A network expert system management system for multiple domains

Journal of Information Science http://jis.sagepub.com

A network expert system management system for multiple domains Brenda L. Killingsworth, Michael B. Hayden and Robert Schellenberger Journal of Information Science 2001; 27; 81 DOI: 10.1177/016555150102700203 The online version of this article can be found at: http://jis.sagepub.com/cgi/content/abstract/27/2/81

Published by: http://www.sagepublications.com

On behalf of:

Chartered Institute of Library and Information Professionals

Additional services and information for Journal of Information Science can be found at: Email Alerts: http://jis.sagepub.com/cgi/alerts Subscriptions: http://jis.sagepub.com/subscriptions Reprints: http://www.sagepub.com/journalsReprints.nav Permissions: http://www.sagepub.com/journalsPermissions.nav

Downloaded from http://jis.sagepub.com at PENNSYLVANIA STATE UNIV on April 12, 2008 © 2001 Chartered Institute of Library and Information Professionals. All rights reserved. Not for commercial use or unauthorized distribution.

The effect of postings information on searching behaviour

1 2 3 4 5 6 7 8 9 1110 1 2 3 4 5 6 7 8 9 20 1 2 113 4 5 6 7 8 9 30 1 2 3 4 5 6 7 8 9 40 1 2 3 4 5 6 7 8 9 50 1 2

A network expert system management system for multiple domains Brenda L. Killingsworth East Carolina University, USA

Michael B. Hayden Rhode Island College, USA

Robert Schellenberger

thus making it suitable for use in self-directed team environments. Furthermore, these concepts also provide support for dynamic organizational change. An illustration is provided to show how such a network of expert systems can assist a manager attempting to identify principal labour needs within an organization.

1. Introduction

East Carolina University, USA Received 29 July 2000 Revised 4 December 2000

Abstract. In an increasingly competitive global business environment, the fast pace of technological change, widely dispersed expertise and strong competition demands proactive decision-making by today’s corporations. Global information management has become increasingly important, as businesses strive to leverage their information resources for competitive advantage. Innovative expert systems that enhance and speed up the decision-making process are one avenue for improving an organization’s competitive position. However, most expert systems are designed as standalone products and, as such, they do not reflect actual decision-making within an organization. This paper discusses a design that extends the concept of a hierarchical expert system management system into a network of expert systems from multiple domains. This model is designed to bring the best source of knowledge to the decision-making process. Further, it is intended to use these resources to quickly make adjustments to the rapidly changing business environment. The model is less restrictive, Correspondence to: B.L. Killingsworth, East Carolina University, School of Business, General Classroom Building 3134, Greenville, NC 27858, USA. Tel: 1 252 328 6893. Fax: 1 252 328 4092. E-mail: [email protected]

Organizations face a number of trends that demand more responsive and comprehensive information systems and expert systems. Among these are: ● a global economy; ● an increased use of technology; ● improved organizational effectiveness; and ● delegation of more responsibilities to self-directed work groups. These trends have been in part responsible for the growth of the American economy. Much of this growth was due to improvements in productivity. The global business environment requires the decision maker to have quick access to the knowledge necessary to respond effectively to these trends. Innovative expert systems that can meet these challenges are one avenue for improving an organization’s competitive position in the marketplace. However, while most expert systems are designed as stand-alone products, organizations are moving towards teamoriented decision-making as a means of increasing worker productivity. Therefore, most expert systems (which are designed to emulate human decisionmaking) do not accurately reflect organizational decision-making processes. This poses a number of challenges for the management of information and knowledge within an organization. A number of models have been proposed that link expert systems. Recent research has been conducted on integrating corporate databases with expert systems [1, 2, 3]. There have also been attempts to resolve multiple

Journal of Information Science, 27 (2) 2001, pp. 81–92 Downloaded from http://jis.sagepub.com at PENNSYLVANIA STATE UNIV on April 12, 2008 © 2001 Chartered Institute of Library and Information Professionals. All rights reserved. Not for commercial use or unauthorized distribution.

81

A network expert system management system for multiple domains

conflicting expert strategies within a specific knowledge domain [4, 5, 6]. However, these attempts have only addressed problems associated with the management of knowledge among experts within a given domain. They have not addressed problems directly related to sharing knowledge among experts in multiple domains. Team orientation adds impetus to the need for a network of expert systems that can support knowledge from multiple domains. By modeling manual procedures in the form of expert systems that are networked to each other, and to the corporate database, the network can be used to better emulate organizational decision-making. There have been proposals for a hierarchical expert systems management system (ESMS) for multiple domains [7]. However, a hierarchical ESMS is restrictive and reflects the hierarchical decision-making found in traditional management approaches. As such, a hierarchical ESMS only supports Hackathorn and Keen’s [8] personal and organizational categories of decision making. However, this paper extends the idea of a hierarchical ESMS by generalizing the framework to include a network of expert systems that supports team decision-making. This design allows for knowledge from multiple domains within a network and so supports decisions requiring knowledge from multiple experts. The design limits knowledge only by the number of experts within the system and it supports unstructured decisions that might require unanticipated links. The design is modular and permits expansion of the network as necessary. Confidentiality of knowledge is both recognized and supported. The structure of the paper is as follows. Section 2 provides the foundation for the conceptual model by reviewing research on decision making within organizations and expert system design issues. Section 3 proposes a framework for viewing how expert systems ‘fit’ within an organization’s information system structure. Section 4 presents the design of a network ESMS that addresses knowledge search strategies necessary for implementing a network of expert systems. Section 4 also illustrates how a network ESMS can support an organization’s management of labour needs. Section 5 offers final remarks and discusses directions for further research.

