JP2924619B2 - Knowledge base creation support device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an expert such as an automatic diagnostic apparatus for estimating a patient's disease from findings and a data analyzing apparatus for estimating a chemical structural formula of a substance from a given molecular formula and a mass spectrum. The present invention relates to a knowledge base creation support device that automatically creates a knowledge base used for a system in place of a KE (Knowledge Engineer).

[0002]

2. Description of the Related Art An important point in constructing a knowledge base of an expert system is how to extract knowledge useful for constructing a knowledge base from an expert, and a specialized engineer (KE) exists for that purpose. Conventionally, when constructing a knowledge base, the KE has been working to highly organize knowledge drawn from experts and to make rules.

In addition, the likelihood (certainty) of each rule at the time of constructing this knowledge base is usually determined arbitrarily by trial and error in order to adjust the entanglement between the rules.

[0004]

However, the prior art having such a structure has the following problems. The construction of a knowledge base of a system for making intellectual decisions, such as an expert system, is an extremely complicated task. However, conventionally, this task is performed by KE, that is, manually, and the burden of KE is enormous. Things.

In addition, the certainty factor of each rule is not determined based on a specific norm such as an arithmetic expression or theory, but KE itself is arbitrarily determined by trial and error. Since it is difficult to specify the meaning of the certainty itself, it is difficult to determine the certainty of each rule. Further, if one certainty is incorrect, for example, the reliability of the knowledge base itself and, consequently, the entire expert system There are also problems such as a decrease in

The present invention has been made in view of such circumstances, and determines the certainty factor of each rule based on a specific criterion, and automatically creates a highly reliable knowledge base using the certainty factor. The purpose of the present invention is to provide a knowledge base creation support device for creating a knowledge base.

[0007]

The present invention has the following configuration to achieve the above object. That is, the present invention is used for an expert system.
What is claimed is: 1. A knowledge base creation supporting apparatus for supporting creation of a knowledge base in which a relationship between a conclusion item and a condition item is expressed by a predetermined rule and stored, wherein (a) a plurality of elements (affiliation elements) belonging to the predetermined conclusion item ) And a plurality of elements (non-affiliation elements) that do not belong to the conclusion item are obtained based on actual data (affiliation data) indicating whether or not the affiliation element meets a predetermined condition item. Determined based on the existence rate of the belonging element that conforms to the condition item (affiliation element condition existence rate) and the result data (non-affiliation result data) indicating whether the non-affiliation element conforms to the condition item.
A condition existence rate fetching means for acquiring the non-affiliation element existence rate (non-affiliation element condition existence rate) which conforms to the condition item in the non-affiliation element; and (b) an affiliation element condition acquired from the condition existence rate acquisition means. Based on the existence rate and the non-affiliation element condition existence rate, for the propriety of the condition item of the new element, a belonging probability calculation means for calculating a belonging probability of whether the new element belongs to the conclusion item or not. (C) a certainty factor determining means for determining a certainty factor in the relationship between the conclusion item and the condition item by using an entropy definition formula based on the membership probability calculated by the membership probability calculating device;
Based on the affiliation probability calculated by the affiliation probability calculating means and the credibility determined by the credibility specifying means, a rule for a relationship between the conclusion item and the condition item to be stored in the knowledge base is created. Rule creation means.

[0008]

The operation of the present invention is as follows. The condition presence ratio capturing means captures, for a relationship between a predetermined conclusion item and a condition item, a belonging element condition existence ratio and a non-affiliation element condition existence ratio obtained from past performance data. The affiliation probability calculating means determines whether the new element belongs to the conclusion item based on the affiliation element condition existence rate and the non-affiliation element condition existence rate. Find the probability. That is, based on the statistical data obtained from the past performance data, for the propriety of the condition item of the new element, the belonging probability of whether the new element belongs to the conclusion item is obtained, The affiliation probability is supported by statistical data and is highly reliable.

Then, the certainty factor specifying means obtains the certainty factor using the entropy definition formula based on the belonging probability. The credibility obtained in this way indicates the certainty as information supported by the information theory.

