Context-Sensitive Conversation Patterns

for Agents in Wireless Environments

Santtu Toivonen

, Heikki Helin

Mikko Laukkanen

, and Tapio Pitk

aranta

VTT Information Technology

P.O.Box 1203, FIN-02044 VTT, FINLAND

TeliaSonera Finland

P.O.Box 970, FIN-00051 Sonera, FINLAND

Abstract. We outline an approach for adapting software agent conversations be-

tween a mobile client and a service provider. The adaptation takes advantage of

the context information characterizing the client’s current situation. The service

provider takes this context information into account when selecting / creating a

suitable interaction protocol for communicating with the client. Both the context

information and the interaction protocol descriptions are based on ontologies,

which are assumed to be known by the conversating agents.

1 Introduction

In ubiquitous environments, services can be accessed in various contexts. Context infor-

mation is anything that can be used to characterize the situation of an entity. An entity,

in turn, can be a person, place, or object that is considered relevant to the interaction

between a user and an application, including the user and application themselves [1].

Also details of the device used to access the service, and characteristics of the network

over which the services are accessed, are contextual information. In this paper, we focus

especially on wireless networks, and how they can be utilized when customizing agent-

based services for users. Particularly, network Quality of Service (QoS) is investigated.

When performing a handshake between a client and the service provider, context

information is taken into account before the actual interaction. Context information can

have inﬂuence on what services are provided to the client, or how a speciﬁc service

is tailored to best facilitate the interaction between the client and the service provider.

More speciﬁcally, context information may have impact on the chosen interaction pro-

tocol according to which the service is accessed. Interaction protocols set the rules that

govern the interaction [2] and are often used to steer software agent conversations. They

typically consist of ordered sets of messages, as well as restrictions on which participant

can send a message, of what type, to whom, and in what state of the protocol.

In [3], the authors discuss a bilateral negotiation protocol. Conversating agents can

utilize it for adapting to the quality of the network they are using. In contrast to their

approach, we make a distinction between the client and the service provider, and do not

Toivonen S., Helin H., Laukkanen M. and Pitkäranta T. (2004).

Context-Sensitive Conversation Patterns for Agents in Wireless Environments.

In Proceedings of the 1st International Workshop on Ubiquitous Computing, pages 11-17

DOI: 10.5220/0002676800110017

 SciTePress

Network

Class

WirelessNetwork

Class

WirelineNetwork

Class

disjointWith

subClassOf

GSM

Class

subClassOf

GPRS

Class

subClassOf

WLAN

Class

subClassOf

IEEE802.11a

Class

IEEE802.11b

Class

subClassOf

UMTS

Class

subClassOf

QoSProperty

Class

Service

Class

hasService

ObjectProperty

range

hasIdentifier

ObjectProperty

Operator

Class

operatedBy

ObjectProperty

domain

range

availableAt

ObjectProperty

domain

Location

Class

range

qosProperties

ObjectProperty

domain

Identifier

Class

range

domain

PacketSwitched

Class

CircuitSwitched

Class

subClassOf

WAN

Class

LAN

Class

PAN

Class

BlueTooth

Class

subClassOf

hasGPRS

ObjectProperty

range

subClassOf

Restriction

subClassOf

1

onProperty

maxCardinality

range

Fig. 1. The core concepts of the network ontology

design a framework where the agents could negotiate as equal peers about how the in-

teraction should take place. Instead, in our approach the service provider is responsible

for the selection / creation of the interaction protocol to be used.

We are unaware of any research in ubiquitous environments, which investigates the

relationships between context information and the selection / creation of interaction

protocols used to access services. We believe, however, that such relationships could be

found and explicated. One rule of thumb might be, that the longer round-trip times for

messages, the simpler interaction protocol should be. Another one could emphasize the

relationship with the device’s and operating system’s ability for parallel processing and

the number of concurrent protocols under execution.