2. Literature review A number of areas must be addressed prior to the development of a framework for classifying expert systems within an organization. These areas include decision 82

making within an organization, current knowledge base principles and the designs and limitations of previous expert systems. This section discusses these issues and provides a foundation for the conceptual framework of the system. 2.1. Decision making in an organization Before developing a framework that assists in understanding how expert systems can be developed and networked, it is necessary to first understand the implications of organizational decision-making. Decision making within an organization can be categorized in several ways: the hierarchy of decision making, the degree of structuredness of the decision-making task and the nature of decision support. One early classification scheme of decision making was presented by Anthony [9]. Anthony classified decisions according to the type of decision required: strategic, tactical or operational in character. Strategic decisions are concerned with establishing goals, policies and guidelines for an organization. They are characterized as non-routine, semi-structured decisions. Strategic decisions require primarily external data, have an uncertain outcome, are long-ranged in focus and are performed on an infrequent basis. Tactical decisions are concerned with implementing policies established by strategic planners. Tactical decisions are characterized as routine, structured to semi-structured decisions that require primarily internal data with a limited amount of external data. The decisions contain an element of uncertainty, are mid-ranged in focus and are performed on a frequent basis. Operational decisions are concerned with the daily operations of an organization. Operational decisions are characterized as routine, structured decisions that primarily require internal data, are relatively certain in nature, are shortranged in focus and are performed on a frequent basis. Simon [10] developed another classification scheme of decision making, based on the structuredness of a decision. In this classification scheme, decisions can be categorized as structured, semi-structured and unstructured. Simon [11] also later introduced the concept of programmed and non-programmed decisions. Programmed decisions are those described by a clear set of rules, making it possible to replace the judgement of a human decision-maker with an automated, computer-based system. Non-programmed decisions are typically unstructured and require the use of heuristics to arrive at a solution. These kinds of decisions have complex problem spaces that are not easily describable. Journal of Information Science, 27 (2) 2001, pp. 81–92


111 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 20 1 2 3 4 5 6 7 8 9 30 1 2 3 4 5 6 7 8 9 40 1 2 3 4 5 6 7 8 9 50 1 2

BRENDA L. KILLINGSWORTH, MICHAEL B. HAYDEN AND ROBERT SCHELLENBERGER

111 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 1120 1 2 3 4 5 6 7 8 9 1130 1 2 3 4 5 6 7 8 9 1140 1 2 3 4 5 6 7 8 9 50 1 112

Hackathorn and Keen [8] classified decisions based on the type of interaction required to produce a solution. Three types of decision support include personal support, group support and organizational support. Personal support is directed at a user or class of users in addressing a specific decision task, such as selecting a stock, which is relatively independent of other tasks. No interaction with domain experts is required to produce a solution. Group support is directed at a group of individuals participating in separate, but highly interrelated tasks (e.g. strategic planning or selfdirected team decision-making). This level of support implies that a pool of experts must work together concurrently to produce a solution. Organizational support is directed at a task or activity composed of a sequence of operations that require multiple experts (e.g. capital budgeting). This type of interaction requires a sequential transfer of knowledge or information between experts to produce a solution to a problem. 2.2. Knowledge base design principles This section discusses the two design principles that will be incorporated into the conceptual framework of the ESMS. These topics deal with a networked versus hierarchical organization of knowledge and with the design of functionally adjacent expert systems. 2.2.1. Network versus hierarchical organization of knowledge A hierarchical organization of knowledge is often used to handle multiple strategies within a given expert system (see Fig. 1). For example, the Hearsay-II system [12] is built around the notion of a committee with members and a chairman. The ‘members’ generate several suggestions in the form of hypotheses and then the ‘chairman’ of the committee evaluates and selects which hypothesis will be ‘pursued’. Such a system is an example of a ‘blackboard-based’ expert system (i.e. a system that adopts a blackboard-based model of problem solving). The blackboard model partitions the domain knowledge into separate knowledge bases, with each having its own inference engine. This reduces the cohesion between components, allowing for a more flexible and dynamic behaviour. Many applications of blackboard models are within inherently hierarchical domains. The OPIS system [13] implemented a hierarchical structure for a scheduling system that incorporated two types of knowledge systems: ‘manager’ knowledge

System

Knowledge Management

OPIS

Hierarchical

Graph

(Manager and Scheduler)

Hearsay

Committee (Chairperson and Members)

ESMS

Self-Directed Teams (Facilitator and Members)

Fig. 1. Knowledge management approaches.

