AN AGENT-BASED APPROACH TO SUPPORT PERFORMANCE

MANAGEMENT FOR DYNAMIC AND

COLLABORATIVE WORK

Nora Houari

and Behrouz H. Far

Department of Electrical and Computer Engineering, University of Calgary, Calgary, Alberta, Canada

Keywords: Multiagent support systems, modelling, roles, integration, collaboration, coalition.

Abstract: This paper describes a comprehensive multiagent based modelling approach for collaborative and dynamic

organizational roles support. The method is a role centred one, where agents collaborate to assist human

users through collaboration within the same role, with other roles of the same team, as well as roles of

different teams that share tasks dependencies. Agents in the system are not restricted by predefined schemes

they can join and/or leave the coalition. We identify the key elements of the role model as rules, agents, and

relationships. Our role model integrates both the operational functionalities and the performance

management towards specific goals, the former involve human individuals in the loop, whereas the later are

performed by software agents to assist in monitoring, control, and adaptation of the performance in dynamic

organizations.

1 INTRODUCTION

Software (agent-based) information and support

systems have a significant influence on the product,

and services, an organization offers. Organizations

increasingly rely on computational systems, these

systems need to be designed, constructed, and

integrated as such to support the feature that justifies

this reliance. At the same time, they need to reveal

the changes that occur in the environment, such as

operational tasks changes, and individuals’

reallocations. Hence, this appeal for an explicit

examination of knowledge which is generally

implicit, for example, who is doing what with

whom, etc.

In this paper, we propose a unified and adaptable

organizational model for integrating the operational

functionality activities and the performance towards

specific goals, as well as their dynamics. The agent

assistance mechanisms described in this research

provides proactive monitoring and control, in

addition to keeping track of current events (e.g. task

milestones delayed), the agents can automatically

join and/or leave their role coalition. Such explicit

exploration of this knowledge supply human

individuals with an actual view of their progress and

allow them take appropriate actions accordingly.

The contribution of this work is threefold: first,

we provide an integrated structure of a role that

enables us to combine operational activities with

performance indicators, second we develop a formal

framework for supporting individuals in

organization, and third, we outline the process of our

approach.

The technical focus of this paper is on

developing organizational modelling mechanisms

for collaboration and coalition dynamics. Although

the presented computational support system was

built for software engineering organization, we

believe that the method is general enough to be

applied to any collaborative environment involving

distributed tasks and teams.

This paper is structured as follows: Section 2

briefly discusses the problem and a case study. In

Section 3, we illustrate the specification of the role

concept in our modelling approach. Section 4

describes our organizational formal model, and in

Section 5 we delineate the model process steps.

Section 6 summaries related works, and finally

conclusions and perspectives of future works are

provided in Section 7.

178

Houari N. and H. Far B. (2007).

AN AGENT-BASED APPROACH TO SUPPORT PERFORMANCE MANAGEMENT FOR DYNAMIC AND COLLABORATIVE WORK.

In Proceedings of the Ninth International Conference on Enterprise Information Systems - AIDSS, pages 178-184

DOI: 10.5220/0002411801780184

 SciTePress

Figure 1: The dimensions of our model.

2 PROBLEM STATEMENT

The application case study considered in this paper

is a software engineering organization dealing with

the development of software projects development.

Software projects are usually multi-team and multi

stage development. They can belong to one

organisation or be shared among partners. Due to the

wide range of distributed data, information and

knowledge, it is impossible to effectively locate,

keep track, and adapt to changes acquired in these

projects. Individuals in these organizations are

involved in organizational and projects tasks, they

may be reassigned tasks (added and/or substituted),

the design and implementation of their support

system must reveal these dynamics, and be able to

provide awareness about who is doing what when

and who is interacting and collaborating with whom.

The aim of this work is to provide a support

system that assists human users performing their

roles by taking care of the performance management

of their roles execution. Since the processing

capabilities of human are limited; the distribution of

information, resources, and activities among

organization members must be set up to guarantee

that management of the tasks of each individual

remains below his/her capability. By delegating

performance management to agents we are (semi)-

liberating human individuals from routine tasks and

consenting them to focus on processing and

execution of the task.

Figure 1 depicts our organizational model

dimensions that contain: the organization and its

policy, roles, agents, their relationships and

dynamics.