Both context information and interaction protocol descriptions are speciﬁed accord-

ing to ontologies. The ontologies can be utilized for adaptation by both the service

provider and the client. The service provider’s agent can adapt the service based on the

context information received from the client. The client, in turn, can modify its behavior

to conform with the interaction protocol description received from the service provider.

An example on how the client-side adaptation to interaction protocol descriptions can

take place is described in [4].

In the next section, we brieﬂy address the ontologies that are relevant for adapt-

ing interaction protocol based services according to network characteristics and net-

work QoS attribute values. We then consider an example of such adaptation. Finally,

we present some concluding remarks and future directions.

2 Context Ontologies for Service Adaptation

This section summarizes three ontologies. These are intended for describing networks,

network QoS, and interaction protocols, respectively. These ontologies are needed in

order to perform the adaptation of services according to the context information char-

acterizing the client. There are other kinds of relevant contextual information that could

have impact on the service adaptation. Information about the person accessing the ser-

vice is one potential candidate. Information which is more permanent in nature, such as

gender and nationality, could result in completely different services provided to differ-

ent people. In addition, dynamic information such as social context and user’s current

mood, could be taken account when adapting the services. Properties of client device,

such as screen size, operating system, and available memory, could also have impact on

the adaptation [5]. However, in this but in this paper we focus on the network, especially

on its QoS.

Figure 1 concentrates on the wireless branch of the network ontology [6], which

is more important in typical ubiquitous scenarios, than the wireline branch. WIRE-

LESSNETWORK has several subclasses such as WLAN and UMTS. The notions of

CIRCUITSWITCHED and PACKETSWITCHED can be used to deﬁne the data connection

mode of the network. Furthermore, each network is available at some LOCATION. In

this paper, we do not consider how locations are expressed, but assume that there is

some other ontology, which can be used to describe such issues.

Figure 2 depicts a network QoS ontology, which is based on the FIPA Quality of

Service Ontology [7]. It is contains for example the following properties. LINERAT E

is used to express the bandwidth in one direction over a measured component. DELAY

is the (nominal) time required for a data segment to be transmitted to a peer entity.

RTT, round-trip time, is the time required for a data segment to be transmitted to a

peer entity and a corresponding acknowledgment sent back to the originating entity.

FRAMEERRORRATE is the probability, that a data segment is not transmitted correctly

over a measured component.

Interaction protocol ontology [8] plays a different role in the service adaptation than

the previous two ontologies. The network ontology and the network QoS ontology were

used to describe the context information, whereas the interaction protocol ontology

is used to describe how the services are accessed. The offered interaction protocol is

selected / created based on the contextual details.

Figure 3 depicts the concepts forming the core of the interaction protocol ontology.

INTERACTIONPROTOCOL concept contains general information about the interaction

protocol, such as information about the participating agents. Every protocol has one

INITIATOR and one or more PARTICIPA N Ts.

Every interaction protocol has exactly one PROGRESS component that deﬁnes pro-

gression of the interaction protocol. PROGRESS consists of STATEs that constitute the

ﬂow of the protocol. STATEs can have decision points called CHOICEs that in turn have

OPTIONs to choose from. Only one OPTION can be chosen in a given STATE. A MES-

SAG E can be sent or received as a consequence of choosing an option in a particular

state. In the core set of the ontology, the message only contains an identiﬁer and the

sender and receiver information. However, it can be extended with concepts that enable

