 |
Claims  |
|
|
What is claimed is:
1. A method for retrieving optimum case information for a current problem,
comprising the steps of:
retrieving case candidates having features common to features of the
current problem from information representing a plurality of cases stored
in a case base in response to presentation of the current problem, wherein
the case base is stored in a storage unit and each case includes a case
problem and a solution or solving method for the case problem;
determining ann optimum case candidate from the case candidates using a
group of features of the current problem, a group of common features and a
group of features of the case problems of said case candidates in
accordance with influence relation information indicating a feature group
influenced by other feature groups, said determining including determining
reference feature groups of respective case candidates from the group of
features of the current problem, the group of common features and the
group of features of the case problems of respective case candidates in
accordance with tine influence relation information, wherein features of
one reference feature group are common to the current problem and the case
problem and being subject to no influence by non-common features wherein
said non-common features are features included in only one of the current
problem and the case problem, and selecting one case candidate which has
the largest number of features among the reference feature groups as the
optimum case candidate; and
retrieving case information corresponding to the optimum case candidate
from information representing the plurality of cases.
2. A method according to claim 1, wherein a step of determining said
reference feature group includes the steps of:
(a) determining an intersection of a problem feature set representing the
group of features of a current problem and a case feature set representing
the group of features of the case problems of said case candidates;
(b) determining a difference set obtained by removing the intersection from
a union of the problem feature set and the case feature set;
(c) determining an influenced feature set wherein said influenced feature
set is a set of features influenced by elements in the difference set by
reference to the influence relation information;
(d) determining a reference feature set by removing the influenced feature
set from the intersection; and
(e) repeating steps (a) to (d) over all the case candidates.
3. A method according to claim 1, further comprising the steps of:
storing information obtained in the step of retrieving case candidates and
the step of determining the optimum case candidate in a retrieving process
information storing unit as retrieving process information; and
representing the retrieving process information or influence relation
information from the retrieving process information storing unit in
response to an explanation request.
4. A system for retrieving optimum case information for a current problem,
comprising:
a case base storing unit for storing a case base which stores information
representing a plurality of cases;
candidate retrieving means for retrieving case candidates having features
common to features of the current problem from information representing a
plurality of cases stored in the case base by accessing said case base
storing unit in response to input of the current problem, said case
candidates including case problems and solutions or solving methods for
the case problems;
optimum candidate determining means for responding to said candidate
retrieving means and for determining an optimum case candidate among said
case candidates based on a group of features of the current problem, a
group of common features and a group of features of the case problems of
said case candidates in accordance with influence relation information
indicating a feature group influenced by other feature groups, said
optimum case candidate determining means including a reference feature
group determining means for determining reference feature groups of said
case candidates from the group of features of the current problem, the
group of common features and the group of features of the case problems of
said case candidates in accordance with the influence relation
information, wherein features of the reference feature group are common to
the current problem and the case problem and being subject to no influence
by non-common features wherein said non-common features are features
included in only the current problem or the case problem, and selecting
means for selecting one case candidate having the largest number of
features among the reference feature groups as the optimum case candidate;
and
optimum case information retrieving means for responding to said optimum
candidate determining means and for retrieving optimum case information
corresponding to said optimum case candidate from information representing
the plurality of cases.
5. A system according to claim 4, wherein said reference feature group
determining means includes:
operating means for executing an operation for determining an intersection
of a problem feature set representing the group of features of the current
problem and a case feature set representing the group of features of the
case problems of said case candidates, determining a difference set
obtained by removing the intersection from a union of the problem feature
set and the case feature set, determining an influenced feature set
wherein said influenced feature set is a set of features influenced by
elements in the difference set by reference to the influence relation
information, and determining a reference feature set wherein said
reference feature set is obtained by removing the influenced feature set
from the intersection; and
means for having said operating means execute one or more operations on all
the case candidates.
