JPH08137696A - Case-based reasoning device - Google Patents

Abstract

PURPOSE: To obtain the case.based reasoning device from which an excellent solution is always obtained by selecting similarity between a problem case and a past case properly. CONSTITUTION: The device A is provided with a problem case input device 1 inputting a problem case a, a case storage device 2 storing a past case, a similar case selector 3 selecting a similar case c similar to the received problem case a among past stored cases by referencing a similarity definition prepared in advance defining the similarity between the problem case and the past case, a similarity case correction device 4 correcting solution data of the selected similar case c by referencing a correction knowledge prepared in advance to correct solution data of the similar case in response to the difference between the similar case and the problem case and a similarity adjustment device 5 revising the definition of similarity in response to the addition and revision of the correction intelligence. Thus, an excellent solution is obtained at all times through the configuration above.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a case-based reasoning apparatus, and more particularly to a case-based reasoning apparatus that can be used for machine design, machine failure diagnosis, manufacturing process design, scheduling in general, and the like. .

[0002]

2. Description of the Related Art As shown in FIG. 3, a conventional case-based reasoning device (CBR device) A0 inputs a problem case a, which is data characterizing a current problem, as a problem case input device 1
And a case storage device 2 that stores a past case that is a pair of data characterizing a past problem and answer data of the past problem, and a similarity between the problem case a and the past case are defined in advance. A similar case selecting device 3 for selecting a similar case c similar to the input problem case a from the stored past cases with reference to the prepared similarity definition, a similar case and a problem case Refer to the correction knowledge prepared in advance to correct the answer data of similar cases according to the difference of
The similar case correction device 4 for correcting the answer data of the selected similar case c. Further, the conventional device A0 stores the modification knowledge addition / modification device 8 which additionally modifies the modification knowledge, the similarity definition storage device 6 which stores the previously prepared similarity definition, and the previously prepared modification knowledge. It comprises a correction knowledge storage device 7 for storing and a solution output device 9 for outputting the corrected answer data of the similar case as a solution b for the problem case a. The operation will be described below by taking the case of solving the fare determination problem from Osaka using the conventional apparatus A0 as an example. First, problem case a is
It is a set of pairs of attributes and attribute values that characterize the problem.
For example, the destination is Tokyo, the means is a train, and the departure date is Obon as the problem case a. As a prerequisite, the case storage device 2
It is assumed that there is at least one piece of data (past case) having data having the same structure as problem case a + fare information (= answer data) from Osaka. The similar case selection device 3 refers to the similarity definition and selects a similar case c similar to the problem case a from the case storage device 2. The similarity definition storage device 6 already contains the definition of the similarity between the problem case a and the past case. The correction knowledge storage device 7 already contains knowledge for changing the fare (= answer data) based on the difference between the similar case and the problem case. The similar case correction device 4 uses the correction knowledge to create a fare (= answer data). The modification knowledge addition and modification device 8 additionally modifies modification knowledge as necessary.

Consider now that a system for estimating a fare when traveling from Osaka to another city is constructed using the CBR device A0. Each case has three types of attributes, a1, a2, and a3, and represents a city name (= destination), means, and time, respectively. For example, "go to Hiroshima by bus at the end of the year"
Case C is expressed as C = (Hiroshima, bus, year-end). In addition, the attribute value of the attribute'time 'of case C is a3 (C)
Shall be represented by. In this case, a3 (C) = 'end of year'. Further, the similar case selection device 3 selects the case having the highest similarity SD to the given problem case P, and the similarity definition between the problem case P and a certain case C is expressed by the following equation. Shall be done.

[Equation 1] Here, sdi () indicates the degree of similarity between the attribute values ai (P) and ai (C) of both cases. Furthermore, the similarity SD
Is defined as follows.

[Equation 2] That is, the similarity between cases is considered to be equal to the number of attributes having the same attribute value. For example, for problem case P = (Tokyo, train, Obon), past case C1 = (Tokyo, airplane, Obon) (similarity = 2) is past case C2 (Yokohama, train, June) (similarity It is closer to the problem case than = 1). Such a method of defining the degree of similarity conforms to the daily intuition of human beings and is a method often used in the conventional CBR system.

It is also assumed that the correction knowledge storage device 7 has the correction knowledge of the empirical similar case shown in Table 1 below.

[Table 1] The IF part of the rule indicates the difference in attribute value between the problem case and the similar case, and the THEN part indicates the degree of fare modification in the similar case. For example, Rule 1 indicates that when using a train, the discount ticket cannot be used during the busy season and the regular fare needs to be paid, so that it is about 10% higher than the normal season. Conventionally, the fare from Osaka to the destination is estimated as described above.