3 THE ROLE SPECIFICATION

We define a role as “a collection of duties and

rights” (Biddle and Thomas, 1966). Duties represent

the tasks and interactions that the role is obligated to

perform, whereas rights represent the permissions to

utilize information entities to perform tasks or

interactions. Thus, the concept of role in our system

is essentially an abstraction on one hand, for the

tasks that are necessary to be performed and/or the

interactions that need to occur with other roles to

achieve an individual goal, and on the other hand,

the information that needs to be accessed or will be

generated during the course of performance of those

tasks/interactions, and the knowledge that is needed

for the successful execution of tasks and interactions

towards achievement of the goal.

We identified two types of roles: an

organizational abstract role and a conceptual role.

An abstract role refers to operational

functionality tasks and their strategic performance

goal; by operational we mean processing and

communication towards the execution of the tasks,

strategic performance on the other hand deals with

measuring the performance of the goal. In our

system, an abstract role is assigned to one or more

agent (human, software), according to their

capabilities, an agent may carry more than one

abstract role. Figure 2 delineates the capabilities of

an organizational abstract role. Any abstract role

requires two types of capabilities: (i) the processing

capabilities theses are execution, communication,

decision making, etc. (ii) and strategic capability

these are the performance management of time,

product/service quality, cost, effort etc.

The second type of role we introduced in our

model is a conceptual role, which describes a set of

capabilities and responsibilities necessary to perform

an activity. A conceptual role, often involves

relationships (i.e., collaboration, conflict resolution,

etc) among abstract roles.

The main idea of our work is to integrate both

capabilities of an abstract role in a coherent and

unified form (see Figure 3). Our technical approach

to this integration is based on building the

collaboration and coalition of agents for a dynamic

conceptual role (activity). The objective here is that

human will be able to assign the performance

management of their activities to software agents so

that they can fully focus their cognitive capabilities

on processing, decision making, etc. Our approach

allows an effective shift and delegation of

responsibilities to agents playing an abstract role as

well as the dynamics in a conceptual role.

AN AGENT-BASED APPROACH TO SUPPORT PERFORMANCE MANAGEMENT FOR DYNAMIC AND

COLLABORATIVE WORK

179

Figure 2: A Dynamic abstract role.

Figure 3: The components of an abstract role.

4 THE FORMAL MODEL

In order to implement our approach that integrates

organizational processing functionalities and

performance management in a dynamic

environment, we first identified key elements such

as agents, collaboration, environment and goals in a

formal model. Based on this formalism, we highlight

the essence of our multiagent support system (in

terms of its components and its process). Our

proposed role model is a tuple < Rules, Agents,

Relationships>. In the next subsections we review

briefly these components.

4.1 Rules

Organizational rules are used to constrain one or

more of the following: (i) the assignment of agents

to roles, (ii) role’s capabilities requirement, (iii)

role’s relationships, and (iv) role's goal achieve.

These rules must satisfy the accuracy and

functionality requirement, and often are application

and domain dependent.

4.2 Agents

Software agents are computational entities that run

in an environment, sense it and act upon it to achieve

a specific goal. Each agent must possess the required

capabilities to play a particular role. We identify

two categories of agents the operational and the

strategic agents. Operational agents deal with

processing and communication to execute the tasks,

in our current version of the system they are carried

by human individual; however he/she can delegate

those routine tasks to software agents as in

(Chalupsky et al., 2005). Strategic agents, on the

other hand are an added value to the operational

agents to keep an eye on the mission goal

performance, this category of agent uses metrics for

each specific goal.

4.3 Relationships

Relationships refer to instance interaction

relationships, the dynamics of activities, the roles,

and agents performing them towards achieving a

mission goal. In our system we consider the

collaboration for activities among team members,

and the dynamic in joining and leaving the activity.

5 THE MODEL PROCESS STEPS

The process for building our support system consists

of the following steps:

1. Identify the performance indicators (PIs) for a

particular goal; for each operational task we identify

the performance metrics towards a particular goal.

2. Assign agents: define the required agents to

support each task, and their grouping for an abstract

role.

3. Identify the type of interactions: Interactions are

constrained by organization law, the application

specific process, and individual privacy.

4. Build a collaboration algorithm between teams’

members leading to accomplishment of the goal.

5. Construct the coalition method that automatically

manages the performance in the case of an agent

leaving or joining the coalition.

Because of the limitation of the paper length, details

of some steps are provided in (

Houari and Far,. 2007).

The individual process steps are briefly summarised

in the subsequent sub-sections.

ICEIS 2007 - International Conference on Enterprise Information Systems

180

5.1 Performance Indicators

Identification

For any given task, we identify the key performance

indicators (PIs) to accomplish a particular goal.

Details of an example of metrics for achieving the

goal of delivering work on time were presented in

(Houari and Far, 2007).