6. A system according to claim 4, further comprising:
a retrieving process information storing unit;
means for storing information obtained from said retrieving case candidates
step and from said determining the optimum case candidate step in said
retrieving process information storing unit as retrieving process
information; and
representing means for representing said retrieving process information or
influence relation information from said retrieving process information
storing unit in response to an explanation request.
7. A method for inferring an optimum solution for a current problem,
comprising the steps of:
retrieving case candidates having features common to features of the
current problem from information representing a plurality of cases stored
in a case base in response to a presentation of the current problem, a
storing unit storing the case base and the plurality of cases including
case problems and solutions or solving methods for the case problems;
determining an optimum case candidate among said case candidates based on a
group of features of the current problem, a group of common features and a
group of features of the case problems of said case candidates in
accordance with influence relation information showing feature groups
influenced by other features groups, said determining said optimum case
candidate includes determining reference feature groups of said case
candidates from the group of features of the current problem, the group of
common features and the group of features of the case problems of said
case candidates, wherein features of a reference feature group are common
to the current problem and the case problem and being subject to on
influence by non-common features wherein said non-common features are
features included in only one of the current problem and the case problem
and selecting one case candidate having the largest number of features
among the reference feature groups as the optimum case candidate;
retrieving optimum case information corresponding to the optimum case
candidate from information representing the plurality of cases; and
inferring a solution for the current problem from a solution or a solving
method for the case problem in the optimum case information in accordance
with a case using rule group.
8. A method according to claim 7, wherein the step for determining said
reference feature group includes the steps of:
(a) determining an intersection of a problem feature set representing the
group of features of the current problem and a case feature set
representing the group of features of the case problems of said case
candidates;
(b) determining a difference set obtained by removing the intersection from
a union of the problem feature set and the case feature set;
(c) determining an influenced feature set which are features influenced by
elements in the difference set by reference to the influence relation
information;
(d) determining a set obtained by removing the influenced feature set from
the intersection; and
(e) repeating the steps (a) to (d) over all said case candidates.
9. A method according to claim 7, further comprising the steps of:
storing information obtained in the step of retrieving case candidates and
the step of determining an optimum case candidate in a retrieving process
information storing unit as retrieving process information; and
representing the retrieving process information or influence relation
information from the retrieving process information storing unit in
response to an explanation request.
10. A system according to claim 7, wherein:
said case using rule group is comprised of one or more case using rules
wherein said one or more case using rules include a group of features and
processing steps for solving respective portions of a case problem
represented by the group of features, and
said inferring step includes the step of executing said one or more case
using rules of the case using rule group.
11. A system for retrieving optimum case information for a current problem,
comprising:
a case base storing unit for storing a case base which stores information
representing a plurality of cases;
candidate retrieving means for retrieving case candidates having features
common to features of the current problem from information representing
said plurality of cases stored in said case base by accessing said case
base storing unit in response to an input of the current problem, said
plurality of cases including case problems and solutions for solving
methods for the case problems;
optimum candidate determining means for responding to said candidate
retrieving means and for determining an optimum case candidate among said
case candidates based on a group of features of the current problem, a
group of common features and a group of features of case problems of said
case candidates in accordance with influence relation information
indicating a feature group influenced by other feature groups, said
optimum candidate determining means including a reference feature group
determining means for determining reference feature groups of said case
candidates from the group of features of the current problem, the group of
common features and the group of features of the case problems of said
case candidates in accordance with the influence relation information,
wherein features of the reference feature group are common to the current
problem and the case problem and being subject to no influence by
non-common features wherein said non-common features are features included
in only the current problem or the case problem, and selecting means for
selecting one case candidate having the largest number of features among
the reference feature groups as the optimum case candidate;
optimum case information retrieving means for responding to said optimum
candidate determining means and for retrieving optimum case information
corresponding to the optimum case candidate from information representing
the plurality of cases; and
inferring means for inferring a solution for the current problem from the
solution or the solving method for the case problem in the optimum case
information in accordance with a case using rule group.