QoSProperty

Class

LineRate

Class

subClassOf

Delay

Class

subClassOf

Throughput

Class

subClassOf

value

DataTypeProperty

Direction

Class

direction

ObjectProperty

Collection

oneOf

InBound OutBound

range

xsd:nonNegativeInteger

range

RateValue

Class

hasRateUnit

ObjectProperty

hasRateValue

ObjectProperty

range

RateUnit

Class

range

Collection

oneOf

bits/s

kbits/s

mbits/s

Restriction

onProperty

subClassOf

1

cardinality

Restriction

subClassOf

onProperty

1

cardinality

Restriction

onProperty

subClassOf

1cardinality

TimeValue

Class

hasTimeUnit

ObjectProperty

TimeUnit

Class

range

Collection

oneOf

ms

s

m

h

Restriction

subClassOf

1

cardinality

onProperty

Restriction

subClassOf

1

cardinality

onProperty

Restriction

subClassOf

1

onProperty

maxCardinality

ProbabilityValue

Class

Restriction

subClassOfRestriction

subClassOf

1

maxCardinality

onProperty

RTT

Class

subClassOf

MeanUpTime

Class

subClassOf

ConnSetupDelay

Class

subClassOf

TimeQoSProperty

Class

subClassOf

hasTimeValue

ObjectProperty

domain

range

RateQoSProperty

Class

subClassOf

domain

ProbQoSProperty

Class

subClassOf

hasProbValue

ObjectProperty

domain

range

BER

Class

subClassOf

OmissionRate

Class

subClassOf

FrameErrorRate

Class

subClassOf

ConnSetupFailureProb

Class

subClassOf

1

cardinality

Fig. 2. The network QoS ontology

more detailed descriptions. For example, in [8], we introduced a FIPA extension to the

ontology.

3 An Adaptation Example

We now describe one way of adapting a service according to network conditions. Lo-

cation of a mobile user changes as he moves around. By having location information

of the user, the service provider can conclude where the user is heading and how fast.

These information may be useful for the service provider when adapting services [9,

10]. Knowledge of the direction, for example, can result in the system providing the

user with information related to things that are ahead of user, rather than behind [11].

InteractionProtocol

Class

Agent

Class

Initiator

Class

Participant

subClassOf

Class

State

Class

Progress

Restriction

subClassOf

minCardinality

ObjectProperty

hasState

onProperty

range

Restriction

subClassOf

cardinality

ObjectProperty

hasProgress

onProperty

Restriction

subClassOf

cardinality

ObjectProperty

hasInitiator

onProperty

Restriction

subClassOf

minCardinality

ObjectProperty

hasParticipant

onProperty

range

ObjectProperty

follows

Class

PredThing

range

sendBy

ObjectProperty

recvBy

ObjectProperty

range

choices

ObjectProperty

range

hasTimeout

ObjectProperty

TimeOut

Class

range

Message

Class

Exception

Class

subClassOf

hasMessage

ObjectProperty

range

Option

Class

subClassOf

domain

domaindomain

domain

Fig. 3. The core concepts of the interaction protocol ontology

Knowledge of speed and direction of the user can be utilized also when specifying

the interaction protocol for accessing the services. The service provider can estimate, for

example, how soon the user is expected to be out of network coverage. This estimation

can in part determine the chosen interaction protocol. If the service provider estimates

that the user is expected to be out of network coverage soon, it tries to provide the

user with an interaction protocol description with less roundtrips. Instead, if the user

is expected to stay longer within the network coverage area, the service provider can

provide him with a more sophisticated version of the interaction protocol. This kind of

reasoning could, for example, result in providing different interaction protocols to users

on foot than to users on a tram or driving a car.

Should the service provider have the user’s current network and QoS information,

it can perform further reasoning besides that based on the speed and direction. For

example, if the expected round-trip time is short enough, the service provider can decide

to go forward with a protocol having more round-trips, even if the user is soon heading

out of the network coverage area. With a potential for more round-trips, the service

provider can engage a conversation trying to ﬁnd out the user’s preferences and interest

of the services it is providing.

Furthermore, should there be multiple networks to choose from, the service provider

can select the most suitable one to best facilitate the user. The service provider can also

perform handovers, for example from WLAN to GPRS networks and back [12]. While

in WLAN coverage area, the service provider and the client can engage a conversa-

tion along a more sophisticated interaction protocol, and shift to a more stripped-down

version when the user is heading out of the WLAN hotspot. In addition to mere net-

work types, the service provider could also take advantage of the real-time QoS values,

as discussed in [13]. For example, should the user enter a crowded WLAN hotspot,

continuing the conversation with GPRS connection can prove to be more appropriate.

