oriented messages to each neighbour

(based on fact (1) about the WIS and its instances;

see section 3.1). For example, if the passenger would

like to know if her/his flight is on time, the RA sends

the query to the ISA which executes on the WIS of

the airline (for this example, we call it “OneISA”)

and to the ISA which executes on the WIS of the

airport and manages flight departure and arrival

schedules (for this example, we call it

“DAFlightsISA”). In this case, we can find in the KB

of the RA the following facts which allow it to

redirect the query to the OneISA and the

DAFlightsISA:

(assert (WIS

(name AirlineOneIS)(agentID OneISA)

(device server)

(info_items “departures” “arrivals”

“prices”)))

(assert (WIS

(name AirportIS)(agentID DAFlightsISA)

(device server)

(info_items “departures” “arrivals”)))

The RA must then compile answers obtained from

different agents and select the most relevant ones

according to the established dependency criteria.

The mechanisms for compiling results are not

explained in this paper.

5 RELATED WORKS

We present here some agent-based architectures or

frameworks for adapting information to users.

CONSORTS Architecture (Kurumatani, 2003) is

based on ubiquitous agents and designed for a

massive support of MDs. It detects user locations

and defines user profiles to adapt their information.

The CONSORTS architecture proposes a mechanism

to define the relations that hold between agents (e.g.,

communication, hierarchy, role definition), with the

purpose of satisfying user requests. Unlike PUMAS,

it does not consider distribution of information

between MDs (which could improve response time)

nor user preferences.

The work of (Gandon et al., 2004) proposes a

Semantic Web architecture for context-awareness

and privacy. This architecture supports automated

discovery and access of a user’s personal resources

KNOWLEDGE MANAGEMENT FOR ADAPTED INFORMATION RETRIEVAL IN UBIQUITOUS

ENVIRONMENTS

subject to user-specified privacy preferences.

Service invocation rules along with ontologies and

profiles enable identification of the most relevant

resources available to answer a query. However, it

does not consider that information which can answer

a query can be distributed between different sources.

The PIA-System (Albayrak et al., 2005) is an agent-

based personal information system for collecting,

filtering and integrating information at a common

point, offering access to information by WWW, e-

mail, SMS, MMS and J2ME clients. It combines

push and pull techniques in order to allow the user

on the one hand, to search explicitly for specific

information and, on the other hand, to be informed

automatically about relevant information. However,

the PIA System only searches information in text

format. It does not consider the adaptation of

different kinds of media to different MDs, nor user

location.

(Sashima et al., 2004) propose an agent-based

coordination framework for ubiquitous computing. It

coordinates services and devices to assist a particular

user in receiving a particular service in order to

maximize her/his satisfaction. This framework

chooses proper resources from numerous sources,

coordinates those resources on behalf of users and

assists them in accessing resources of ubiquitous

computing environments. These authors take into

account the contextual features of nomadic users,

especially, the location. Unlike PUMAS, this

framework does not consider the adaptation of

information according to the access devices nor the

possible distribution of data among devices.

6 CONCLUSIONS AND FUTURE

WORK

In this paper, we have described knowledge

managed and exchanged by the Information and the

Adaptation MAS of PUMAS to support the

adaptation capabilities and the Query Routing

process. PUMAS is a framework which retrieves

adapted information according to user profiles and

technical capabilities of MDs used to access the Web

Information Systems (WIS). We have also described

the strategies followed by the Router Agent to

perform the Query Routing process. In PUMAS, this

process is composed of three activities: analysis of

the query, selection of the information sources and

redirection of the query. Finally, we have presented

each activity and we have also illustrated them

through a scenario supported by a WIS in an airport.

We are currently implementing and testing each

MAS of PUMAS. For this purpose, we have chosen

JADE-LEAP (http://jade.tilab.com/), a FIPA

compliant platform. We intend to define, on the one

hand, algorithms for each activity of the Query

Routing process and, on the other hand, extensions

of an Agent Communication Language (ACL,

http://www.fipa.org/specs/fipa00061/SC00061G.htm

l) in order to consider nomadic user characteristics

like location and connection time. For this purpose,

we want to introduce in ACL, primitives like query-

when, query-where, query-close.

ACKNOWLEDGMENTS

The author Angela Carrillo-Ramos is partially

supported by the Universidad de los Andes

(Colombia). She thanks Nicolas Lopez-Giraldo for

his comments.