12. A system according to claim 11, wherein said reference feature group
determining means includes:
operating means for executing an operation for determining an intersection
of a problem feature set representing a group of features of the current
problem and a case feature set representing a group of features of
problems of respective case candidates, determining a difference set
obtained by removing the intersection from a union of the problem feature
set and the case feature set, determining a set of influenced features
which are influenced by elements in the difference set by reference to
influence relation information, and determining a set obtained by removing
the set of the influenced features from the intersection as a reference
feature set; and
means for having said operating means execute one or more operations on all
the case candidates.
13. A system according to claim 11, further comprising:
a retrieving process information storing unit;
means for storing information obtained from determining the optimum case
candidate in said retrieving process information storing unit as
retrieving process information; and
representing means for representing the retrieving process information or
influence relation information from said retrieving process information
storing unit in response to an explanation request.
14. A system according to claim 11, wherein said case using rule group is
comprises of one or more case using rules wherein said one or more case
using rules include a group of features and processing steps for solving
respective portions of a case problem represented by the group of
features.
15. A method for retrieving optimum case information for a current problem,
comprising the steps of:
retrieving case candidates having features common to unsolved features of
the current problem from information representing a plurality of cases
stored in a case base in response to presentation of the current problem,
wherein the case base is stored in a storage unit and each case includes a
case problem and a solution or solving method for the case problem,
wherein the current problem comprises at least one solved part, an
unsolved part, and a satisfied part of the solution for the problem being
described in the case;
determining a reference feature group of each case in accordance with
influence relation information indicating a feature group influenced by
other feature groups, each of the features of the reference feature group
being not influenced by common features which are common to the current
problem and the case problem and are included in the unsolved part of the
current problem and are included in the part of the case problem having
the solution described therein and are included only in the current
problem or case problem, by the features which are included in part of the
solved part of the current problem difference in solution from the case,
and by the features which are included in part of the case problem having
no solution described therein;
selecting as an optimum case candidate one of the case candidates having a
highest number of features matching the features of the reference feature
group; and
retrieving case information corresponding to the optimum case candidate
from information representing the plurality of cases.
16. A system for retrieving optimum case information for a current problem,
comprising:
a case base storage unit for storing a case base, the case base comprising
a plurality of cases;
candidate retrieving means for accessing the case base storage unit in
response to a presentation of the current problem and for retrieving
candidates of the cases which have features common to unsolved features of
the current problem on a bases of information of the plurality of cases
stored in the case base, each of the cases include a case problem and a
solution or solution method for the case problem, wherein the current
problem comprises at least one solved part, an unsolved part, and a
satisfied part of the solution for the problem being described in the
case;
optimum candidate deciding means, in response to the candidate retrieving
means, for determining a reference feature group of each case information
in accordance with influence relation information indicating a feature
group influenced by other feature groups, each of the features of the
reference feature group being not influenced by the features which are
common to the current problem and the case problem and are included in the
unsolved part of the current problem and are included in the part of the
case problem having the solution described therein and are included only
in the current problem or case problem, by the features which are included
in part of the solved part of the current problem different in solution
from the case, and by the features which are included in part of the case
problem having no solution described therein, the optimum candidate
deciding means also selecting as the optimum case candidate one of the
case candidates in one of the reference features groups which is highest
in the feature number; and
optimum case information retrieving means for retrieving case information
corresponding to the optimum case candidate from information representing
the plurality of cases.
17. A method for retrieving optimum case information for a current problem,
comprising the steps of:
retrieving case candidate having features common to unsolved features of
the current problem from information representing a plurality of cases
stored in a case base in response to a presentation of the current
problem, wherein the case base is stored in a storage unit and each case
includes a case problem and a solution or solving method for the case
problem, wherein at least one part of the current problem is already
solved and a satisfied part of the solution for the problem being
described in the case;
determining a reference feature group of each case information in
accordance with influence relation information indicating a feature group
influenced by other feature groups, each of the features of the reference
feature group being not influenced by the features which are common to the
current problem and the case problem and are included in the unsolved part
of the current problem and are included in the part of the case problem
having the solution described therein and are included only in the current
problem or case problem, by the features which are included in part of the
already-solved part of the current problem different in solution from the
case, and by the features which are included in part of the case problem
having no solution described therein;
selecting as an optimum case candidate one of the case candidates from one
of the reference feature groups which has a highest feature number;
retrieving an optimum case information corresponding to the optimum case
candidate from information representing the plurality of cases; and
inferring a solution of the current problem from the solution or solving
method of the case problem in the optimum case information in accordance
with a group of case application rules.
