Multi-agent WWW - based Environment for Distributed Team Decision Making Support

Alexander V. Smirnov, Mikhail P. Pashkin, Irina O. Rakhmanova

St.-Petersburg Institute for Informatics and Automation of the Russian Academy of Sciences

39, 14th line, St. Petersburg, 199178, Russia
Tel.: +7-(812)-218-8071
Fax.: +7-(812)-218-0685

Abstract. This paper is devoted to one of the main problems of collaborative engineering - decision making in distributed team. It describes main functions of multi-agent WWW-based environment WWW-MULTIEXPERT: communication functions of information exchange, co-ordination functions of timely informing the leader about changes in individual preferences of the experts, etc.

Keywords: decision support systems, agent, WWW, environment, distributed system


New market opportunities demand constant increase of product quality and decrease of its cost in rapidly changing environment. This trend has become more prevalent during 90's. In fact, manufacturing companies need to change conventional images and introduce new strategic objectives and tools. In order to cope with these new paradigms, companies need to deeply transform both their product development structure and the structure of business processes. In the recent decade a big number of engineering and manufacturing concepts were developed. Some of them are: Concurrent Engineering, Collaborative Engineering, Co-operative Engineering, Agile Manufacturing, Customer Drive Manufacturing, World Class Manufacturing, Virtual Enterprises, etc. And one of the main problems of this concepts is distributed team decision making. Computer technologies which can realise interaction of the members of working group: co-ordination and communication under consider of this paper.


Multi-agent Technology

Recently one of roughly developing information technologies, ensuring co-ordination and communication in distributed teams, is multi-agent systems (MAS) technology. WWW and MAS are in wide use now. In this article discussed one of the approaches of work organisation in distributed team using these notions.

Intelligent agent is an autonomous software entities that can navigate heterogeneous computing environment and can, either alone or working with other agents, achieve some goals [1]. It have to enjoy the following properties [2]:

Multi-agent system technology can be considered as the basis for decision support system. WWW technology and intelligent agents are widely used in Co-operative Engineering and Electronic Commerce. Intelligent agents also used for manufacturing management.

Fig. 1 presents one of the classifications of the software agents [1]. Allocate the following agents:

Fig. 1. Classification of the software agents

The computer decision support system should provide simultaneous work of the several users with one or several projects. It is possible that the experts involved in the distributed team territorially removed from each other. This fact requires special functions in the system for work in distributed multi-user environment support, such as:

The Environment Development

Conventional decision making general scheme consists of the following steps:

The WWW-MULTIEXPERT system supports all this steps in some extend.

The first version of MULTIEXPERT supported group expert work in the local network. This environment was not neither distributed nor multi-agent system. This system was developed using Microsoft FoxPro 2.6 and it was completed multi-user environment [3].

But at present a set of specialists from different parts of the world are involved into decision making processes. Hence, the system have to support decision making process in the distributed computer network. In this cases there are a lot of functions have to be added into the environment. Main of them - post functions for information exchange. Also in the distributed environment the members of process have to spend own time to additional work. The leader have to check readiness of each expert in every task, the experts have to look for new messages for them, and there are a lot of other problems in this type of work. To avoid this problems on the second step of system developing we propose a multi-agent environment for decision making system in the distributed network. On this step of system development Leader Agent and Expert Agent were developed. They were facilitators and carried out only post functions. In more detail algorithms and the functions of these agents are described in [4]. The basic lack of the given realisation of system - necessity of the special software installation on the expert computer and consequently, dependence on a platform, on which user works.

On these reasons the basic principles of the third stage of development of multi-agent multi-user environment were developed. In this version the information exchange bases on Internet and TCP/IP protocol. New type of agent appeared and a set of function of Leader and Expert agent was extended. The Leader and Expert Agents decomposed in to levels: the expert level and the communication layer. The first layer is responsible for agent activity and the communication level assures the transparency of agents exchange [5]. At present the system under construction. This approach allows to organise the work in system independently on the platform on which the expert works. WWW - browser is a software on the expert computer. Using it, the experts addresses to the leader's WWW - server. During dialogue they fill special HTML forms and results of their work are processes on the Leader's database server. For an information exchange with the users, which have not an access to Internet the post-packet technology is still used. For organisation of work inside a local network Intranet technology is used.