systems and ‘scheduler’ knowledge systems. Manager knowledge systems were responsible for directing, controlling and coordinating the problem-solving activity of subordinate knowledge systems. Scheduler knowledge systems were responsible for carrying out the manager knowledge systems directives. This was done by making decisions that extend the partial solution to a problem or by collecting information about the state of the problem. The manager knowledge systems are concerned only with the level of knowledge systems directly below them. This approach facilitates the design and maintenance of large systems. The proposed extension of the hierarchical ESMS provides an organization of knowledge that includes support for the concept of self-directed teams composed of expert systems. It is based on the features of the Hearsay-II system, in that it is built around the notion of a committee using a ‘facilitator’ (parent node) and a blackboard for recording the hypotheses of ‘team members’. It differs, in that the facilitator serves as a problem-solving manager of the team. The facilitator is treated as an equal member of the team, with the team making the final decision (rather than a chairperson). The hierarchy of teams is modeled around the organizational concept of a core team of executives and subordinate teams, each containing team members and a facilitator. The members generate several suggestions in the form of hypotheses. The facilitator records all of these hypotheses on a blackboard and then requests a rank order of the hypotheses from the team members. The committee members evaluate and then select


83


which hypotheses will be ‘pursued’. Selection is based on a given set of criteria established by teams and/or facilitators higher up in the hierarchy. The facilitator serves the team in the role of directing, controlling and coordinating the problem-solving activities of the team members. The core team members (or top hierarchical team) serve as coaches for subordinate teams. They are responsible for providing criteria to be used by subordinate teams in selecting hypotheses to pursue. Similarly, a team on a lower level can establish criteria for subordinates directly below it, assuming the criteria do not conflict with the ones established by its own superiors. 2.2.2. Functionally adjacent expert systems Functionally adjacent management information systems often exist within an organization. Such systems share common data elements, generate data necessary for each other to function, and have functional overlap [14]. If one views the organization as a whole, and recognizes that knowledge is rarely used for just one specific purpose, then the importance of managing knowledge stored in separate knowledge bases is apparent. Expert systems usage should also be evaluated based on this concept. Decision support system researchers Hackathorn and Keen [8] recognized that decision support should be provided for personal decision-making, group decision-making and/or organizational decision-making. Personal decision support offers guidance for specific tasks that do not cross task boundaries. Group decision support coordinates the decision-making activities of several decision-makers working on a given problem at the same time. Organizational decision support offers assistance for decision making that requires knowledge generated throughout an organization on a sequential basis. Expert system researchers should also attempt to develop methods for coordinating decision-making activities. This may be accomplished by ensuring that, for each of these decision support categories, knowledge can be shared between different expert systems. 2.3. Previous designs to integrate expert systems Most expert systems have been developed as standalone systems. Such systems are not able to share knowledge or make decisions that require multiple domain expertise. However, many decisions often require the knowledge of multiple experts to produce a

84

solution. Recently, research has begun to address these limitations by integrating and networking expert systems. Application areas span a wide range of interests, including manufacturing machine control [15], handwriting recognition [16], forestry resource management [17] and spacecraft control systems [18]. This section discusses its use in the more traditional corporate decision-making process. 2.3.1. Multiple experts within a given domain There have been several conceptual frameworks proposed to network multiple expert systems. These frameworks have ranged from supporting fixed links between specific functional areas [19] to more general frameworks that permit flexible access to multiple expert systems [6, 12, 20, 21, 22]. 2.3.1.1. Supporting multiple experts with fixed-link protocols A number of models (and systems) have been developed that utilize multiple expert systems with fixed communication links. For example, Digital Equipment Corporation implemented a network of expert systems [19] for the customer-sales-manufacturing-distributioncustomer cycle, where the interfaces between expert systems within the network are fixed. Jacob and Pirkul [6] designed a networking framework based on communication models found in office information systems research. In this model, knowledge outside a given expert system’s domain is identified as ‘peripheral’ knowledge. The primary limitation of this model is that predefined communication protocols must be established for communication to be permitted within the network. If a link has not been predefined, then an expert cannot access that knowledge. The network cannot handle situations not anticipated in the original network design. Both Ho, Hong and Kuo’s [23] society model and Woo and Lochovsky’s [24] model focus ‘on problem solving within an office, and models the office as a society of interacting agents’ or objects. The models indicate that an agent (or object) may need personal domain knowledge, as well as knowledge belonging to other agents (or objects). The pre-specification of links also makes these models inflexible for unstructured decisions that might require unanticipated links. Interestingly, there has been renewed interest in the concept of cooperating ‘interacting agents’ in the context of the Internet and distributed objects.


111 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 20 1 2 3 4 5 6 7 8 9 30 1 2 3 4 5 6 7 8 9 40 1 2 3 4 5 6 7 8 9 50 1 2


111 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 1120 1 2 3 4 5 6 7 8 9 1130 1 2 3 4 5 6 7 8 9 1140 1 2 3 4 5 6 7 8 9 50 1 112

2.3.1.2. Supporting multiple experts with flexible links An early general framework for communicating between experts was the contract net [20, 21]. In a contract net, the node requiring information would broadcast its need and the other nodes in the network would communicate back to it their ability to supply the needed information. The main limitation to the contract net framework is its inefficiency. A node must evaluate all other nodes’ abilities before entering into a contract with the node that can best provide the information it needs. The Hearsay-II system [12, 21] is an expert system that is designed to understand speech. Its design incorporated the concept of multiple cooperating experts using a blackboard for communication within the network. A primary limitation with this design is that it assumes that all knowledge is communal and that there is no issue of privacy. This paper offers a network design that permits sequential or group access to any expert system’s knowledge within the network, unless it has been identified as ‘sensitive and unavailable’. This design expands the domain of knowledge available to a given expert to include the whole network of experts. It does not restrict access to predefined peripheral knowledge or specific domain knowledge. With these extensions, the design will support each category of Hackathorn and Keen’s decision classification scheme.