18. A system for retrieving optimum case information for a current problem,
comprising:
a case base storage unit for storing therein a case base which stores
therein information indicative of a plurality of cases;
candidate retrieving means for the case base storage unit in response to a
presentation of the current problem and for retrieving candidates of the
cases which have features common to unsolved features of the current
problem on the basis of the information of the plurality of cases stored
in the case base, each of the cases include a case problem and a solution
or solution method for the case problem, wherein at least one part of the
current problem is already solved and a satisfied part for the solution of
the problem being described in the case;
optimum candidate deciding means, in response to the candidate retrieving
means, for determining a reference feature group of each case information
in accordance with influence relation information indicating a feature
group influenced by other feature groups, each of the features of the
reference feature group being not influenced by the features which are
common to the current problem and the case problem and are included in the
unsolved part of the current problem and are included in the part of the
case problem having the solution described therein and are included only
in the current problem different in solution from the case, and by the
features which are included in part of the case problem having no solution
described therein, the optimum candidate deciding means also selecting as
an optimum case candidate one of the case candidate in one of the
reference feature groups which has a highest feature number;
optimum case information retrieving means for retrieving case information
corresponding to the optimum case candidate from information representing
the plurality of cases; and
inferring means for inferring a solution of the current problem from the
solution or solving method of the case problem in the optimum case
information in accordance with a group of case application rules.
19. A method for retrieving optimum case information for a current problem
having features, comprising the steps of:
determining a current problem feature set comprised of the features of the
current problem;
retrieving case candidates from a plurality of cases stored in a case base
having features common to the features of the current problem in response
to a request to solve the current problem, wherein the case base is stored
in a storage unit and each case includes a case problem and a solution or
solving method for the case problem;
generating a case problem feature set comprised of features of the case
problem of the case candidate retrieved;
generating a common feature set wherein the common feature set is an
intersection set between the current problem set and the case problem set;
generating a non-common feature set wherein the non-common feature set is
obtained by subtracting the common feature set from a union of the current
problem feature set and the case problem feature set;
generating an influenced feature set wherein the influenced feature set is
a set of features influenced by features in the non-common feature set
based on influence relation information;
generating a reference feature set wherein the reference feature set is
equal to the common feature set minus features contained in the influenced
feature set;
selecting one case candidate having a greatest number of features which
match the features of the reference feature set as an optimum case
candidate; and
retrieving case information corresponding to the optimum case candidate
from the plurality of cases. |
|
 |
|
 |
Claims |
|
|
|
Description |
|
|
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to a method of retrieving an optimum case for
a given problem from a case data base and inferring a solution of the
problem by using the retrieved case, and a system therefor.
2. Description of Related Art
Case information means data composed of data representing a problem and
data representing a solution of the problem or a way to solve the problem,
i.e., a solving method. One of the objects of using case information is
such that a system retrieves a solved case of a problem similar to a
problem to be currently solved (an analogous case) and a user refers to
the solution or the solving method in that case. Another object of using
case information is such that the system retrieves the analogous case and
further performs an inference by using the analogous case, thereby to
obtain a solution of a current problem. Thus, a technical subject for
using case information is to develop a method of retrieving an analogous
case which furnishes with much information for the current problem out of
a plurality of accumulated cases and a method of solving the problem by
using the analogous case.
A conventional method of retrieving and using an analogous case is
discussed in "Case-Based Reasoning Workshop", (1988) pp. 21-30, for
instance. A problem is represented therein by using goals which are
desired to achieve in the problem and constraints which are to be
satisfied in solving the problem. First, the following processing method
has been proposed for analogous case retrieval.