So, the power of the system consecutively growth. WWW-MULTIEXPERT system relates with other Microsoft Office products. Using ODBC drivers leader can import new projects from other systems and return results of the work (weights of projects). For computation of dependent complex attributes the experts can use Microsoft Excel. And for result visualisation Microsoft Graph is used.

WWW-MULTIEXPERT Environment Structure

The structure of multi-agent environment supports an expert group co-ordinated work for quality evaluation of organisational and technical decisions. It consists of three types of agents:

In dependence on information exchange this agents performs a different functions. Using agent classification we relate our agents to the next type of agents: Expert and Coordinating agents are regular agents. They react to environment conditions and always know what to do. Leader Agent is a planning agent. On some steps of decision making it have build-in algorithm for decision making.

Describes main functions of entity agents:

Leader Agent:

  1. Provides for a set of procedures, to facilitate the leader's work in distributed system;
  2. Executes an automation procedure of attributes ranking;
  3. Makes conformation control of decisions;
  4. Provides tools for visualisation of the expert group results. The Leader can see group decision or every local decision by graphic, table, excel - sheet, etc;
  5. Prepares messages for the Coordinating Agent;
  6. Executes post functions in distributed system:

Coordinating Agent

  1. Provides the step by step decision making algorithm:
  2. Supports common databases integrity on the group level, makes all necessary changes there;
  3. Prepares dialogue forms for information exchange by Internet.

Expert Agent

  1. Supports a step execution of the current task:
  2. Prepares messages for the Coordinating Agent;
  3. Supports integrity of local databases;
  4. Executes the post functions in the distributed environment:

Fig. 2 demonstrates the step by step decision making algorithm of attributes ranking using entered agents. The leader operates with the following data for formation of the task: by the attribute templates, containing parameters qualities and their values; by knowledge of the system (lists of the experts and tasks existing in system, decisions, which it is necessary to consider). The task became accessible for proceeding when the leader prepares a list of attributes for evaluation. Coordinating Agent receives incoming message (Minp) about task readiness for processing. It defines and prepares necessary information, for expert local databases. Using internal set of functions (F) it prepares messages for Expert Agent. When the expert registers in the system his (or her) agent checks existing tasks (analyses incoming messages Mij, where j is a unique number of the expert and i is a number of the process step) and prepares corresponding screen forms. The expert works (do a set of actions Fj). Expert Agent increments readiness degree of the

Fig. 2. Agent interaction general structure

task on the local level when the expert finishes (transition S from readiness degree N into readiness degree N+1). It sends a message (Moj) to Coordinating Agent. The Coordinating Agent analyses whether the similar messages on readiness from all other members of group have come. If someone is not ready there is a wait of whole group readiness. The readiness of the task has incrementing when everybody are ready (transition S from readiness degree N into readiness degree N+1). As a rule, after a stage of opinions making by each expert a stage of the conformation follows. The Coordinating Agent sends the message (Mout) to the Leader Agent that a stage of the conformation follows. Depending on the coordination type - automatic control (by the Coordinating Agent) or personal control (by the leader), it is possible two ways of the result proceeding:

The appropriate decision is sent to the Coordinating Agent, which advances the task per unit of further, or returns the experts on the previous stage. The experts use local data, which are representation of data, prepared by the leader, and each agent at local and group levels works with the own internal representations.

An Automation Procedure of Attributes Ranking

As a rule, it is too difficult for all experts to elaborate the same solution. For this reason the partially conformed private opinions are used for group decision making.