3. Conceptual framework A number of conceptual frameworks have been developed and used within the realm of information systems research [14]. There is a benefit to research continuity if the same framework can be applied across research topics. In fact, the degree of robustness is a critical factor in the applicability of a given framework. A generally well-accepted framework is the Gorry and Scott-Morton [25] framework (see [26] for validation of this framework). 3.1. Applicability of the Gorry and Scott-Morton framework to expert system research The Gorry and Scott-Morton framework model is based upon two dimensions of decision making. The first dimension is the level of the decision [9]: operational, tactical or strategic. The second dimension is the type of decision as identified by Simon [10]: structured, semistructured or unstructured. Several authors have since

identified the areas or cells within the Gorry and ScottMorton model that are applicable to electronic data processing (EDP) systems, management information systems (MIS) and decision support systems (DSS) [14]. 3.2. Applicability of MIS, DSS and expert systems to the Gorry and Scott-Morton framework Operational structured decisions typically fall within the realm of EDP. The primary reason for this is that EDP is mostly transaction-based processing. Tools that are common to these types of decisions are decision trees, decision tables and simplistic expert systems. Expert systems that are used in this capacity are highly structured, with very narrowly defined problem domains. In reality, the use of expert systems within this capacity is often considered an inefficient use of resources. An MIS system is typically considered for use with semi-structured tactical decisions, often under the domain of middle management. The information used is typically derived from EDP systems. It is further summarized and processed to provide information to support tactical decisions. Some more complex expert systems are used for tactical decision-making, as are DSS. Expert systems are considered to be a subordinate of DSS. Therefore, it would only be logical to identify positions of expert systems within the realm of DSS positioning within the Gorry and Scott-Morton framework. DSS are usually thought of as providing the greatest aid to strategic decision-makers, especially if the decision maker is operating with unstructured decisions. It is this area that provides the greatest potential for the use of expert systems. However, the biggest problem with the application of expert systems to this area is the degrees of interaction between multiple domains and functional areas. A third dimension, involving degrees of interaction, must be added to the Gorry and Scott-Morton framework to properly place all types of information systems. Hackathorn and Keen’s [8] decision interaction classification scheme provides a vehicle for defining this third dimension (see Fig. 2). A logical, though complex, alternative to traditional DSS and expert system design would allow for interaction between multiple domains and functional areas. The design should provide support for multiple domains, multiple sources of data, a high degree of uncertainty and a high degree of integration between function units (both internal and external). Such a system would require multiple expert systems and would require some method of managing these experts. What is needed is an expert system management system.


85


Type of Decision

Degree of Interaction

Degree of Structure

Fig. 2. A framework for classifying expert systems.

4. ESMS design This section discusses the design of the proposed network ESMS. Discussion begins with a description of the system components and basic assumptions of the network model. This is followed by a description of the method of knowledge representation and methods used to traverse the network. A concluding example illustrates use of the network to identify labour needs within an organization.

(4) An expert node can communicate with any other node in the network via direct access links (i.e. with its parent (facilitator) and its siblings) or via other expert systems acting as facilitators. (5) The interface between two experts is assumed to be standardized and consistent (i.e. common language to eliminate ambiguities, e.g. homonyms, synonyms, etc). (6) An expert node, if it lacks expertise in the area questioned and consults its neighbours, trusts its neighbours completely. (7) A request may be split into sub-requests if it requires the knowledge from a union of domains. (8) An expert node will always be available for group decision-making. (9) Each facilitator node is also linked to every other node, but these links are activated only when required in a specific team setting. The traversal benefits of a traditional hierarchy exist with the flexibility to assemble or disassemble teams and thus permit direct communication between experts when decisions require a team effort. (10) The link between a facilitator node and sibling node is bi-directional. 4.2. Example of ESMS

4.1. System components and assumptions The ESMS model is based upon an organized network of expert systems. Each node in the network is an expert system and the nodes are connected via links. At the top of the network is a root node with meta-knowledge of all domains within the hierarchy. Remaining nodes are referred to as sibling nodes. Communication between nodes is accomplished via the traversal of activated links. All other links in the network are inactive until invited into a team session by a facilitator node. The network organization is therefore based upon a hierarchical overlay of activated links that represent the traditional flow of decision making within an organization. Facilitator nodes represent organizational decision-making in a team setting. This conceptual framework for the network ESMS is based upon the following ten assumptions. (1) The network is noise-free. (2) The domain of each node (expert system) is independent of other domains within the network, except for functional dependency [27] with ancestral domains (i.e. no redundancy and no inconsistencies). (3) A node contains meta-knowledge about the domains of its sub-tree of siblings. 86

