Data processing: artificial intelligence – Knowledge processing system – Knowledge representation and reasoning technique
Reexamination Certificate
2000-07-18
2003-04-22
Follansbee, John A. (Department: 2121)
Data processing: artificial intelligence
Knowledge processing system
Knowledge representation and reasoning technique
C707S793000
Reexamination Certificate
active
06553361
ABSTRACT:
FIELD OF THE INVENTION
The present invention relates to a knowledge based system and, in particular, to a knowledge based system which is implemented by a computing system.
BACKGROUND OF THE INVENTION
It is known to provide a knowledge based system for carrying out a task such as generating an output conclusion based on an input series of attributes. Such systems are useful because once the necessary knowledge is stored in the system, it is possible for a non-expert to solve a task without the need to consult an expert. However, a disadvantage with the majority of such knowledge based systems is that it is necessary to acquire the knowledge from the expert and utilise a knowledge engineer to correctly store and organise the acquired knowledge in the knowledge based system.
In order to overcome difficulties building and maintaining this type of system, an alternative knowledge based system referred to as a Ripple Down Rules System (RDR) has been devised.
Ripple Down Rules is a process for building and maintaining knowledge based systems which utilises the normal process of knowledge acquisition. The process was conceived as a result of a realisation that when asked why a certain conclusion applies in a given situation, an expert generally does not explain how the conclusion was reached but, rather, gives a justification for why that the conclusion is correct. This realisation is critical since for each case presented to an expert, the expert provides and justifies a conclusion in a particular and specific context. The conclusion may not be appropriate for other contexts that the expert did not consider. In other words, the normal process of knowledge acquisition involves continual refinement and correction.
In an RDR system, therefore, knowledge is acquired by accumulating conclusions made by experts and associating with each changed conclusion a justification for the change of conclusion.
An RDR system is structured in a way which may be represented in the format of a tree, with each node of the tree being associated with a particular conclusion. In addition, each child node of a preceding parent node is associated with a conclusion which is a refinement of the conclusion associated with the parent node. The RDR system is constructed on a case by case basis.
An example of a known RDR knowledge based system (10) is shown in
FIGS. 1-4
of the accompanying drawings.
In
FIGS. 1-4
, the RDR system is shown at different stages of construction. The system is constructed by presenting “cases” to the knowledge base, a “case” being a set of attributes which are to be used in determining an appropriate conclusion.
Initially, the system contains a single “null” first node
12
as shown in FIG.
1
.
Irrespective of the attributes of the input case, the first node
12
is always satisfied and when executed returns the conclusion “null”.
In relation to RDR systems a “rule” is a set of criteria which define the justification for the difference between the case associated with a child node and the case associated with its parent node. A rule may include one or several “conditions” which define the justification.
In operation of the system, a node is said to be “satisfied” when all rules between the node and the first node are satisfied. A node is said to be “executed” and the conclusion associated with the node returned when all rules between the node and the first node
12
are satisfied and no rules between the node any of its child nodes are satisfied.
A first case including the attributes “barks”, “four legs”, “warm blooded” and live born” is presented to the knowledge base and the system returns a conclusion “null”. The domain expert disagrees with the conclusion. In order to provide a correct conclusion for the case, the domain expert indicates to the system that the expert disagrees with the conclusion and the system adds a second node
14
, as shown in FIG.
2
. The second node
14
is a child node of the first node
12
. The expert supplies a conclusion to be associated with the second node
14
and, in this example, the conclusion is chosen to be “dog”. The expert is also required to provide a justification for the difference between is the given conclusion “null” and the chosen conclusion “dog”. In this example, the justification is chosen to be “four legs=true” and this is added between the first node
12
and the second node
14
as a rule.
The case on which the second node
14
is based is stored in the knowledge base as a cornerstone case associated with the second node
14
.
A second case including the attributes “lives in water” and “gills” is presented to the knowledge base and the system returns a conclusion “null” since the first node
12
is satisfied but its child node is not satisfied.
The domain expert disagrees with this conclusion. In order to provide a correct conclusion for the case, the domain expert indicates to the system that the expert disagrees with the conclusion and the system adds a third node
16
, as shown in FIG.
3
. The third node
16
is a child node of the executed first node
12
. The expert supplies a conclusion to be associated with the third node
16
and, in this example, the conclusion is chosen to be “goldfish”. The expert is also required to provide a justification for the difference between the given conclusion “null” and the chosen conclusion “goldfish”. In this example, the justification is chosen to be “lives in water” and this is added to the link between the first node
12
and the third node
16
as a rule.
The case on which the third node
16
is based is stored in the knowledge base as a cornerstone case associated with the third node
16
.
A third case including the attributes “four legs”, “whiskers”, “warm blooded” and “live born” is presented to the knowledge base, and the system returns a conclusion “dog”, since the case presented to the knowledge base satisfies the condition “four legs=true” and the second node
14
does not have any child nodes at this stage. The domain expert disagrees with this conclusion. In order to provide a correct conclusion for the case, the domain expert indicates to the system that the expert disagrees with the conclusion and the system adds a fourth node
18
, as shown in FIG.
4
. The fourth node
18
is a child node of the executed second node
14
. The expert supplies a conclusion to be associated with the fourth node
18
and, in this example, the conclusion is chosen to be “cat”. The expert is also required to provide a justification for the difference between the present case and the case associated with the given conclusion “dog”. In this example, the justification is chosen to be “barks=false” and this is added to the link between the second node
14
and the fourth node
18
as a rule.
The case on which the fourth node
18
is based is stored in the knowledge base as a cornerstone case associated with the fourth node
18
.
It will be understood from the diagrams in
FIGS. 1
to
4
that each new node that is added to the knowledge base is a node which is linked by a rule to a previous node, the previous node being a parent node and the new node being a child node. It will also be understood that the new node is associated with a conclusion which is a refinement of the conclusion associated with its parent node.
This type of Ripple Down Rules system allows only a single path through the tree to be followed at any one time, and therefore only one conclusion to be returned for each case presented to the knowledge base. Such an RDR system is known as Single Classification RDR (SC-RDR).
A flow diagram showing a method of construction of the Single Classification RDR system described above is shown in FIG.
5
. The method includes the steps of presenting a case to a knowledge base
19
, interpreting the case by the knowledge base and outputting a conclusion
21
, entering a decision by an expert as to whether the interpretation is correct
23
, receiving expert approval if the interpretation is correct
25
, adding a new node to the knowledge base if the interpretation is n
Compton Paul Justin Lloyd
Edwards Glenn Andrew
Harries Michael Bonnell
Lazarus Leslie
Peters Lindsay Andrew
Davis & Bujold P.L.L.C.
Follansbee John A.
Hirl Joseph P.
Pacific Knowledge Systems Pty Ltd.
LandOfFree
Knowledge based system does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Knowledge based system, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Knowledge based system will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3033359