[0005]

The conventional case-based reasoning apparatus A0 as described above has the following problems. That is, when the correction knowledge as described above is adopted, the similarity definition represented by the equation (2) is not appropriate. This is because Rule3, which has a low certainty when the case C1 is selected by the similar case selection device 3 as described above.
This is because the case is corrected by (confidence factor 0.5), and there is a high possibility that an appropriate solution cannot be obtained. This example shows that the correction knowledge of similar cases affects the similarity definition. What is also important is that the problem did not occur because the definition of equation (2) was incorrect. No matter how accurately the similarity definition is determined, it is necessary to change the similarity if there is additional change in the correction knowledge of similar cases. In the extreme case, even in two cases where the three attribute values are all significantly different, if the correct (= high confidence) knowledge is added to correct it so that it matches the problem case, the similarity of the cases However, the problem is that the definition of the similarity in the conventional device is fixed and the flexibility to adjust the similarity according to the problem case is lacking, even though the similarity must be changed so that there were. The present invention improves a case-based reasoning apparatus to solve the problems in the conventional arts,
It is an object of the present invention to provide a case-based reasoning device that can always obtain a good solution by making the similarity between a problem case and a past case appropriate.

[0006]

In order to achieve the above object, the present invention provides a problem case input means for inputting a problem case which is data characterizing a current problem, data characterizing a past problem, and It is stored as described above with reference to a case storage unit that stores a past case that is a pair with answer data of a past problem, and a similarity definition prepared in advance that defines a similarity between a problem case and a past case. From similar past cases, similar case selecting means for selecting a similar case similar to the input problem case, and in advance for correcting the answer data of the similar case according to the difference between the similar case and the problem case In a case-based reasoning apparatus equipped with similar case correction means for correcting the answer data of the selected similar case with reference to the prepared correction knowledge, the similarity definition is defined in accordance with the additional change of the correction knowledge. Is configured as a case-based inference apparatus, characterized in that provided a similarity definition change means for further.

[0007]

According to the present invention, a problem case, which is data characterizing the current problem, is input by the problem case input means. A past case, which is a pair of data characterizing a past question and answer data of the past question, is stored in advance in the case storage means. The similarity definition prepared in advance that defines the similarity between the problem case and the past case is referred to, and from the stored past cases, the similar case that is similar to the input problem case is the similar case. Selected by the selection means. Based on the difference between the similar case and the problem case, the correction knowledge prepared for correcting the answer data of the similar case is referred to, and the answer data of the selected similar case is corrected by the similar case correction means. It At this time, the similarity definition changing means changes the similarity definition according to the additional change of the correction knowledge. As a result, the similarity between the problem case and the past case becomes appropriate, and as a result, a good solution can always be obtained.

[0008]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiments embodying the present invention will be described below with reference to the accompanying drawings for the understanding of the present invention. The following embodiments are examples of embodying the present invention and are not intended to limit the technical scope of the present invention. 1 is a schematic diagram showing a schematic configuration of a case-based reasoning apparatus A1 according to an embodiment of the present invention, and FIG. 2 is a schematic diagram showing a schematic configuration of a case-based reasoning apparatus A2 according to another embodiment of the present invention. It is a figure. The same reference numerals are used for elements common to the schematic diagram showing the schematic configuration of the example of the conventional case-based reasoning apparatus A0 shown in FIG. As shown in FIG. 1, a case-based reasoning apparatus (CBR apparatus) according to this embodiment.
A1 is a problem case input device 1 (corresponding to a problem case input means) for inputting a problem case a which is data characterizing the present problem, data characterizing a past problem and answer data of the past problem. The case storage device 2 (corresponding to a case storage unit) that stores a past case that is a pair and a similarity definition prepared in advance that defines the similarity between the problem case and the past case are referred to, and the storage is performed. Depending on the similar case selecting device 3 (corresponding to similar case selecting means) that selects the similar case c similar to the input problem case a from the past cases that have been input, and the difference between the similar case and the problem case. The similar case correction device 4 (corresponding to similar case correction means) for correcting the answer data of the selected similar case c is referred to by referring to the correction knowledge prepared in advance for correcting the answer data of the similar case. In terms of having Is the same as the coming examples. However, the present embodiment is different from the conventional example in that a similarity adjusting device 5 (corresponding to similarity definition changing means) that changes the similarity definition according to the additional change of the correction knowledge is provided.

