Knowledge Bases are a product of KM (Knowledge Management). KM is a murky, multidisciplinary area that covers business management, sociology, document management and cooperative computing solutions. Although KM includes CRM (Customer Relationship Management), ERP (Enterprise Resource Planning) and business intelligence, each of these applications supports a collection of different business processes that involve multiple tasks as well as a focus on either the employee or the customer. Knowledge Bases focus on the creation, storage and reuse of documents as a tangible resource for the enterprise and they can be part of a KM solution.

A Base for Knowledge

Creating knowledge in an enterprise is not unlike creating knowledge for an individual. We all learn from personal experience and from the experience of others. An enterprise is different in degree, not in kind. It learns from the collective experience of its employees and by sharing expertise. To learn from collective experience, enterprises need to cultivate ideas and provide incentives for employees to collaborate on solutions to problems.

Many enterprises today foster competition among workers to improve their individual worth and value to the business through rewards and recognition. In this model, there is no benefit gained from collective knowledge, and employees may feel threatened by sharing knowledge with coworkers; parting with it could mean parting with their job. To create knowledge as a business resource, management needs to foster collaboration by providing new incentives and rewards.

Helpdesk staff members often pride themselves on the knowledge they impart to customers and end users. That knowledge often leads to rewards and bonuses. Just as often, the end result may be incomplete reporting -- real solutions leave the enterprise when employees "walk." To create a Knowledge Base, incentives and rewards must be changed; employees must do more than simply resolve incidents, they record them, so that solutions can be shared and reused across the enterprise. Once the enterprise becomes a base for knowledge, it's ready for a Knowledge Base.

Knowledge Base as Expert System

A helpdesk Knowledge Base retrieves and reuses support solutions that have worked for similar situations in the past. This CBR (Case-Based Reasoning) approach came from research and development in XPS (expert systems) and begins with a collection of experiences or cases stored in a repository or case base. Each case contains a problem description and solution enhanced by descriptors for case numbers, context and evaluations pertaining to the reusability of the case. In addition, there may be attachments where solutions can be described and delivered to end users in multimedia formats. Knowledge Bases can incorporate hardware and software manuals, FAQs and newsgroup files. Remedy Corp.'s Remedy Help Desk, Computer Associates International's Unicenter TNG Advanced Help Desk and Peregrine Systems' Knowlix use Knowledge Bases in their helpdesk software solutions that add groupware and workflow functionality.

Knowledge Bases can be broken down into components. These components include: the case base; the vocabulary used to describe cases; the similarity measures used to compare cases (representation); and the rules or algorithms used to transfer and integrate new solutions into the case base. There is no strict separation between these four components with respect to the overall problem-solving task. Each component can lead to possible solutions to problems. In fact, the entire system could be implemented using an OLCP (Online Complex Processing) database such as Oracle8 or MS-SQL Server that provides high performance and availability to run large queries and issue long reports while allowing end users to access and update cases.

Retrieving information from Knowledge Bases involves different search strategies, depending on the size of the case base and the complexity of case representation. For Knowledge Bases with structured representation and multiple document types, searching can be done using multiple indexes and decision tree models. For example, suppose the helpdesk just received a call about a mail server allowing unrestricted relay in the enterprise. A staff person would access the Knowledge Base user interface and create a new case.

CASE NUMBER: NWC999
DATE: 03-10-01
CONSULTANT: sdoherty@nwc.com
CUSTOMER: <Directory Information>
PROBLEM DESCRIPTION: Mail server GUINEVERRE (Sendmail 8.8) allowing unrestricted mail relay AKA open mail relay.

Once the new case is created, the helpdesk searches the Knowledge Base for previous problems relevant to the description for SMTP servers acting as mail relays. This search may involve documents represented by certain classes (such as network, e-mail or server) from cases, manual pages or product documentation. The retrieved set then would be compared using configurable rules to analyze content description. In Knowledge Bases where case representation is less structured, linear search models may be used to search classes and document types. Using simple keyword matching or statistics about word frequency, however, may not be sufficient where documents span multiple classes and document types. Simple keyword searching for product names, error messages or codes may lack sufficient meaning to include variable terms associated with similar concepts across the Knowledge Base.

After retrieving cases similar to the SMTP mail relay problem, possible solutions will need to be applied to the new case. Frequently, some adaptation is necessary, where there are differences in the suggested solution as applied. Retrieving cases and documents that are related to our SMTP configuration problem could return numerous cases and main pages. One document could contain the correct configuration settings, but fail to show the reader how to verify that the problem exists or to remind the reader that the sendmail process needs to be restarted before the changes take effect. Problem verification and completion may come from other documents or from the staff person's experience. Adding this information to the new case for future use allows the Knowledge Base to adapt and learn.

Once the new case is saved to the Knowledge Base, it can be represented by class -- for example, e-mail and server -- and by document type. Running the problem description through a filter can normalize and index words using a controlled vocabulary. Without filters, the full text of the problem description can also be indexed; however, such an index may grow to unreasonable lengths and degrade document retrieval. Adaptation and new case integration also can be achieved by parameter adjustments to cases and documents according to configurable rules, constraints and document references, but sophisticated adaptation mechanisms have yet to be realized.