Fig. 3 provides an illustration of the ESMS in the context of an organizational setting. The expert system nodes are as follows: the Chief Executive Officer (CEO) expert is the root node and has domain knowledge in strategic planning. The CEO’s siblings include experts in administration, production and marketing. Administration’s expertise is in financial analysis and human resources. The administration node has two siblings: a finance director and a human resource manager. The finance director’s expertise is in financial analysis and projecting cash flow. The human resource manager’s expertise is in managing the human relations aspects of the company. Production’s expertise is in facility planning. The production node has four siblings: a receiving manager, a processing manager, a final inspection manager and a post-production manager. The receiving manager’s expertise is in controlling and ensuring traceability of a product with respect to the raw materials. The processing manager’s expertise is in assessing and managing production rates and rejection levels. Marketing’s expertise is in sales forecasting. The marketing node has two siblings: an advertising manager and a sales manager. The advertising manager’s expertise is in tracking and assessing consumer preferences. The sales Journal of Information Science, 27 (2) 2001, pp. 81–92


111 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 20 1 2 3 4 5 6 7 8 9 30 1 2 3 4 5 6 7 8 9 40 1 2 3 4 5 6 7 8 9 50 1 2


111 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 1120 1 2 3 4 5 6 7 8 9 1130 1 2 3 4 5 6 7 8 9 1140 1 2 3 4 5 6 7 8 9 50 1 112

Inactive Link

CEO

Active Link

Root Node

M

Administration

Production

Marketing

E1

E2

E3

Finance E4

Human Resource E5

Receiving E6

Processing E7

Inspection E8

Post Production E9

Advertising E10

Sales E11

Fig. 3. A network ESMS (with hierarchical overlay of activated links).

manager’s expertise is in evaluating the sales force performance. In Fig. 3, solid lines indicate activated links and inactive links are indicated by dotted lines. To facilitate readability, only one facilitator (production) is linked. The marketing, administration and CEO nodes would also have dotted lines linking them to the other nodes in the network.

knowledge. In Fig. 3, there are three siblings: administration, production and marketing. The ‘parent’ slot is a pointer to the parent node, which, in the case of a master frame, is null. ‘Activity table’ is a procedure that updates two tables used to track the requests/responses that are sent/received over the network via packets. A request packet is represented as:

4.3. Knowledge representation

Request Packet {Qm,Tm,N,Qm,s,Tm,s,[Pe1,e2,..en]}

A frame-based representation of knowledge [28] was chosen, because frames permit a modular specification of knowledge. Ideally, frames are designed to represent objects, actions or events (procedures can be linked to the frame itself). A special type of frame, called the master frame, is associated with the root node. A master frame differs from any other frame in the system, in that it does not have a parent. Fig. 4 illustrates the frame-based description of knowledge in the master frame for the root node of the last example. Fig. 5 depicts the same for a sibling node. In each of these frames, ‘self-knowledge’ represents the domain of knowledge of the expert in question. As listed in the assumptions, all domains of knowledge within the hierarchy are independent. ‘Sibling knowledge’ refers to the domain of knowledge of the node’s siblings and can be incorporated as a facet of domain

Where: ● Qm represents the main question sent; ● Tm represents the type of request: organizational support, group support or combined. Combined indicates that a team of experts will convene once organizational support knowledge has been received by one or more experts; ● N represents the number of question subcomponents; ● Qm,s represents a specific question subcomponent; ● Tm,s represents the type of request (organizational support or group support) of a specific question subcomponent. This parameter indicates the responding expert system’s role (organizational support or group support). Its inclusion allows for cases in which a request requires organizational support from some experts and group support from others;


87


Frame:

Knowledge in Master Frame (for M)

Own Slot:

CEO activities in managing environmental issues, policy issues, etc.

Self Knowledge:

Strategic planning knowledge

Sibling Knowledge 1:

Administration, node E1


Production, node E2


Marketing, node E3

Parent:

Null

Activity Table:

(Request Slot, Response Slot)

Fig. 4. Sample master frame.

Frame:

Knowledge in Sibling Frame (for E2)

Own Slot:

Production

Self Knowledge:

Production Management knowledge


Receiving, node E6


Processing, node E7


Inspection, node E8


Post Production, node E9

Parent:

CEO, node M

Activity Table:

(Request Slot, Response Slot)

Fig. 5. Sample sibling frame.

PQ[e1,e2,..en] represents the path of the expert nodes queried. A response packet is represented as:

●

●

Response Packet {Qm,Tm,Qm,s,Tm,s,Rm,s,PR[e1,e2,..en]}

●

Where: Qm represents the main question sent; ● Tm represents the type of request (organizational support or group support) of the main question sent; ● Qm,s represents a specific question subcomponent; ● Tm,s represents the type of request (organizational support or group support) of a specific question subcomponent; ●

88

Rm,s represents the response (i.e. ability to participate in the case of group support or the knowledge passed back for organization support) to a specific question subcomponent, Qm,s; PR[e1,e2,..en] represents the path of the expert nodes sending or forwarding a response.

