AKRI

Knowledge: Knowledge representation

The way that knowledge is stored and organised is a very important area. Work within the science of Artificial Intelligence has shown that knowledge representation is far from trivial and the representation method chosen can have a significant impact of the way knowledge is applied.

Even people can represent knowledge differently and consequently draw different conclusions from its application. There is knowledge in books but it needs extracting (learning). The book simply provides the material necessary to learn. Other methods such as rules and maps can offer representation options that make knowledge easier to apply directly from the representation framework. There is also the idea of invisible or tacit knowledge. It is not really clear whether this is real knowledge or not but there is a lot of evidence that it is applied as such. The difficulty is in representing this sort of knowledge.

Representation of Knowledge in Machines

Machine Knowledge

It is difficult to say whether a machine can really possess knowledge. Does the knowledge held by a machine comply with accepted definitions for knowledge? Probably not. However, it is possible for machines to both possess and apply something that would be considered knowledge if a human possessed and applied it. From this point of view it is reasonable to talk about knowledge that is contained in a machine. Much more reasonably in fact than it is possible to say that there is knowledge in a book.

Machines are very useful knowledge tools. They can be used to store knowledge in a variety of ways and knowledge can easily be disseminated amongst machines. Even more impressively, through the use of various software techniques, machines can actually learn or generate knowledge from data. A machine can learn things about an environment or a machine can learn new things from studying complex company data bases.

Knowledge based systems and knowledge engineering are now accepted options for dealing with organisational knowledge.

Data is stored in computers using several methods. For example, numbers may be stored a integers, as real numbers or in exponential form. Characters and words are stored differently. Various structures are used to give data (information) some meaning and to facilitate access.

Knowledge is more complex and requires more detailed treatment. For knowledge to be represented, it must be done in such a way as to make it functional. The representation must be able to explain functionality and justify decisions. Several key representations are listed:

Semantic Networks

Semantic Networks are networks of Nodes and Arcs. Nodes represent particular concepts or elements and arcs represent relationships between concepts or elements. They are a powerful knowledge representation system, easy to understand by humans and can be used in automated processing systems. This means that they can also become a vehicle to archive company knowledge. A typical semantic network that represents knowledge concerning an electric space heater could be:

Semantic Network to describe an Electric Heater

Electric Heater
                Diagram

In this simple network, nodes are specific items and links show relationships between items.

It would be possible for an automated system to answer questions about items contained within the network by following links (provided that it could understand the questions).

  • How does power get to the heating element?
  • What is the purpose of the lamp?
  • It would also be possible to a computer to construct a textual statement about the knowledge contained in the network.

Concept Diagrams

Concept diagrams are closely related to semantic networks. Concept Diagrams are also composed of nodes and arcs and the nodes and arcs have similar functions. Concept diagrams can be used to describe fairly complex concepts and are suitable for both machine and human interpretation. They are seen as a knowledge representational method that employs graphical structures (Sowa 1984). There is a body of work relating to concept diagrams and their use as a graphical logic. (Sowa 1993). This offers interesting opportunities for work on knowledge mapping by creating the framework that could allow knowledge maps to be transformed into other machine understandable representations such as the Knowledge Interchange Format (KIF) (Genesereth 1992).

Graphs & Maps

Diagram : Terminology
                of Graphs and maps

Diagram : Terminology of Graphs and maps

Graphs and maps are very much part of this group of representation methods. They are essentially nodes joined by arcs. The meaning of the nodes and arcs can be varied according to the uses that the maps or graphs are put to. Maps and graphs can be categorised by the way that nodes are allowed to be connected up by arcs. Simple non overlapping connected structures are referred to as maps. Some allow only two child arcs and can be known as binary maps. Graphs may have a more complex and overlapping connectivity structure. The figure above identifies some of the commonly used terminology.

Language

Several specific languages have been developed to support Artificial Intelligence and the development of systems. The most popular ones are LISP and PROLOG. Many expert systems are developed in these languages because they contain natural ways of representing knowledge and making inference.

Scripts

A script is a typical way in which actions are carried out or situations occur. A typical example of a script is the restaurant script. In certain types of restaurant people entering may expect to be seated and may expect to be given a menu. etc. They have a script for it.

Expert Systems

Knowledge can also be represented in a computer based expert system or Knowledge Based System. These systems have some fundamental differences and some important similarities to human experts and although no substitute for human experts, they can be a sensible, cost effective and practical option for a range of specific business and industrial problems.

Rules

Rules are reasonably easy to understand by humans and are also a powerful machine based knowledge representation scheme. Rule based systems that could apply human knowledge and function at the level of a human expert were famously pioneered by E.H. Shortliffe in the system 'Mycin' (Shortliffe 1976). Rules require knowledge to be identified as attribute value pairs. They take the general form:

if attribute A1 has value V1

and attribute A2 has value V2

then attribute A3 has value V3

Attributes can represent internal data items, they can represent input or output systems or they can initiate a response from the user. Once knowledge is represented as a rule set, it is relatively easy to construct an engine that can make use of the rules in an automated reasoning system.

Attributes can represent internal data items, they can represent input or output systems or they can initiate a response from the user. Once knowledge is represented as a rule set, it is relatively easy to construct an engine that can make use of the rules in an automated reasoning system.

In addition, the rules themselves can be archived and updated as necessary. This would be a knowledge archive rather than an information archive since the rules can be directly used in automated reasoning.

Exception systems are similar to rules in that they can also be archived, understood by humans and used directly in automated reasoning systems. Exceptions may take the form:

attribute A1 has value V1

unless attribute A2 has value V2

and attribute A3 has value V3

Frames

Frames are also a powerful knowledge representation system that are accessible to both humans and machine. A frame is a collection of information and associated actions that represents a simple concept. It would be possible to represent a person (in a simple way) by the use of a frame.

Frame: Elery Stone

Specialisation of: Frame Person

Date of Birth: 30:04:62

Sex: Male

Nationality: British

Home Town: St. Helens

Occupation: Tailor

Health: (Consult Medical system)

In the simple frame shown above, most of the slots have values but one slot requires an independent system to be called to find a value. Frames are a mixture of information, calls to information derivation functions and output assignment. Frames can be used to represent complex pieces of knowledge and can also be archived and edited as required.

Cases

A case is a record of an instance of an activity or event. Cases are similar to frames in that they require consistent index identifiers that allow cases to be compared with each other. For instance, a case containing information about a specific car sale would include information about the buyer, the actual product bought, when it was bought etc. A collection of cases of similar type could be used by an organisation for a variety of purposes. It could help identify sales targets, it could help match customers with products etc.

Links

Knowledge Representation at Vassar College, N.Y

Imperial College, London