[0010] Based on the affiliation probability and the certainty factor, the rule creating means creates a rule for a relationship between a predetermined conclusion item and a predetermined condition item to be stored in the knowledge base. And a highly reliable knowledge base is automatically created.

[0011]

An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing the overall configuration of an automatic diagnosis device (expert system) including a knowledge base creation support device according to one embodiment of the present invention. In this embodiment, a case where a knowledge base for an automatic diagnosis apparatus is created will be described as an example. However, the present invention can be similarly applied to a case where a knowledge base of another expert system is created.

The automatic diagnosis apparatus ESP is roughly divided into a knowledge base creation support apparatus 1 which is a main part of the present invention.
And a knowledge base 2, an inference mechanism 3, a database 4, a data input / output device 5, and the like. The knowledge base creation support device 1 includes a communication device 11, a belonging probability calculation unit 12, a certainty factor identification unit 13, and a rule creation unit 14. In addition, the code | symbol JDP shows the performance data processing apparatus which processes the performance data obtained by clinical practice,
The performance data processing device JDP includes a data input / output device 6
1, a data processing unit 62 and a storage device 63. The configuration of each unit will be described in detail below.

The performance data processing device JDP, when performance data obtained clinically is inputted from the data input / output device 61 as needed, stores the performance data in a storage device 63 such as a magnetic disk device. The data processing unit 62 calculates the belonging element condition existence ratio and the non-affiliation element condition existence ratio from the stored and accumulated result data, and stores the results in the storage device 63. As described above, the belonging element condition existence ratio and the non-affiliation element condition existence ratio that are calculated and stored in advance are taken into the knowledge base creation support apparatus 1 via the communication device 11. This communication device 11
This corresponds to the condition existence ratio capturing means in the present invention.

Here, the belonging element condition existence rate and the non-affiliation element condition existence rate will be described with reference to FIG. FIG.
Indicates whether a patient belonging to “disease A” (for example, pneumonia), which is one of the conclusion items in this example, and a patient who does not belong to the condition item, “fever”, which is one of the condition items Shows actual data.

[0015] Diagnostic results obtained in the clinic (for example,
The disease A) and the findings at that time (eg, fever) are input for each patient from the data input / output device 61 of the performance data processing device JDP. The storage device 63 stores one disease (one conclusion item) and one condition item for each condition item.
Two records are created. As described above, when the patient's performance data is input from the data input / output device 61, the data is stored in a predetermined record, and as shown in FIG. Is performed.

In FIG. 2, a patient corresponds to an element in the present invention, and patients Y1 to Ym are patients identified as a disease A, that is, patients belonging to a conclusion item (disease A), and a patient N1. -Patient Nn is a patient who has not been identified as disease A, that is, a patient who does not belong to the conclusion item (disease A). That is, the patients Y1 to Ym and the patient N1
To patient Nn correspond to the belonging element and the non-affiliated element in the present invention, respectively. Further, in the figure, for each patient, whether or not there was fever, ie, a condition item (heat fever)
"+" And "-" indicate whether or not the condition is met.
For example, patients Y1, Y2, Ym and patients N1, N3 are:
Since it is “+”, it indicates that heat generation has occurred, that is, it meets the condition items, and the patient Y3, the patients N2, and Nn
Is "-", so there was no heat generation, that is,
Indicates that the condition item is not met. The result data indicating whether or not the condition item is appropriate for each of the patients Y1 to Ym and the patients N1 to Nn corresponds to the belonging result data and the non-affiliation result data in the present invention.

As described above, when the result data indicating whether or not the patient A belongs to the disease A and the plurality of patients who do not belong to the disease A accumulate the heat or not, the data processing unit 62 of the result data processing apparatus JDP. Calculates the belonging element condition existence rate and the non-affiliation element condition existence rate as follows.

