Features - Enterprise Data Insights:

KNOWLEDGE MANAGEMENT AND IT: THEORY, PRACTICE, AND PROMISE
By Langhorne Carter Bond

Introduction

Knowledge Management, like knowledge itself, can be difficult to define. While there have been numerous writings on the subject in the last decade, comprehensive overviews of the subject are few. Part of the problem is that the field of Knowledge Management is relatively new and evolving. In addition, contributions to the field come from dispersed sources. While writers such as Davenport and Prusak are well known in the United States, the contributions of their European counterparts have only recently begun getting the attention they deserve by management experts in North America.

One source of contributions to the field of Knowledge Management from within the IT industry is the IBM Systems Journal. While it is not an academic peer reviewed journal, it nevertheless has become a force in Knowledge Management for identifying emerging trends. However, not all of industry's contributions are looked upon in a favorable light. In particular, the IT industry is criticized for selling "Knowledge Management" in a box. Often technology is implemented with little regard to the actual business processes that they are supposed to complement or capture.

The ironic result of all this is that the field of knowledge management is fragmented. Diffusion of information about Knowledge Management is currently a problem. This is a problem for organizations wishing to educate themselves not only about the theoretical aspects of Knowledge Management, but also about the current IT options available. Organizations need to know not only the theoretical underpinnings of Knowledge management, but also how to implement Knowledge Management in the field. A useful tool would be an overview that addressed both aspects of Knowledge Management. Such an overview would need to be current. In addition, knowing what new technologies are currently in the development will help IT implementers plan for the future. The aim of this paper is not to be a comprehensive overview of Knowledge Management theory, but to capture snapshots of three aspects of Knowledge Management and thereby create a generalized understanding of Knowledge Management.

Framework Taxonomy -- Theory

Michael Earl provides a good overview of the various theoretical schools of Knowledge Management in his article: "Knowledge Management Strategies: Toward a Taxonomy." There are a number reasons to start with this taxonomy. First, it appears to be the only taxonomy of its kind published to date. Second, as will be addressed shortly, it is a fair and balanced taxonomy. The taxonomy was compiled from data from four basic sources, much of which is original data. Of particular note are interviews with 20 Chief Knowledge Officers. The other sources he draws from include workshop discussions, professional journals, and case study research in six companies.

The taxonomy lists three broad schools of knowledge management: Technocratic, Economic and Behavioral. Each school can have multiple sub types, or attributes. Michael Earl labels the first of the three schools Technocratic and identifies three sub types, Systems, Cartographic and Engineering. The Technocratic schools have their basis in information or management technologies . The second classification is that of the economic school, which has only one subtype, Commercial. The Economic School views and manages knowledge as if it were an asset like stock or bonds. The final classification is the Behavioral schools, which include the Organizational, Spatial and Strategic attribute types. These schools are more behavioral and tend to view knowledge exchange as a fundamentally social process.

The first of the Technocratic schools, the Systems school, is the best known of the schools and is the longest established. This school of thought uses Knowledge Bases to codify knowledge. Examples of this are Xerox and its Web-based maintenance knowledge base. In it are issues from bugs to patches and other articles pertaining to the fixing and maintenance of its copy machines as well as "best practice solutions." The Systems schools tend to be strongly domain specific. The principal IT contribution is its knowledge base systems.

The second Technocratic school is the Cartographic school. This school uses knowledge maps of organizations that list knowledgeable people within the organization that possess certain skills. The article lists the IT contribution as "Profiles and Directories on Internets as the principal contribution by IT. Interestingly enough, IBM's Lotus Discovery Server is a product that fits this school almost perfectly. Thus if someone in marketing needed some one with Oracle database expertise then they simply consult the knowledge map and locate the particular individual within the organizations.

The final school under the Technocratic banner is the Engineering school. The Engineering schools utilize business process re-engineering. Michael Earl lists two important ideas that drive this school. The first is that performance of business processes can be enhanced by providing the operating personnel with knowledge relevant to their tasks. The second is that management processes are inherently more knowledge intensive than business process. Ergo, the provision of not only decision-relevant information, but also contextual and best practice knowledge should be beneficial. This school seeks to use IT to share databases across tasks, levels, departments to all knowledge workers in the process. The example given by Michael Earl is Hewlett-Packard, where there is a strong emphasis on writing up assignments and codification of every work processes and then sharing them on the company intranet. The idea is to share best practices across the company.