To facility the leader work Leader Agent has an internal procedure Auto_Range for automatic attributes ranking. It receives the following set of parameters:

  1. Number_Of_Steps - number of iteration steps. If this parameter is equal to 0 the decision making process may be limited only by time.
  2. End_Time - date and time of the process termination. If this parameter is omitted the decision making process may be limited only by the number of iteration steps.
  3. Array_Of_Attributes - a set of E attributes. The experts have to evaluate this attributes not leaving the reassigned range. If this list is empty, the experts have to work out full conformed solution. This set is a list of couples - an internal number of attribute and a value in percents of the maximum difference of the private evaluation from the group evaluation. Any of attribute may be included in this list.
  4. Type_Of_Action - a kind of the Leader Agent's action in the case when a group decision is not formed. If the group decision is not formed to the End_Time or during Number_Of_Steps Leader Agent makes a list of the experts, whose private opinions differ from the group opinion. According to the value of this parameter, Leader Agent may performs one of the following actions:

After finishing this procedure the leader receives a report about results of the group working. This report is an electronic diary. It includes:

  1. Number of the iterations steps;
  2. Time of the group work on the project and time of each expert work on the project;
  3. list of discharged experts;
  4. on every iteration step a list of the experts, whose private opinion differ from the group opinion, and list of attributes with differences.

The leader can print this report, view on the screen and save it in Excel - table for future work.

At the local level the experts rank the attributes and send the results of their job to Leader Agent. Auto_Range works according to the following algorithm:

  1. Forms a group decision.
  2. Checks a contents of Array_Of_Attributes.
  3. If this list is empty then it compares private and group opinions. If any of attributes difference value is more than 5% the group solutions is not completed. In other case the group decision is fully conformed.
  4. If Array_Of_Attributes is not empty it compares only attributes from this set. If any of attributes difference value is more than given percent then the group solution is not completed. In other case the group solution is E - conformed.
  5. During attributes difference value comparison Leader Agent forms a list of the experts, whose private opinion differ from the group opinion.
  6. If a group opinion was not formed, then number of iterations increments and compares with Number_Of_Steps.
  7. If a group opinion was formed, Leader Agent sends a message to Coordinating Agent. Coordinating Agent makes a special changes in project bases and changes the degree of the task readiness. Coordinating Agent sends to the Leader Agent a special information about projects. In this case Leader Agent makes actions according to the forth parameter of the procedure.


Using multi-agent WWW-based group decision support system reduces time of decision making and increases quality of organisational and technical decisions evaluation and choice.

A rapid developing of Internet-based technologies with steadily increasing easiness in accessing any kind information through World Wide Web allow to add mobile agents into WWW-MULTIWEXPERT system for quick search an information related to discussed problems. In the future we are going to use the multimedia tools for visualization of results.


  1. S.Franklin, Is it an agent, or just a Program?: A Taxonomy for Autonomous Agents, in: Proceedings of the Third International workshop on Agent Theories, Architectures, and Languages, Springer-Verlag (1996).
  2. M.J.Wooldridge, N.R.Jennings, Agent Theories, Architecture, and languages: A survey, in: Intelligent Agents, Proceedings of the Workshop on Agents Theories, Architecture, and languages (ECAI-94), Springer-Verlag (1995).
  3. M.Pashkin, Some Special Features of Data Proceeding in Distributed Computer Systems, in: Information Technologies and Intelligent Methods, St.-Petersburg: SPIIRAS, Russia, (1995) (in Russian).
  4. M.Pashkin, A.Smirnov, I.Rakhmanova, Multiexpert: an Internet bases support environment for solution evaluation, in: Proceedings of the Special Day on pan-European Co-operation and Technology Transfer, Zakopane, Poland (1996).
  5. C.Sandru, V.Negru, A Distributed Task-Oriented Problem Solving System, in: Proceedings of the International Workshop «Distributed Artificial Intelligence and Multi-Agent Systems», St.Petersburg, Russia (1997).

BENEFIT Special Report Main Page