4.4. Traversal through the ESMS Consider a request made of a node in the network. If the node’s domain of knowledge is not appropriate, the request is transferred to the facilitator node. If the request is for organizational support, the facilitator searches its domain, and its siblings’ domains, to route the request to an appropriate expert node. If it is Journal of Information Science, 27 (2) 2001, pp. 81–92


111 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 20 1 2 3 4 5 6 7 8 9 30 1 2 3 4 5 6 7 8 9 40 1 2 3 4 5 6 7 8 9 50 1 2


111 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 1120 1 2 3 4 5 6 7 8 9 1130 1 2 3 4 5 6 7 8 9 1140 1 2 3 4 5 6 7 8 9 50 1 112

unsuccessful in identifying a node that can supply the needed expertise, the request is transferred to the next level up and that node searches its domain and the meta-knowledge of its siblings’ domains. If the request reaches the root node, and the request cannot be answered within the domain of the ESMS, then the root will declare the request as illegitimate and send a response back to the source. If the request is legitimate, then the root node transfers the search to another subtree until an appropriate expert is found to respond. Once a response is established, the path stored in a request packet is reversed and stored in a response packet. This response packet is sent back using the reversed path. If the request is for group support, the facilitator searches its domain and its siblings’ domains to identify the experts needed on the ‘team’. The facilitator then routes the request for a team meeting to the appropriate experts. If it is unsuccessful in identifying a node that can supply the needed expertise, the request is transferred to the next level up and that node searches its domain and the meta-knowledge of its siblings’ domains. If the request reaches the root node and the request cannot be answered within the domain of the ESMS, then the root will declare the request as illegitimate and send a response back to the source. If the request is legitimate, then the root node transfers the search to another subtree until an appropriate expert is found to participate. Once a participant is found, then the path stored in a request packet is reversed and stored in a response packet. This response packet is sent back using the reversed path. Once the facilitator receives responses from all participants, then it ‘calls a meeting’ by activating the appropriate links to establish a group decision-making session (or new team setting). This approach permits the modeling of a dynamic organization of teams, rather than the static set of teams found in traditional hierarchical models (i.e. parent and siblings). 4.4.1. Illustration of traversal through the ESMS For illustration, let us assume that the production node (E2) must identify labour needs for the next quarter. To accomplish this, the production expert node determines that its request should be disaggregated and sent to each of its siblings. This is done to obtain knowledge about the labour market and to query an expert that understands labour availability. Node E2 sends a request to its sibling nodes E6, E7, E8 and E9 to participate in a team meeting. The objective of the meeting is to identify the top labour needs for the next quarter.

Additionally, node E2 sends a request to the root node (its facilitator) for the identification of an expert outside of its domain that is available to participate in the meeting. This expert is required to provide information on the labour market. Initially, node E2’s request table includes five entries (note: a status of S indicates that the request packet has been sent): Production (E2) Request Table Sent to Request Packet (Q0,T0,5,Q0,1,T0,1,[E2]) E6 (Q0,T0,5,Q0,2,T0,2,[E2]) E7 (Q0,T0,5,Q0,3,T0,3,[E2]) E8 (Q0,T0,5,Q0,4,T0,4,[E2]) E9 M (Q0,T0,5,Q0,5,T0,5,[E2])

Status S S S S S

Nodes E6, E7, E8 and E9 search their domains and determine that they can satisfy E2 requests. Their respective request tables include the transaction of receiving the request along with a status, R, to indicate a response had been made to the request. Their respective request tables appear as: Receiving (E6) Request Table Sent to Request Packet Ø (null) (Q0,T0,5,Q0,1,T0,1,[E2,E6])

Status R

Processing (E7) Request Table Sent to Request Packet Ø (Q0,T0,5,Q0,2,T0,2,[E2,E7])

Status R

Inspection (E8) Request Table Sent to Request Packet Ø (Q0,T0,5,Q0,3,T0,3,[E2,E8])

Status R

Post Production (E9) Request Table Sent to Request Packet Ø (Q0,T0,5,Q0,4,T0,4,[E2,E9])

Status R

Production (E2) then receives the response packets sent by E6, E7, E8 and E9. Its response table is updated to reflect the receipts of these packets as follows: Production (E2) Received from E6 E7 E8 E9

Response Table Response Packet (Q0,T0,Q0,1,T0,1,R0,1,[E6,E2]) (Q0,T0,Q0,2,T0,2,R0,2,[E7,E2]) (Q0,T0,Q0,3,T0,3,R0,3,[E8,E2]) (Q0,T0,Q0,4,T0,4,R0,4,[E9,E2])

Node M (CEO) also searches its domain to determine if it can satisfy the request for labour market information.