(1) Goals of the current problem and goals of the problem in a case are
compared with each other, so as to select cases including plenty of common
goals.
(2) When a plurality of cases are obtained in the processing (1), the
constraints of the current problem and the constraints of the problem in
the case, i.e., the case problem are compared with one another so as to
select cases which hold one or more important constraints in common.
(However, information relating to significance of constraints is given to
the system in advance.)
This system is based on the following thoughts in substance.
(1) The more features the current problem and the case problem includes in
common, the more similar those cases are.
(2) In order to reflect not only the number of features held in common but
also the significance thereof in processing, weighting of features is
applied.
Further, the following processing method has been proposed for the use of
the analogous case.
(1) Regarding a partial problem represented by a feature group which is
shared by the current problem and the case problem as a partial problem
common to the current problem and the case problem, a portion
corresponding to the partial problem is taken out of the solution of the
case.
(2) The taken out partial solution is modified so as to be adapted to the
current problem and is used as a part of the solution of the current
problem.
A plurality of partial problems involved in one problem are not independent
of one another in various fields. Namely, a plurality of features
representing the problem are not independent of one another. Since an
influence relation, an interference relation or a dependency relation
among these features has not been considered in the conventional system,
it has happened sometimes that a case having less utility for solving the
current problem is retrieved.
For example, positioning of individual equipment may be considered as a
partial problem in a layout problem in a computer room. Here, the
positioning of a card reader and a card puncher for katakana characters
(hereinafter referred simply as a card puncher) depends on the positioning
of a console display. That is, an influence relation exists among these
partial problems.
Now, when a set of equipment names to be arranged are used as a feature
group representing the problem, it is assumed that the current problem is
the positioning problem represented by {card reader, card puncher, console
display}, viz., the layout problem of a card reader, a card puncher and a
console display. Further, it is assumed that the case includes a card
reader, and a card puncher, but does not include a console display (for
example, it is arranged in another room sometimes). In this case, the
current problem and the case problem include a card reader and a card
puncher in common, but the positioning of these units in the case is not
applicable in the current problem. Because, the layout of the card reader
and the card puncher varies depending on existence of the console display.
On the other hand, it is assumed that the current problem includes a card
reader and a card puncher and does not include a console display. Further,
it is assumed that the case problem includes a card reader, a card puncher
and a console display. On this occasion, a case can neither be used for
the positioning of the card reader and the card puncher because of the
same reason as described previously. In this manner, the solution of the
case cannot necessarily be used on common features. Therefore, the case
including much common features is not necessarily a useful case.
Further, the existence of the console display determines usefulness of a
case in the above-described example. However, the console display is not
important sometimes depending on the problem. For example, the positioning
of a communication control processor and a power supply unit does not
depend on the positioning of the console display. Accordingly, it is not
important whether the case includes a console display or not when a case
is retrieved in the layout thereof. In this manner, the significance of
the features depends on the combination of a problem and a case, and is
not fixed. Thus, a useful case cannot necessarily be retrieved even when
weighting of features is used.
According to a conventional method of retrieving a case, a case of little
usefulness is retrieved sometimes as described above. Thus, there have
been such problems as waste of time due to repetition of retrieval and
mistakes in determination based on an inappropriate case.
Further, in a conventional method of using a case, all partial solutions of
cases corresponding to common features in a current problem and a case
problem have been used as the partial solution of the current problem.
Thus, it has been required to correct the effect of influence among
features. For example, since the layout in a case of a card reader and a
card puncher cannot be used as it is under the influence of a console
display, the layout thereof has been required to be corrected in
above-described example. Thus, in a conventional method of using a case,
there have been such problems as complicated processing is required for
correcting influence among features, the development period and the
development cost for an inference system gets longer and higher, and
performance and reliability are lowered.