However, Hewlett-Packard has had its' own problems with the knowledge management frameworks it adheres to. Take for example Hewlett-Packard's WIN program. At the core of the program was a strong emphasis on social knowledge exchange between knowledge experts and knowledge seekers. Some of the basic components of WIN included networking, peer consulting, and an environment where workers feel safe in innovation with out fear of retaliation. However, in 1995 Hewlett- Packard upper management abruptly and dramatically scaled back the resources and commitment to the WIN program. Nevertheless, Hewlett-Packard has stayed fundamentally committed to the Engineering school, as attested by its recent investment in IT infrastructure designed to support that school. It would be interesting to know if Hewlett-Packard, with its current use of IT technology, could have made the WIN program a better success back in 1993 and 1994.

Under the economic classification, Michael Earl only identifies one type of school, the commercial school. This school seeks to manage knowledge as if it were an asset like stocks and bonds. The example Earl gives is that of Dow Chemical with its more than 25,,000 patents. Dow continually analyzes its patent database, pruning out low performing patents by licensing, outright sales, or donation to the public domain. By using this process, the principal contribution of IT is in the registering and capturing of patent data. The article also lists IBM under this classification, and IBM does manage its patent portfolio in this manner. However, IBM's knowledge management philosophy is not neatly cubby holed into one category. IBM practices in several other schools simultaneously.

The first of the Behavioral schools is Organizational, which focuses on knowledge pooling through networking. This can occur on both the IT level, and on the social level. Groupware and online communities typify the IT component. The social component can consist of social networking that leads to the exchange of information. One example is that of Shell, which constantly seeks to improve its drilling operations.

Shell brings together experts to share their knowledge in multiple ways. The Oil industry is legendary for being a tight community, so they share the knowledge socially.

Nevertheless, in addition, there are online communities and groupware. Interestingly enough, even though this under the behavioral school, this is where a lot of new IT is being used. One IT tool that Shell utilizes extensively is live video streaming across their intranet. In this case, the IT tool reinforces the social transfer of knowledge. The second school under the Behavioral category is the Spatial school, which takes place more on the interpersonal level than the IT level. The example given by Earl is the British Airways headquarters near Heathrow airport in London. The architecture of the building includes spaces that encourage group meetings and encounters. Knowledge is exchanged through this social process. The Spatial, or "social school," contributes technologically in the area of "representational tools"; i.e. the use of PowerPoint to illustrate a point.

Finally, the last school under the behavioral category is the Strategic school. The strategic school views knowledge management as "a dimension of competitive strategy." The example given is Skandia, which has "positioned intellectual capital as the corporation's core capability." That is, knowledge is a key resource. In terms of IT's contribution to this school, it can be very broad. The Strategic school can act as an umbrella encompassing all the other schools. Therefore, a number of IT knowledge management tools incorporate Intranets to online knowledge bases.

Using the Taxonomy -- A Model for the Model

Now that the taxonomy has been established and defined, we return to the original question posed at the beginning of the article. How do organizations proceed from here?

In answer to this, Michael Earl presents a model for using the taxonomy:

        ---> 1. Knowledge Business Vision? -----------------------
        |                                                        |
        |    2. Business Performance Gap?                        |
        |                     |                                  |
        ---- 3. How Could Knowledge Make a Difference?           |
                              |                                  |
             4. Alternative Knowledge Management Initiatives? <---
                              |
             5. Degree of Fit and Feasibility?
                              |
             6. Knowledge Management Program

The model suggests that the starting point for organizations could be an analysis of business performance gap. The author suggests doing a SWAT analysis, and then asking, "How could knowledge make a difference." In analyzing this question, it is important to include alternative knowledge management strategies. Then narrow the field down by analyzing the degree of fit and feasibility of each school. Finally, pick a knowledge management strategy. It is import to remember that the schools presented are not mutually exclusive and thus can be mixed and matched.