89


When it realizes that it cannot satisfy the request from its own domain, it searches its siblings’ domains to determine if it can route the request to another expert. The search indicates that the request should be sent to Administration (E1). Thus, node M’s request table appears as: CEO (M) Request Table Sent to Request Packet (Q0,5,T0,5,1,Q0,5,1,T0,5,1,[E2,M]) E1

Status S

Administration (E1) searches its domains (and siblings) and determines that its sibling E5 (Human Resources) can provide the necessary knowledge. E1’s request table appears as: Administration (E1) Request Table Sent to Request Packet (Q0,5,1,T0,5,1,1,Q0,5,1,1,T0,5,1,1,[E2,M,E1]) E5

Status S

In turn, node E5 searches its domain, determines that it can satisfy the request, sends a response packet back to its facilitator and changes the status of its request table entry to R. Its request and response tables appear as: Human Resource (E5) Request Table Sent to Request Packet Status Ø (Q0,5,1,T0,5,1,1,Q0,5,1,1,T0,5,1,1,[E2,M,E1,E5]) R Human Resource (E5) Response Table Received from Response Packet Ø (Q0,5,1,T0,5,1,Q0,5,1,1,T0,5,1,1,R0,5,1,1,[E5]) Administration (E1)’s status in the request table is updated to reflect the receipt of a response from its sibling (status R). E1’s response table appears as: Administration (E1) Response Table Received from Response Packet (Q0,5,1,T0,5,1,Q0,5,1,1,T0,5,1,1,R0,5,1,1,[E5,E1]) E5 Node M’s status in the request table is updated to reflect the receipt of a response from E1 (status R). M’s response table appears as: CEO (M) Response Table Received from Response Packet (Q0,5,T0,5,Q0,5,1,T0,5,1,R0,5,1,[E5,E1,M]) E1 Node M routes the final response to Production (E2); E2’s response table appears as: 90

Production (E2) Request Table Received from Response Packet (Q0,T0,Q0,1,T0,1,R0,1,[E6,E2]) E6 (Q0,T0,Q0,2,T0,2,R0,2,[E7,E2]) E7 (Q0,T0,Q0,3,T0,3,R0,3,[E8,E2]) E8 (Q0,T0,Q0,4,T0,4,R0,4,[E9,E2]) E9 M (Q0,T0,Q0,5,T0,5,R0,5,[E5,E1,M,E2]) Production (E2) updates its request table to indicate the change in status from S to R for Q0,5. Then, node E2 aggregates that information with the other four packets received from its siblings to determine if all necessary experts are available to address the top labour needs for the next quarter. Node E2 identifies the outside expert by the path stored in the response packet. Node E2 then activates the direct link to this outside expert and initiates a team session with it and Node E2’s siblings. The session begins by the facilitator (E2) requesting hypotheses on the labour needs from each of the team members. The members generate suggestions in the form of hypotheses. The facilitator denotes all of these hypotheses on a blackboard. At the end of the ‘brainstorming session’, the facilitator requests a rank order of the hypotheses from the members. This process allows the team members to evaluate and select which hypotheses will be pursued, given a set of criteria established by teams and/or facilitators higher up in the hierarchy. Fig. 6 provides a diagram illustrating the activated links in the network during the passage of this team meeting.

5. Final remarks This model does permit access to all segments of the organization. It can potentially utilize knowledge and expertise throughout the organization. If designed and used effectively, it can permit rapid access to such information. This paper has demonstrated the design of a network of expert systems that is flexible with respect to the retrieval of knowledge. This network design also supports each level of decision-making based upon Hackathorn and Keen’s [8] classification. The significance and contribution of this design, over previous designs for networking expert systems, involves the support of unstructured decision-making (such as strategic planning) and the flexibility of supporting different decision settings. Specifically, this framework addresses: (1) the aggregation and disaggregation of problems requiring knowledge from multiple domains; Journal of Information Science, 27 (2) 2001, pp. 81–92


111 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 20 1 2 3 4 5 6 7 8 9 30 1 2 3 4 5 6 7 8 9 40 1 2 3 4 5 6 7 8 9 50 1 2


111 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 1120 1 2 3 4 5 6 7 8 9 1130 1 2 3 4 5 6 7 8 9 1140 1 2 3 4 5 6 7 8 9 50 1 112

CEO

Team meeting Overlay of Activated Links

M

Administration

Production

Marketing

E1

E2

E3

Finance E4

Human Resource E5

Receiving E6

Processing E7

Inspection E8

Post Production E9

Advertising E10

Sales E11

Fig. 6. A network ESMS (with hierarchical and team meeting overlay of activated links).

(2) knowledge acquisition that is not restricted to predefined links within the network; and (3) support for organization and group decisionmaking. However, this model makes a number of assumptions that will not always hold true. More research is needed in the area of networking expert systems to address communication problems that may arise within the network. Examples of such problems include overloading the network (flooding), deadlocks when two experts are waiting for each other to respond, plus the occurrence of noise in the network (which would require the management of acknowledgements). In addition, several interesting knowledge base issues arise that involve anomalies resulting from lack of knowledge base integrity, redundancy of knowledge between experts and inconsistencies in the knowledge bases [5]. Other interesting areas to investigate would include the ability to handle unexpected (dynamic) situations that might arise in the midst of the decision-making process, the propagation and handling of run-time errors and real-time scheduling deadlines for decision making. The issues of what determines optimal group size and the impact of network design on large-scale management of information also require attention. Lastly, the ideas presented here may be applied to other decision-making scenarios. For example, Jacobson and

Levin [29] have been investigating conceptual frameworks for network learning environments, where information flows through the network based on decisions made by ‘mediator’ nodes. In this context, decision making is used to filter the tremendous amounts of information that may flow through a network (such as the World Wide Web).