5.2 Assignment of Agents

We identify two categories of agents to support

operational functionalities. The first category

support the execution, and the second the

communication. In this work we consider only the

formal. The later (communication) was addressed by

works like (Chalupsky et al., 2005). Future

extension of this work will integrate both categories

of support. For an abstract role we assigned five

support agents: H

: Human Proxy; R

: Role

Scheduler; R

: Role Bookkeeper; R

: Role product

Deliverable; R

: Role Cost. There is also an

additional agent support for the team leader: Team

Leader Proxy (H

TLP

This set of agents capture the information of all

the tasks assigned to the role. We set the human

proxy agent to consolidate the other agents of the

role, and allow it as the only agent to talks to the

human user.

5.3 Identify the Interactions

In our system, we identified three types of

interactions (see Table 2 and Figure 4).

Table 2: Agent interaction types.

Between Definition

1 H

∝ H

Interaction between human proxy

agent and the human individual it

supports

2 R

∝ R

Interaction among agents that

support the same human

individual

∝

Interaction between one human

proxy and another human proxy

: Human Proxy

: Role Scheduler

: Role Bookkeeper

: Role Deliverable

: Role Cost

: Human Proxy

: Role Scheduler

: Role Bookkeeper

: Role Deliverable

: Role Cost

Figure 4: The types of agent interactions.

5.4 Building the Collaboration

The set of an individual’s assigned project and

organization tasks are specified by P

respectively. Team’s project and team’s organization

tasks are denoted by P

and O

correspondingly.

The members of the team are T

and task’s

dependencies function is defined by tdf. Each

individual is assisted by five agents: the proxy,

scheduler, bookkeeper, product deliverable, and cost

) respectively (step 2). These

agents are assigned the individual’s organization and

project tasks, and their dependencies (steps3). For

every task (project or organization) we allow the

interaction among the supporting agents of the

individual role according to the interaction types

identified in section 5.3 (steps4-5). If the task has no

dependencies we add the collaboration with the team

leader proxy (step 6). If the task has dependencies

(i.e. carried by more than one individual) we allow

interaction between proxy agents of the individuals

sharing the task as well as between the agents of the

teams’ leaders (step7). The next level of

collaboration, deals with the team leader (step 11-

14), where he is assisted by a proxy, scheduler,

product deliverable, and cost support agents

(step12). The team’s project and organization tasks

of the team members are assigned to the supporting

agents of the team leader.

Figure 5 outlines the pseudo code of the

collaboration of agents within the role and between

roles.

Figure 5: The pseudo code of the collaboration.

Algorithm 1: The Collaboration Algorithm

Inputs:P

..t

]; O

..t

]; P

..t

];

..t

];T

..H

]; tdf

Outputs: Human’s (H

) support agents’

collaboration binding

1.for each individual

2.H

:=<H

Pa;

3.<H

Pa;

>:=[P

Ta,

tdf]

4.for

∀

[ t

, t

]

∈

, O

5. (H

∞

)

∧

∞

)

6. if(t

∃

tdf) then (H

∞T

LPa

)

7. else(H

∞H

)

∧

∞H

)

∧

∞R

)

∧

∞T

LPa

)

∧

LPa

∞ T

LPk

)

8. end if

9. go to 11

10.end for

11.for

∀

[ t

, t

]

∈

< P

, O

12.H

:=< H

TLP;

13.<H

TLP;

> := [P

TT,

]

14. end for

AN AGENT-BASED APPROACH TO SUPPORT PERFORMANCE MANAGEMENT FOR DYNAMIC AND

COLLABORATIVE WORK

181

5.5 The Coalition Dynamics

The last step in the process of our computational

support system consists of building the conceptual

role coalition dynamics mechanism. Our approach in

building the dynamic coalition is rule driven; getting

agents involved in tasks triggers automatic activity

coalition creation, deletion or integration. Automatic

coalition management enhances the dynamics of

collaboration: agents are not restricted by predefined

schemes and they can join or leave the coalition.

5.5.1 Formal Notation

A role in our computational system is a task

centred one. A task represents anything which has a

role involvement, attributes, properties and role

means of interactions. Our conceptual role coalition

is founded on the notion of activity that

associates a goal with a set of individuals and with a

set of tasks. According to the following semantics;

the individuals involved in the activity have a

common goal to achieve through the

accomplishment of the tasks of the activity.

Performing a task requires a binding to it. A

conceptual role coalition dynamically binds a set of

tasks and agents from a given activity. This binding

is normalized by rules specified bellow and managed

by an algorithm illustrated through an example in

the next subsection. An agent a assigned to activity

V performs a task t from V, if and only if, he has a

binding on t.