Models of the Emerging IT Knowledge Workplace -- Practice

Now that an overview of the frameworks available has been present, the question becomes, "what are the trends and major technologies that are currently being used in the IT market." Clearly, one the biggest trends in IT tools that connect workers with knowledge is the Knowledge Portal.

What exactly is a "portal?" What makes a portal a "knowledge portal? The term "portal" is very ambiguous. Originally, this term described an application (usually web based) that provides a single point of access to distributed online information. Types of information include news channels, web sites and other internet documents. To facilitate navigation of the information, the model features search capabilities and index type directories.

Portals use different techniques and technologies to help connect the user to information. A common feature of most portals is keyword search of a database. However, having a simple keyword search function on a portal site does not automatically make it a Knowledge Portal. A better designation for the basic portal model would be Information Portal. In fact, the first generations of portals were often called information portals because of their emphasis on information.

Portals usually employ various other techniques to help connect the user with information. The Yahoo home page presents the user with a collection of very broad subject heading such as computers, health, business, and other knowledge categories. Selecting a topic takes you to another Yahoo page that gives a more detailed break down of the category, in essence, a table of contents for the Internet. It is interesting to note that Yahoo has been the number one home page used by people since the Internet first emerged.

The concept of the knowledge portal ties into the knowledge work environment. Knowledge Portals and the Emerging Digital Knowledge Workplace presents a current framework. In it, the authors Mack, Ravin and Byrd start by defining knowledge work as "solving problems and accomplishing goals by gathering, organizing, analyzing, creating, and synthesizing information and expertise "Additionally, knowledge work is done by individuals who belong to "communities of interest ". In this framework, knowledge workers gather information relevant to a task, organize it, search it, analyze it, synthesize solutions with respect to specific task goals, and then share, and distribute what they have learned. One example is that of a day in the life of a consultant interacting with a customer in various contexts, such as marketing. Some of the task steps are gathering information about the customer and researching the options available to meet the customer's needs. Finally, Knowledge Management refers to the methods and tools for capturing, storing, organizing, and making accessible both knowledge and expertise .

For a broader definition, the authors refer directly to the outline of knowledge presented by Prusak. That is, knowledge refers to all the tools, technologies, practices, and incentives deployed by an organization to "know what it knows."

A knowledge portal must be able to access as much of the information in the knowledge work environment as possible. Integrating access to different data sources helps capture as much of the knowledge in the work environment as possible. For example, an important discussion about a customer may occur via email. A new company representative may take over handling that customer and want to find relevant information about the customer. If the knowledge portal does not access to the email database, that knowledge will not be taken into the system. Thus, at the heart of Knowledge Management system is the ability to capture and gather information by crawling through the data and building the database. In the knowledge environment, there are many different types of files to analyze, from html web pages to Microsoft Word files created by employees. One example used by the authors is that of IBM's Lotus Discovery Server, that parses Lotus Notes databases, html files, and employee created word processing files. Of course, real world implementation of such a system requires a degree of control to achieve uniformity across the domain. Using the Discovery Server as an example, all the workers would use Lotus Notes and IBM word processing software. In addition, the html documents need a certain degree of uniformity to prevent the actual code of the page from seeping into the database.

An important distinction between information and knowledge is the concept of clustering and gathering ideas into coherent and logical categories. One common feature employed by portals is "context thesaurus" which is used to implement prompted query refinement. This helps the user drill down to more effective searches by suggesting terms to add to their search. In the gathering phase, the search engine analyzes 2 to 3 sentences around the key words and tries to find groups of patterns. The keywords employed by the user are compared to the thesaurus and popular contexts are returned. Other refinements include keeping track of popular links and weighing them more heavily and the "more hits like this" feature.

To move to more advanced document analysis the authors emphasize the importance of using "intelligent agents" to help the database extract textual features from the documents. An example of this would be a system that can recognize proper names. Much more promising are systems that can recognize grammatical features. From simple grammatical features, you can then determine the contextual meaning of words, such as Java, which could mean "coffee" or it could refer to the programming language. It is important to note that this is a very different approach from conventional attempts to extract features using message-understanding conventions. Message understanding conventions try to find similarities in knowledge patterns. The example given by the authors is of a subject like mergers in a business periodical database.