REFERENCES

Agostini, A., Moro, G., 2004. Identification of

Communities of Peers by Trust and Reputation. In

AIMSA 2004, 11th Int. Conf. in Artificial Intelligence:

Methodology, Systems, and Applications. LNCS,

3192, Springer Verlag, 85-95.

Albayrak, S., Wollny, S., Varone, N., Lommatzsch, A.,

Milosevic D., 2005. Agent Technology for

Personalized Information Filtering: The PIA-System.

In SAC 2005, 20th ACM Symposium on Applied

Computing. ACM Press, 54-59.

Carrillo-Ramos, A., Gensel, J., Villanova-Oliver, M.,

Martin, H., 2005a. PUMAS: a Framework based on

Ubiquitous Agents for Accessing Web Information

Systems through Mobile Devices. In SAC 2005, 20

ACM Symposium on Applied Computing. ACM Press,

1003-1008.

Carrillo-Ramos, A., Gensel, J., Villanova-Oliver, M.,

Martin, H., 2005b. A Peer Ubiquitous Multi-Agent

Framework for providing nomadic users with adapted

information. In AP2PC 2005, 4th Int. Workshop on

Agents and P2P Computing. Eds: Despotovic, Z.,

Joseph, S. and Sartori, C., 166-179.

El Fallah-Seghrouchni, A., Suna, A., 2004. CLAIM: A

Computational Language for Autonomous, Intelligent

and Mobile Agent. In PROMAS 2003, 1st Int.

Workshop on Programming Multi-agent Systems

Languages and Tools. LNAI, 3067, Springer-Verlag,

90-110.

Gandon, F., Sadeh, N., (2004, Oct). Semantic Web

Technologies to Reconcile Privacy and Context

Awareness. In Journal of Web Semantics, 1 (3).

Retrieved Nov. 9, 2005, from

http://www.websemanticsjournal.org/ps/pub/2004-17.

WEBIST 2006 - INTERNET TECHNOLOGY

Indulska, J., Robinson, R., Rakotonirainy, A., Henricksen

K., 2003. Experiences in Using CC/PP in Context-

Aware Systems. In MDM 2003, 4th Int. Conf. in

Mobile Data Management. LNCS, 2574, Springer-

Verlag, 247-261.

Kurumatani, K., 2003. Mass User Support by Social

Coordination among Citizen in a Real Environment. In

MAMUS 2003, Int. Workshop in Multi-Agent for Mass

User Support. LNAI, 3012, Springer-Verlag, 1–16.

Park, J., Barber, S., 2004. Finding Information Sources by

Model Sharing in Open Multi-Agent System. In

UbiAgents04, Workshop on Agents for Ubiquitous

Computing. Retrieved Nov. 9, 2005, from

http://www.ift.ulaval.ca/~mellouli/ubiagents04/

Pirker, M., Berger M., Watzke, M., 2004. An approach for

FIPA Agent Service Discovery in Mobile Ad Hoc

Environments. In UbiAgents04, Workshop on Agents

for Ubiquitous Computing. Retrieved Nov. 9, 2005,

from http://www.ift.ulaval.ca/~mellouli/ubiagents04/

Ramparany, F., Boissier, O., Brouchoud, H., 2003.

Cooperating Autonomous Smart Devices. In

sOc’2003, the Smart Objects Conference, 182-185.

Sashima, A., Izumi, N., Kurumatani, K., 2004. Bridging

Coordination Gaps between Devices, Services, and

Humans in Ubiquitous computing. In UbiAgents04,

Workshop on Agents for Ubiquitous Computing.

Retrieved Nov. 9, 2005, from

http://www.ift.ulaval.ca/~mellouli/ubiagents04/.

Xu, J., Lim, E., Ng, W.K., 1999. Cluster-Based Database

Selection Techniques for Routing Bibliographic

Queries. In DEXA 99, 10th Int. Conf. on Database and

Expert Systems Applications. LNCS, 1677, Springer-

Verlag, 100-109.

Yang, D., Xu, L., Cai, W., Zhou, S., Zhou, A., 2004.

Efficient Query Routing for XML Documents

Retrieval in Unstructured Peer to Peer Networks. In

APWeb 2004, 6th Asia Pacific Web Conf. LNCS,

3007, Springer-Verlag, 217-223.

KNOWLEDGE MANAGEMENT FOR ADAPTED INFORMATION RETRIEVAL IN UBIQUITOUS

ENVIRONMENTS