We call c

(A, T) a conceptual role coalition of

the activity V binding a set of agents A to a set of

tasks T from V, such that every agent of A is

assigned to a task from T. The conceptual role

coalition combines the collaborations of the agents

with the tasks. A given coalition shows who is

sharing tasks with whom, and which tasks are

shared. The conceptual role coalition is based on the

following rules:

Rule 1: The set A of agents of the conceptual

role is maximum that is A contains all the members

of the activity bound on any of the tasks of the

activity.

Rule 2: The set T of tasks of the conceptual role

is maximum that is T contains all the tasks of the

activity that the members of the conceptual role are

sharing (a task is shared if and only if, more than

one binding exist)

To ensure rules 1 and 2, we define the following

operation.

Operation: Two conceptual roles can be merged

if and only if they have the same agent members or

they have the same tasks.

The result of merging two conceptual role

coalition c

(A;T)and c

(A’;T’) is an other

conceptual role c

A’; T

T’).

For example c

({a

};{t

}) is

mergeable with c

({a

};{t

}) and the result

is c

({a

};{t

}).Similarly

({a

};{t

}) and c

({a

};{t

})

are mergeable in c

({a

}; {t

The binding of agents and tasks control the

evolution of the coalition set of an activity but must

guarantee rules 1 and 2.

5.5.2 Illustrative Example

We demonstrate the coalition dynamic method

through an example, in this example an activity V

has four agent members: a

, a

and

that

are responsible for t

, t

and t

in this

way:

−

, a

and a

are collaborating on the tasks

, t

, and t

−

and a

are collaborating on t

−

is active on t

This is represented by three conceptual role

coalitions

({a

};{t

});c

({a

};{t

}) and c

({

};{t

})(see Figure

4(a)).

5.5.3 The Agent Coalition Binding Scheme

If the agent a

will be involved in task t

, he has

to bind to t

then the conceptual role coalition

involving t

must show that a

joins the group of

members responsible for t

, so a

will be added to

which becomes c

({a

};{t

}).

However this will break the rule 2, because

conceptual role coalitions c

and c

have the

same members. The two conceptual role coalitions

have to be merged; and the coalition set becomes:

({a

};{t

, t

}); and

({a

};{t

})(see Figure 4(b)).

If the agent a

becomes involved on t

then a

has to be added to coalition c

but this will violate

rule 1 because a

is not bound to t

, t

Therefore the coalition c

will be split into c’

({a

, a

};{t

, t

} and

c’’

({a

, a

};{t

}). These two

new conceptual role coalitions are not mergeable

with any of the other coalitions of the activity, so the

ICEIS 2007 - International Conference on Enterprise Information Systems

182

resulting coalition set is: c’

({a

};{t

, t

} and c’’

({a

};{ t

({

};{t

})}.(see Figure 4(c)).

We construct the conceptual role join coalition

algorithm as fellow:

Let (m,t) be the new bindings, in order to

ensure rules 1 and 2, we present the algorithm in two

steps. The first step consist in finding the conceptual

role coalition c

involving task t, if such coalition

exists, and then adding the member m to c

member

set. The second step is the merge of the coalitions.

We arrange the coalitions in three sets:

ToChange: includes the coalition that will be

modified.

Mergeable: contains the possible mergeable

coalitions: it is composed of the remaining coalitions

having agent member m as a member.

UnChanged: contains the coalitions that will

be left unmodified, it is composed of the remaining

coalitions.

Figure 6: Conceptual role coalition dynamics.

5.5.4 The Agent Coalition Release Scheme

When the agent leave the coalition and release the

binding on task t, this must be reflected in the

relevant conceptual role coalition. The first step is to

find the relevant coalition that is finding the

coalition involving the task t. The second step is to

remove the agent from the list of the agent of this

conceptual role coalition, and to guarantee that the

resulting coalition does still correspond to the

coalition definition and still respects rules 1 and 2.

This restriction will imply a reorganization of the

conceptual role coalition. Using the previous

example, we had the following coalition set:

{c’

({a

};{t

}),

c’’

({a

, a

};{t

}), c

({

};{t

})}(See Figure 4(c)).

If agent a

releases the task t

, it has to be

removed from the member associated with t

c’

. Rule 2 implies splitting the coalition into c

({a

};{t

}) and c

({a

};{t

}) which leads the conceptual role

coalition set to be partitioned as follows:

ToChange: Is the set which if not empty,

includes the coalition that will certainly be modified.

Mergeable: contains the coalition that can be

modified to ensure rule1.