References [1] O.R. Liu Sheng and W. Chih-Ping, Object-oriented modeling and design of coupled knowledge-base/database systems. In: Proceedings of the International Conference on Data Engineering (1992), pp. 98–105. [2] H.C. Craig and D.R. Rehak, KADBASE – interfacing expert systems with databases, IEEE Expert (Fall 1989) 65–76. [3] L.M.L. Delcambre, RPL: an expert system language with query power, IEEE Expert (Winter 1989) 51–61. [4] J.F. Nunamaker Jr, B.R. Konsynski, M. Chen, A.S. Vinze, Y.I.L. Chen and M.M. Heltne, Knowledge-based systems support for information centers, IEEE Expert (September 1988) 96–105. [5] R. Dieng and S. Hug, MULTIKAT: a tool for comparing knowledge from multiple experts. In: Proceedings of the 6th International Conference on Conceptual Structures (1998).


91


[6] V.S. Jacob and H. Pirkul, A framework for networked knowledge-based systems, IEEE Transactions on Systems, Man, and Cybernetics 20 (1990) 119–127. [7] B.L. Killingsworth, S. Madabhushi and J. Trumbly, Design of a hierarchical knowledge-based system management system (KSMS) for multiple domains. In: Proceedings of the National Decision Sciences Institute Conference (1992), pp. 34–45. [8] R.D. Hackathorn and P.G.W. Keen, Organizational strategies for personal computing in decision support systems, MIS Quarterly 5 (1981) 21–27. [9] R. N. Anthony, Planning and Control Systems: A Framework for Analysis (Harvard University Graduate School of Business Administration, Cambridge, MA, 1965). [10] H.A. Simon, The New Science of Management Decision (Harper and Row, New York, 1960). [11] A. Newell and H.A. Simon, Human Problem Solving (Prentice-Hall, Englewood Cliffs, NJ, 1972). [12] L.D. Erman, R. Hayes-Roth, V.R. Lesser and D.R. Reddy, The Hearsay-II speech-understanding system: integrating knowledge to resolve uncertainty, Computing Surveys 12 (1980) 213–253. [13] S.F. Smith, OPIS: a methodology and architecture for reactive scheduling. In: M. Zweben and M. Fox (eds), Intelligent Scheduling (Morgan Kaufmann, San Francisco, 1994). [14] R. McLeod, Management Information Systems 6th ed. (Prentice-Hall, Englewood Cliffs, NJ, 1998). [15] P.S.Y. Wu and F.L. Zhao, A framework of cooperating expert system to support CAPP. In: Proceedings of the 1st Annual International Conference on Industrial Engineering Applications and Practice (1996), pp. 120–134. [16] Y.S. Huang and C.Y. Suen, A method of combining multiple experts for the recognition of nonconstrained handwritten numerals, IEEE Transactions on Pattern Analysis and Machine Intelligence 17 (1995) 90–94. [17] D. Charlebois, D.G. Goodenough, S. Matwin, A.S. Bhogal and H. Barclay, Planning and learning in a natural resource information system. In: Proceedings of the

92

[18]

[19]

[20]

[21]

[22]

[23]

[24]

[25]

[26]

[27] [28]

[29]

Canadian Artificial Intelligence Conference (1996), pp. 187–199. S.A. Land, J.T. Malin, C. Thronesbery and D.L. Schreckenghost, Making Intelligent Systems Team Players: A Guide to Developing Intelligent Monitoring Systems (NASA Technical Memorandum 104807). N.B. Rauch-Hindin, Artificial Intelligence in Business, Science and Industry: Volumes I and II (Prentice-Hall, Englewood Cliffs, NJ, 1986). R. Davis and R.G. Smith, Negotiation and metaphor for distributed problem solving, Artificial Intelligence 20 (1983) 68–109. T.W. Malone, R.E. Fikes and M.T. Howard, Enterprise: a market-like task scheduler for distributed computing environments. In: B.A. Huberman (ed.), The Ecology of Computation (Elsevier Science Publishers, San Francisco, 1988), pp. 177–206. M.S. Fox, An organizational view of distributed systems, IEEE Transactions on Systems, Man, and Cybernetics SMC-11 (1981) 70–80. C.S. Ho, J.C. Hong and T.S. Kuo, A society model for office information systems, ACM Transactions on Office Information Systems 6 (1986) 106–131. C.C. Woo and F. H. Lochovsky, Supporting distributed office problem solving in organizations, ACM Transactions on Office Information Systems 6 (1986) 185–206. G.A. Gorry and M.S. Scott-Morton, A framework for management information systems, Sloan Management Review 13 (1971) 55–70. P. Kirs, G. Sanders, R. Cerveny and R. Daniel, An experimental validation of the Gorry and Scott-Morton framework, MIS Quarterly 13 (1989) 183–197. C.J. Date, An Introduction to Database Systems 6th ed. (Addison-Wesley, Reading, MA, 1995). R. Fikes and R. Kehler, The role of frame-based representation in reasoning, Communications of the ACM 28 (1985) 904–920. M.J. Jacobson and J.A. Levin, Conceptual frameworks for network learning environments, Journal of Educational Telecommunications 1 (1995) 367–388.


111 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 20 1 2 3 4 5 6 7 8 9 30 1 2 3 4 5 6 7 8 9 40 1 2 3 4 5 6 7 8 9 50 1 2