One of the techniques for solving the above-described problems has been
proposed by YOSHIURA et al., YOSHIURA being an inventor of the present
application, at the national convention of Information Processing Society
of Japan. The proposal has been made in "An Approach to Knowledge
Acquisition Bottleneck using Case-Based Reasoning (1),--Case Utilization
Method--", Information Processing Society of Japan, the national
convention collection of papers, 4D-7, pp. 274-275, March (1990) and "An
approach to knowledge Acquisition Bottleneck using Case-Based Reasoning
(2),--Application to Computer Room Layout Problem--", Information
processing Society of Japan, the national convention collection of papers,
4D-8, pp. 276-277, March (1990).
The technique proposed by YOSHIURA et al. will be explained hereinafter. In
this technique, dependency relation information is referred to, features
that are to be included in a case to be retrieved and features that should
not be included therein are obtained and cases are retrieved from a case
base using these features.
In accordance with a case and features to be included in a case to be
retrieved, a solution of a current problem is obtained the cases by using
a portion, of a solution in the case, corresponding to a partial problem
represented by these features.
With reference to FIG. 14, a variable T substituted for dependency relation
information between features and a variable FI substituted for a feature
set which should be included in an analogous case are adopted as input
arguments. A portion related to the inputted current problem among
dependency relation information is substituted for the variable T as an
initial value. Further, the whole feature set of the current problem is
substituted for the variable FI.
The contents of processing will be described hereinafter. In a step 1301, a
feature set which should not be included in an analogous case is obtained
and substituted for a variable FE. Here, a procedure P.sub.roc is a
procedure for obtaining a feature set which should not be included in an
analogous case with the feature set FI to be included in an analogous case
as an argument. The contents of processing in the procedure P.sub.roc will
be described in detail later with reference to FIG. 15.
In a step 1302, a case including all features in the variable FI and no
feature in the variable FE is retrieved from a case base. An appropriate
case can be retrieved by using variables FI and FE. In a step 1303, the
existence of a pertinent case is determined. When a case exists, the case
is delivered to a controller in a step 1304 and processing is stopped
thereafter. When a case does not exist, the processing proceeds to a step
1305.
In a step 1305, a list of features which do not appear in the variable T
among features of a problem is formed, which is substituted for a variable
L. The variable L is a list of features to be removed from the variable
FI.
In a step 1306, it is determined whether the variable L is empty or not. If
the variable L is empty, there is no feature to be removed from the
variable FI, that is, a new variable FI cannot be obtained. Therefore, the
processing is brought to a standstill and returned. If the variable L is
not empty, the processing proceeds to a step 1307.
In the step 1307, one of the features is taken out of the variable L and
substituted for a variable F. In a step 1308, the taken out feature is
removed from the variable L. In a step 1309, the result obtained by
removing the variable F from the variable FI is substituted for a variable
FI'. The variable FI' is a new candidate of a feature set to be included
in an analogous case. When an analogous case is retrieved using the
variable FI', the feature set held in common by the problem and the
analogous case is the variable FI'. Here, the feature of the variable F
removed from the variable FI does not appear in the dependency relation
information T. Therefore, the variable F is not the feature exerting an
influence upon the partial problem corresponding to the variable FI'.
Thus, features which exert an influence upon the partial problem
corresponding to the variable FI' and are included in the feature set of
the inputted problem are all included in the variable FI', viz., the
feature set of the retrieved analogous case.
In a step 1310, a table showing the result obtained by removing the
variable F from the variable T is substituted for a variable T'. In a step
1311, the present processing is executed recursively with the sets FI' and
T' as arguments in place of the sets FI and T. When an analogous case is
found in the process of this recursive execution, the analogous case is
outputted in the step 1304, and the operation is stopped. When an
analogous case is not found, the processing returns to the step 1311
through the step 1306. As a result, the processing proceeds to the step
1306. At this time, in the steps 1306 to 1310, the features that are
different from the last occasion are removed from the set FI, and a case
is retrieved again in the step 1311. When every possibility becomes
exhausted, Yes is determined in the step 1306 in the processing at the top
level, and the whole processing returns to the controller.