Toward Smarter Systems -- Promise

In the article "Text Analysis and Knowledge Mining System" the authors Nasukawa and Nagano examine the inherent linguistic complications that are encountered by the IT end user when using the plain keyword Search Engine. Current implementations come up lacking when it comes to ease of use by the IT end-user. Plain keyword Search Engines tend to be problematic when it comes to separating knowledge from text. One example given by the authors of "Text Analysis and Knowledge Mining System" is that of the word "Washington," which can be a person (George Washington), place (Washington D.C.) or perhaps have some other meaning altogether. The context in which words are used is very important. Negation and predicate noun pairs (with their dependencies) are two key examples. Current data extraction techniques use what the authors call the "bag-of-words" approach in which relationships between words are treated only as co-occurrences. According to the authors, the IBM Technical Support Call Center covers the most common 1,000 keywords (not including words such as "the") that cover 80 percent of total words. For Medline, 8,000 words cover 80 percent and patent records, 2,000 words. That is a lot concept words. Moreover, the crux of the problem is that using simple keyword queries usually turn up large numbers of equally weighted documents, most of them irrelevant.

In any keyword search engine scenario there is the ever-present question of how usable is the search engine to the average IT end user? Unless the user wants to spend excessive time wading through stacks of results, not much is accomplished. In theory, end users are expected to know how to use Boolean search queries. Even that presents problems because of a lack of standardization. For instance the "+" function before a keyword in multiple keyword search usually means that term must be included in the result. Search engine software packages such as HTDIG and Altavista's search engine allow the "+" to be switched on or off. Some search engine packages don't even support the "+" functionality, instead requiring written Boolean modifiers such as "AND," or "AND NOT" in place of the "+" and "-" functions. Few sites actually publish documentation that list that search engine's functionality clearly. Using a unique key word, such as a distinctive name is usually very effective, but most of the time a characteristic word is not available. When properly implemented, adding a feedback loop to weigh popular links as a higher match does improve accuracy. However, this is a limited improvement.

Employing Natural Language Processing via intelligent agents to help break down and categorize the data is not new. According to the authors, surprisingly little work has been done beyond the most rudimentary levels. Past techniques suffered from shallow natural language processing. In addition, message-understanding conventions are a very different approach because they focus on finding specific class of events, and then filling in a template. While the technology developed in this article is new, its part in the knowledge management framework is not. The framework specifically listed by the authors for this system is in fact, standard. It contains three main components:

• Concept extraction based on robust natural language processing.
• Data mining for discovering rules and patterns
• Visualization 16

Most of the new technology addressed in this article operates in the first element of this framework. However, the technology, in theory, affects steps two and three by making them more efficient.

The first component is where much of the experimental aspect of this system resides. In the system developed by the authors, concept extraction finds novel patterns in the grammatical structure. Then after discovery of structure, the software can reapply the correct meaning to the word. This is clever, because humans essentially do this in processing natural language. We listen for clues that help us differentiate the context of word. Think about all the context errors where you got the meaning of the word wrong.

For example, if you were to walk up to a Java Developer sitting in a Starbucks in Silicon Valley, sipping a cup of coffee in one hand, while at the same time working on Java code on the laptop, and you said one word "java" there would be no way to gauge the context. In these instances, we simply guess most of the time. This happens a lot more than most people realize. This only represents the times you guessed wrong. In theory, a logical response would be to prompt for more context clues and say; "Do you mean java code or java coffee?" We like to think of ourselves as logical creatures, but in reality this rarely happens; we simply guess most of time. We do weigh words in accordance with the common use in the community. Thus if you were to walk up to a college student sitting in a San Francisco Starbucks and said just one word "java" they would probably assume a default context of referring to the coffee. However, our ability to weight meaning is limited and is frequently faulty.