4 Conclusions and Future Work

We presented ontologies for expressing wireless networks, network QoS, and inter-

action protocols. Such ontologies can in part enable service adaptation according to

context information in ubiquitous environments. We outlined one adaptation example.

Our future work around the area includes considering further properties of networks

and how they could have impact on the chosen interaction protocol. In addition, we are

going to consider device details and user preferences as context information. In this pa-

per we focused on interaction protocols and did not consider individual messages. Fur-

ther work could be done on adjusting message contents and sizes based on the context

information. Such message level adaptation could also be integrated with the interaction

protocol level adaptation.

References

1. Moran, T., Dourish, P.: Special issue on context-aware computing. Human Computer Inter-

action (2001)

2. Jennings, N., Faratin, P., Lomuscio, A., Parsons, S., Sierra, C., Wooldridge, M.: Automated

negotiation: Prospects, methods and challenges. International Journal of Group Decision and

Negotiation 10 (2002) 199–215

3. Paurobally, S., P.J., Turner, Jennings, N.: Automating negotiation for m-services. IEEE

Transactions on Systems, Man and Cybernetics, Part A 33 (2003) 709–724

4. Toivonen, S., Helin, H., Min, J., Pitk

aranta, T.: Using interaction protocols in distributed

construction processes. In: Proceedings of the 6th International Conference on Enterprise

Information Systems (ICEIS 2004), Porto, Portugal (2004)

5. Foundation for Intelligent Physical Agents: FIPA Device Ontology Speciﬁcation, Geneva,

Switzerland. (2000) Speciﬁcation number SI00091.

6. Helin, H., Laukkanen, M.: Wireless network ontology. In: Proceedings of the 9th Meeting

of the Wireless World Research Forum (WWRF), Zurich, Switzerland (2003)

7. Foundation for Intelligent Physical Agents: FIPA Quality of Service Ontology Speciﬁcation,

Geneva, Switzerland. (2000) Speciﬁcation number SC00094.

8. Toivonen, S., Helin, H.: Representing interaction protocols in DAML. In van Elst, L.,

Dignum, V., Abecker, A., eds.: Agent-Mediated Knowledge Management: Selected Papers

from AAAI 2003 Spring Symposium. Volume 2926 of Lecture Notes in Artiﬁcial Intelli-

gence., Berlin, Springer (2004) 310–321

9. Vidales, P., Chakravorty, R., Patanapongpibul, L.: Ubiquitous networking in heterogeneous

environments. In: Proceedings of the 8th IEEE Mobile Multimedia Communications (Mo-

MuC 2003), Munich, Germany (2003)

10. Baus, J., Ding, Y., Kray, C., Walther, U.: Towards adaptive location-aware mobile assis-

tants. In: Proceedings of the IJCAI Workshop on Artiﬁcial Intelligence in Mobile Systems

(AIMS’01), Seattle, USA (2001)

11. Jones, G., Brown, P.: Challenges and opportunities for context-aware retrieval on mobile

devices. In: Proceedings of the SIGIR workshop on Mobile Personal Information Retrieval,

Tampere, Finland, ACM (2002) 47–56

12. Pesola, J., Ponkanen, S.: Location-aided handover in heterogeneous wireless networks. In:

Proceedings of the Mobile Location Workshop (MLW’03), Aalborg, Denmark (2003)

13. Helin, H., Misikangas, P.: Supporting seamless roaming in FIPA nomadic application ar-

chitecture. In: Proceedings of the IJCAI-01 Workshop on Artiﬁcial Intelligence in Mobile

Systems (AIMS’01), Seattle, USA (2001) 25–30