FIG. 15 shows the operation of the procedure P.sub.roc of the step 1301
shown in FIG. 14. The procedure P.sub.roc is for obtaining a feature set
which should not be included in an analogous case with the feature set FI
which should be included in an analogous case as an argument. In a step
1401, the set FI is substituted for a set S.
In a step 1402, dependency feature information having no relation with the
current problem is obtained from a group of dependency feature
information. To be concrete, dependency feature information including what
is not the feature of the current problem is found, and a table composed
of the information is formed and substituted for the variable T1.
In a step 1403, dependency feature information with the right member of
which is included in the variable S among the variable T1 is taken out,
and a table consisting of such information is generated and substituted
for the variable T1'. In a step 1404, a set of features included in the
variable T1' is obtained and substituted for the variable S'. A feature
set exerting an influence directly upon a partial problem corresponding to
the variable FI is substituted for the variable S'.
What is to be noted here is that there is the possibility that features
exert an influence indirectly upon the solution of the problem even if
there were no dependency feature information including features and
problems. For example, when A exerts an influence upon B and B exerts an
influence upon C, a dependency relation that A exerts an influence upon C
exists.
Thus, it is needed to obtain a feature set which exerts an influence upon a
partial problem corresponding to the variable FI including such an
indirect influence. For that purpose, it is only required to repeat
processings in the steps 1403 and 1404 until the variable S converges.
In a step 1405, convergence is determined. In the case of convergence, that
is, when features which are not included in the variable S do not exist in
the variable S', the set in which features in the variable FI have been
removed from the variable S is returned to the main routine shown in FIG.
14 as a feature set which should not be included in an analogous case. In
the case of no convergence, that is, when features which are not included
in the variable S exist in the variable S', a union of the sets S and S'
is obtained a new as a set S in a step 1407 and the processing is returned
to the step 1403.
As described above, according to this technique, the set FI is initialized
to the whole feature set of the current problem. The case corresponding to
at least a part of the feature set common to the case problem and the
current problem can be used for solving the current problem. Hence, the
initial value of the set FI is the largest possible set as the use
portion. The set substituted for the set FI is reduced one element at a
time, and retrieval of the case is applied on all such occasions in the
step 1309. As a result, a case in which the use portion reaches the
maximum is retrieved. A part corresponding to a partial problem
represented by the set FI among the solutions of the case is the part that
can be used without compensation for the influence.
In the next place, another case retrieval processing which has been
proposed at the same time will be explained. In the case retrieval
processing, a part which is related to the current problem is picked out
of the dependency relation information table first and is substituted for
the variable T. Further, the total number of the features of the problem
is substituted for a variable N.
With reference to FIG. 16, a variable I is initialized in a step 1501. This
variable I is a counter for selecting one feature from a feature set of a
problem. In a step 1502, it is determined whether all the featues have
been selected completely or not. In the case of Yes in the step 1502, the
processing is returned assuming that there was no analogous case. In the
case of No, the processing proceeds to a step 1503.
In the step 1503, the variable I is increased by "1" only. In a step 1504,
the Ith feature in the feature set of the problem is selected, and a set
having the Ith feature as an element is set in the set S1. In a step 1505,
a table in which dependency feature information with a part of the
variable set T included in the set S1 is collected is set as a set T2. In
a step 1506, a set composed of features included in the set T2 is set as a
variable set S2. With above-described processing, a feature set which
exerts an influence directly upon a partial problem corresponding to the
set S1 is substituted for the set S2.
In a step 1507, a union of the sets S1 and S2 is set as a set S1'. Here, in
order to obtain a feature set which exerts an influence indirectly upon
the set S1, the steps 1505 to 1507 are repeated until the sets S1 and S1'
converge. In a step 1508, convergence is determined. In the case of
convergence, that is, when S1=S1', the processing proceeds to a step 1510.
In the case of no convergence, the set S1 is replaced with the set S1' in
a step 1509, and the processing is returned to the step 1505.