In addition to the above, the apparatus A1 includes a similarity definition storage device 6 for storing the similarity definition in advance, a correction knowledge storage device 7 for storing the correction knowledge, and a correction knowledge addition for additionally changing the correction knowledge. A changing device 8 and a solution output device 9 for outputting the corrected answer data of the similar case as a solution b are provided. The operation will be described below by taking the case of solving the fare determination problem from Osaka similar to the conventional example using the apparatus A1. Also in this case, it is assumed that the same preconditions as in the conventional example are prepared. Here again, it is considered that a system for estimating a fare when traveling from Osaka to another city is constructed using the device A1. Each case has three types of attributes, a1, a2, and a3, and represents a city name (= destination), means, and time, respectively. For example, the case C of “going to Hiroshima by bus at the end of the year” is expressed as C = (Hiroshima, bus, end of year). Also, the attribute value of the attribute “time” of the case C is represented by a3 (C) (in this case, a3 (C) = “end of year”). Also, the similar case selection device 3 is provided. It is assumed that the case having the highest similarity SD to the problem case P is selected. However, the similarity definition between the problem case P and a case C is expressed by the following expression here instead of the expression (1) in the conventional example.

(Equation 3) Here, “context” is the attribute value of both the problem case and the past case, or the attribute of the attribute value (see Table 1 above).
'Crowded season'and'distance'used in Rule
Etc.) is a set of conditional expressions that take variables.

Further, when the correction knowledge is additionally changed, it is necessary to extract the context from it and reflect it in the similarity definition, but the similarity adjusting device 5 has this function. Basically, the IF part of the added rule is co
The text part and the certainty factor constitute a function s.
It is related to the content of changes in di (). For example, Rule2
Is newly given, the similarity sd1 () of the attribute "destination" which is the only "different" in the IF portion is changed as follows.

[Equation 4] If the similarity sd1 of the above equation (4) is used,
xt1 = {a2 (P) = a2 (C) = train, a3 (P)
= A3 (C)} is satisfied, the attribute value a1 (P)
And a1 (C) are different, the value of the similarity sd1 () is 0.8 (= confidence of Rule2). Therefore, Case C to which Rule2 with a relatively high degree of certainty can be applied
The degree of similarity SD between 2 and the problem case P becomes large, and it becomes easy to use for inference.

Next, another embodiment shown in FIG. 2 will be described. In the case-based reasoning apparatus A2 according to another embodiment, for example, when the similar case selection apparatus 3 and the similarity definition storage apparatus 6 are integrated, or the similar case correction apparatus 4 and the correction knowledge storage apparatus 6 are combined. This is the case when they are integrated. In the latter case, a device 10 for changing the similar case correction device is provided instead of the correction knowledge addition / change device 8. In this case as well, the same operational effects as those of the above-described embodiment apparatus A1 are obtained. The following can be said from the above. For both of the above-described two example devices A1 and A2, the correction knowledge of similar cases affects the similarity definition, because of the reason shown in the conventional example.
If there is an additional change in the correction knowledge of the similar case, the degree of similarity is changed. As a result, even in the extreme case, even if the attribute values of the problem case and the past case are all significantly different, accurate (= high confidence) knowledge to correct it to match the problem case can be obtained. Once added, the similarity can be changed so that the case is more similar. As a result, it is possible to always obtain a good solution by setting the similarity between the problem case and the past case to be appropriate. In the above embodiment, the content of the similarity definition is cout.
Although we decided to reflect the set of conditional expressions called ext,
In actual use, the similarity definition itself may be changed according to additional changes in the correction knowledge.

[0012]

Since the case-based reasoning apparatus according to the present invention is configured as described above, it is possible to always obtain a good solution by setting the similarity between the problem case and the past case to be appropriate. .

[Brief description of drawings]

FIG. 1 is a schematic diagram showing a schematic configuration of a case-based reasoning apparatus A1 according to an embodiment of the present invention.

FIG. 2 is a schematic diagram showing a schematic configuration of a case-based reasoning apparatus A2 according to another embodiment of the present invention.

FIG. 3 is a schematic diagram showing a schematic configuration of an example of a conventional case-based reasoning apparatus A0.

[Explanation of symbols]

 A1, A2 ... Case-based reasoning apparatus 1 ... Problem case input device (corresponding to problem case input means) 2 ... Case storing device (corresponding to case storing means) 3 ... Similar case selecting device (corresponding to similar case selecting means) 4 ... Similar case correction device (corresponding to similar case correction means) 5 ... Similarity adjusting device (corresponding to similarity definition changing means) 8 ... Correction knowledge addition / change device 10 ... Device for changing similar case correction device

Claims (1)

[Claims] 1. A problem case input means for inputting a problem case, which is data characterizing a current problem, and a past case, which is a pair of data characterizing a past problem and answer data of the past problem, is stored. Similar to the input problem case from the stored past cases by referring to the case storage means for storing and the similarity definition prepared in advance that defines the similarity between the problem case and the past case. The similar selected by referring to similar case selecting means for selecting a similar case and correction knowledge prepared in advance for correcting the answer data of the similar case according to the difference between the similar case and the problem case In a case-based reasoning apparatus having similar case correction means for correcting case answer data, similarity definition changing means for changing the similarity definition in accordance with additional change of the correction knowledge is provided. Case-based reasoning device.