How the software weights meaning is straightforward. The software-parsing agent first scans the text and assigns meaning to words with the aid of a semantic dictionary. The semantic dictionary is able determine many parts of speech on the first pass. Of course, words that have multiple meanings and multiple parts of speech cannot be automatically defined on the first pass. In addition, the semantic dictionary has a simple one-word "base form" that is essentially a definition. Words that have multiple definitions are also listed and given multiple base forms equivalent to each of the meanings. At this point, the software compiler has only to worry about one stack of words, the ones with multiple base meanings. Using the mark up, a shallow parse can divide the words into groups based on how they function structurally in a sentence. That is, the parser divides the words into verb and noun groups. Finally using simple heuristic rules the program can map parts of speech patterns and their base forms to determine intention. Two examples the authors give are:

In addition, the authors specifically mention that a domain expert must help tune the dictionary file to that particular community of practice.

The systems analyzed with the text analysis system were varied in nature. The first one was to an IBM customer-service/support call center that received in excess of 50,,000 calls a month. To complicate matters, employees frequently used short hand such as "cu" for customer. Many of these shorthand notes were added to the semantic dictionary. One analysis presented by the authors is for the simple concept association between the two words "liquid" and "problem." The system strongly points to two uses of the word "sticking," for coffee and soda. The system is also able to produce statistics on the grammatical calls it makes, such as a grammatical chart for the words such as "use."

This can be enormously helpful in identifying legitimate trends and patterns in large collections of data. Currently, the customer support records are scanned using IBM support personnel. Because of the vast collection of data, they only have the ability to scan less than 10 percent of the records generated. Using the intelligent agents plus the assistance of one domain expert to help tune the software for the computers to undertake work, enables the coverage of 100 percent of the records.

Conclusion

Looking over the Michael Earl's taxonomy, one consideration is the limit of IT in providing knowledge management solutions. Probably one of the reasons for the difficulties in starting knowledge management programs is that you cannot buy "knowledge management" in a box. In addition, many of the new IT tools, such as groupware, video streaming etc are not IT solutions in of themselves, but reinforce the social aspect of knowledge transfer. If you look primarily to IT for a complete knowledge management solution, you are likely to be disappointed. The fact the Michael Earl's taxonomy focuses heavily on the social nature of knowledge is probably a sign that it is a fair and balanced taxonomy.

His perspective is interesting for a number of reasons. First, there have not been many attempts to define and classify the various schools of Knowledge Management. In addition, Knowledge Management as a field is relatively new and changing. The impact of consistently new technologies and systems will not be clear for some time.

The authors Knowledge Portals and the Emerging Digital Workplace strongly emphasize two important points. The first is the importance of having a good domain expert that can help integrate the knowledge system. The domain expert has to integrate the system on different levels. More broadly, the system must be designed so that it is coupled with actual work process. Knowledge workplace process should be the driver in design. Once the system is set up, the domain expert can then further refine the clustering and gathering categories by analyzing keyword based on the specific community. Then the domain expert can manually adjust how the keywords are weighted and other query refinements. In addition, the domain expert needs to enforce and unify metadata usage across departments and document types. The second important point is make sure to align your Knowledge Workplace process with the technology that will be supporting them.

Finally, it is clear that computers are finally developing intelligence in areas that will significantly impact Knowledge Management. The system present in Text Analysis and the Knowledge Mining System, is in fact very primitive from a linguistic point of view. There are at least three obvious implications for better text analysis systems. The first is a better semantic dictionary. A more comprehensive semantic dictionary would greatly improve analysis. In addition, the dictionary, or dictionaries could be byte compiled a la Linux for better performance. Second, clearly, today's grammar check systems do not talk to the spell check engines to the extent suggested in the proposed system. Such a system would let spell checkers narrow down the list of possible spellings. In many cases, it would be easy to correct many common misspellings like there/their using the data produced by the system outlined above. Third, once the simple heuristic rules have been applied, the sentence can be re-analyzed to further confirm or reject sentence meanings that are on the margin of interpretation. Even though this an experimental system there are some practical considerations. In an age of increasing information overload, the ability of a knowledge management system to pull useful knowledge from ever- increasing piles of data is a very useful trait. The use of "intelligent agents" that can understand grammatical functions to the extent outlined above will help IT in its quest to make systems that allow users to separate knowledge from the ever growing background of data overload.

About The Author

Langhorne Carter Bond works as an Information Technology consultant and writes about how businesses develop dysfunctional relationships with their Informational Technology systems. Comments can be sent to the author at bondl@cox.net.