As described, all the features included in the current problem among those
features that exert an influence upon a partial problem corresponding to
the set S1 are included in the set S1. This set S1 becomes a feature set
to be included in an analogous case. When the analogous case is retrieved
using the set S1, the feature set included in common in the analogous case
and the problem is the set S1. Accordingly, those features that exert an
influence upon a partial problem corresponding to the feature set included
in common in the analogous case and the problem and are included in the
problem are all included in the analogous case, too.
In a step 1510, the procedure P.sub.roc shown in FIG. 15 is invoked with
the set S1 as an argument, and the feature set which should not be
included in the analogous case is obtained and set as the variable set FE.
In a step 1511, cases including the set S1 and including no set FE are
retrieved from the case base. In a step 1512, it is determined whether an
analogous case has been found or not. When it is found, the analogous case
is delivered to the controller in a step 1513, thus completing retrieval
processing. When it is not found, the processing is returned to the step
1502, and above-described processing is repeated using other partial
problems.
(a) An Approach to Knowledge Acquisition Bottleneck using Case-Based
Reasoning (1),--Case Utilization System--:
a new system of case-based inference for solving a problem based on past
cases. In general, case-based inference consists of processes for
retrieving the most effective case for solving a current problem among a
plurality of cases (case retrieval processing) and processes for solving
the problem based on the retrieved cases (case using processing). In the
case using processing of the proposed system, a portion which is
applicable as it is to the current problem is brought down, and that
portion is used. In the case retrieval processing, a case in which a
solvable portion on the current problem reaches the maximum is retrieved.
As to the means of realization, the knowledge related to the problem and
the relation among elements of the cases is used, the problem and the
cases are divided into parts, respectively, and the influence relation
among the parts is analyzed, thereby to retrieve an objective case and
further determine the usable parts therein.
(b) An Approach to Knowledge Acquisition Bottleneck using Case-Based
Reasoning (2),--Application to Computer Room Layout Problem--:
a computer room layout system using above-described system (a system of
arranging various equipments in a computer room) is described.
As described above, according to above-described two proposed techniques,
the cases usable without compensation for the influence can be retrieved.
As it is apparent from FIGS. 14 to 16, however, processings are determined
consecutively on respective features and unnecessary elements are reduced
one by one. Thus, there has been a problem that processing for retrieving
usable cases is complicated and takes a long period of time.
Furthermore, in a conventional method of retrieving and using a case, it
has been assumed that all the solutions of a current problem are obtained
from a case. Actually, however, a user desires sometimes to designate a
part of the solution of the current problem or to obtain a part of the
solution by means of another inference system using no cases. Such
requirements could not be met with a conventional system of retrieving and
using a case.
SUMMARY OF THE INVENTION
It is an object of the present invention to provide a case retrieval method
for retrieving cases having large utility for solving a current problem
with priority and a system therefor.
It is another object of the present invention to provide a case using
method which does not require influence compensation among features
against a solution of a case and a system therefor.
It is still another object of the present invention to provide a case
retrieval method when a part of a solution of a current problem has been
predetermined and a remaining partial solution is obtained and a system
therefor.
Furthermore, it is another object of the present invention to provide a
case using method which does not require influence compensation among
features for a solution of a case when a part of a solution of a current
problem has been predetermined and a remaining partial solution is
obtained and a system therefor.
In order to achieve the above-described objects, a method for retrieving
optimum case information for a current problem, includes the steps of:
retrieving candidates of cases having features common to the features of
the current problem from information representing a plurality of cases
stored in a case base in response to an offer of a current problem,
wherein a storage unit storing a case base and respective cases including
the problem of cases and a solution or a solving method for the problem of
the cases;
determining an optimum case candidate out of case candidates based on a
group of features of a current problem, a group of common features and a
group of features of problems of respective case candidates in accordance
with influence relation information representing feature groups that are
influenced by feature groups; and
retrieving case information corresponding to the optimum case candidate
from information representing a plurality of cases.
According to the present invention, it is possible to retrieve case | | |