UnChanged: contains the coalitions that will

be left unmodified, it is composed of the remaining

coalitions.

6 RELATED WORK

There have been several attempts at formalizing the

concepts of teamwork within organization using

multiagent systems: works of (Chalupsky et al.,

2005), (Sycara et al., 2003) and (Kogut et al., 2004)

propose systems and architectures to assist

organizational human teams in routine tasks, such as

meeting scheduling. In contrast to our approach,

these systems deal with only the communication

support part of the task performance, and ignore the

processing part. We believe that our approach is

more general since it integrates both support for the

processing and communication required in

performing any task. In addition none of these

efforts have included techniques dealing with the

different kind of the possible interactions and

collaborations of team members, and their dynamics.

Some other works such as (DeLoach and

Matson, 2004), (Krauth et al., 2005) and (Xu and

Zhang, 2005) have looked at modelling organization

using the notion of agents roles, goals and

indicators; in distinction to our work, they do not

capture the binding between the human and his/her

tasks, these models fail to take into account the

human perspective, thus is human in the loop, as it is

a key element in any concrete application.

AN AGENT-BASED APPROACH TO SUPPORT PERFORMANCE MANAGEMENT FOR DYNAMIC AND

COLLABORATIVE WORK

183

7 CONCLUSIONS AND FUTURE

WORKS

In this paper, we described a multiagent modelling

approach for collaborative and dynamic

organizational roles support. The method is a role

centric, where agents collaborate to assist human

users by performing collaboration within the same

role, between other roles of the same team, as well

as roles of different teams that share tasks

dependencies. We consider also the dynamics within

the role; that is, agents joining and or leaving the

coalitions. Ongoing work deals with simulation of

the model using the third generation distributed

dynamic decision making DDD-III (Simulator,

2006) for time performance. Future work will

consider incorporating several other performance

indicators such as performance indicators for

quality, effort, cost, etc.

ACKNOWLEDGEMENTS

The Authors would like to acknowledge the

clarifications and suggestions made by Dr. Joerg

Denzinger on early versions of this work. We would

also like to thanks Dr Kevin M. Gildea and Dr.

Georgiy Levchuk of Aptima, Inc. for providing us

with the DDD-III simulator to validate the model.

REFERENCES

Biddle, B.J. and Thomas, E.J., 1966. Role Theory:

Concepts and Research, John Wiley & Son, Inc, 1966.

Krauth, E., Moonen, H. Popova, V. and Schut, M., 2005.

“Agent-Based Modeling of Third-Party Logistics

Performance Indicators”, Intelligent Agent

Technology IEEE/WIC/ACM International (IAT

2005) pp. 262-265, 2005.

Chalupsky, H. Gil, Y. Knoblock, C. A K. Lerman, Oh, J.

Pynadath, D. V., Russ, T. A and Tambe, M. 2005.:

“Electric Elves: Applying Agent Technology to

Support Human Organizations”, IAAI 2001, pp 51-58,

2005.

Xu, H. and Zhang, X. 2005. “A Methodology for Role-

Based Modeling of Open Multi-Agent Software

Systems”, 7

International Conference on Enterprise

Information Systems (ICEIS 3, 2005), pp. 246-253,

2005.

Sycara, K.P. Paolucci, M.,. Velsen, M. V, and Giampapa,

J.A. 2003 “The RETSINA MAS Infrastructure”,

Autonomous Agents and Multi-Agent Systems 7 (1-2),

pp 29-48, 2003.

Houari, N. and Far, B.H. 2007. “Improving Project

Performance with Role Driven Computational Support

System”, IEEE Canadian Conference on Electrical and

Computer Engineering (CCECE 2007), April 2007.

Houari, N. and Far, B.H. 2007. “A multiagent Coalition

Support Tool for Collaborative and Dynamic Roles”,

IEEE Canadian Conference on Electrical and

Computer Engineering (CCECE 2007), April 2007.

Kogut, P.A. Yen, J., Leung, Y., Sun, S, Wang, R.

Mielczarek, T. and Hellar, B., 2004 “Proactive

Information Gathering For Homeland Security

Teams”, Commun. ACM 47(3), pp 48-50, 2004.

DeLoach, S.A., and. Matson, E, 2004 “An Organizational

Model for Designing Adaptive Multiagent Systems”.

The AAAI-04 Workshop on Agent Organizations:

Theory and Practice (AOTP), 2004.

Simulator, 2006. “The DDD-III: Distributed Dynamic

Decision-making” Simulator Provided by Aptima, Inc,

2006.

ICEIS 2007 - International Conference on Enterprise Information Systems

184