The belonging element condition existence rate (YJS)
The number of patients having a fever out of the patients who have been identified as と is divided by the number of patients who have been identified as having a disease A.
On the other hand, the non-affiliation element condition existence rate (NJS) is obtained by dividing the number of patients who have developed heat among the patients who have not been identified as disease A by the number of patients who have not been identified as disease A. For example, eight patients (m =
8) If there are 6 patients with fever, Y
JS is 0.75 (6 ÷ 8), and if 10 patients (n = 10) were not recognized as disease A, among which 4 patients had fever, NJS 0.40 (4 ÷
10).

That is, the belonging element condition existence rate and the non-affiliation element condition existence rate are obtained by statistically calculating the correlation between the disease A and the heat generation from the past performance data. In addition,
The patients N1 to Nn are patients who have not been identified as the disease A, but are not necessarily normal patients, and include, for example, patients who have been identified as another disease B. Therefore, the patients N1 to Nn include cases in which fever has developed due to a disease other than the disease A.

The belonging element condition existence rate (YJS) and the non-affiliation element condition existence rate (NJ
S) is taken into the knowledge base creation support apparatus 1 via the communication device 11.

It is to be noted that the knowledge base creation support apparatus 1 is provided with YJS
The configuration other than the communication device may be adopted in order to take in the NJS and NJS. For example, by inputting the previously determined YJS and NJS from the input device, the knowledge base creation support device 1 may be configured to import the information.
A storage medium (for example, a flexible disk) storing JS and NJS may be read from a predetermined driver (for example, a flexible disk driver) and loaded into the knowledge base creation support device 1. In the case of such a configuration, the input device and the driver (flexible disk driver) serve as the condition presence ratio capturing means.

Next, based on the YJS and NJS fetched via the communication device 11, the affiliation probability calculation unit 12 calculates
With respect to whether or not the new patient (element) has a fever (condition item), the belonging probability of whether or not the new patient should be recognized as the disease A (whether or not belongs to the conclusion item) is determined. The affiliation probability calculation unit 12 corresponds to affiliation probability calculation means in the present invention.

For example, when a new patient is examined and fever develops, the probability that the patient has disease A (affiliation probability: SK +) is determined by substituting the above-mentioned YJS and NJS into the following equation. SK + = YJS ÷ (YJS + NJS) (1) When a new patient has no fever, the probability that the patient has a disease A (affiliation probability: SK−) is calculated by the following equation. Determined by substitution. SK − = (1−YJS) ÷ ((1−YJS) + (1−NJS)) (2) That is, in the belonging probability calculation unit 12, there is a fever (condition item) for a new patient. The probability of the disease A is stochastically determined from past performance data with respect to whether or not. Therefore, this belonging probability is supported by statistical data obtained from past performance data.

The membership probabilities SK + and SK- may be calculated by the following equations instead of the equations (1) and (2). SK + = (YJS × m) ÷ ((YJS × m) + (NJS × n))... (1) ′ SK − = (1− (YJS × m)) ÷ ((1− (YJS × m)) ) + (1− (NJS × n))) ……… (2) ′

Here, m is the number of patients identified as disease A (number of elements belonging to the conclusion item), and n is the number of patients not identified as disease A (number of elements not belonging to the conclusion item). Number). That is, in the above equations (1) ′ and (2) ′, the belonging probability is calculated in consideration of the weight according to the number of each element. For example, a large number of actual data can be reflected on the belonging probability, for example, when m is 10 and n is 10000. In this case, m and n are also taken in from the performance data processing device JDP.

Next, based on the affiliation probabilities (SK +, SK−) obtained by the affiliation probability calculation unit 12, the certainty factor specifying unit 13 calculates
Using the entropy definition formula, a certainty factor in the relationship between the conclusion item and the condition item is obtained. This certainty specifying unit 13
Corresponds to the certainty factor specifying means in the present invention.

For example, when a new patient has a fever, the affiliation probability (SK +) belonging to the disease A is substituted into the entropy definition formula shown below to obtain the entropy E +. E + = − (SK +) × log ₂ (SK +) − (1− (SK +)) × log ₂ (1− (SK +)) where 0 ≦ (SK +) ≦ 1 (3)

Next, E + obtained by the above equation (3) is substituted into the following equation to obtain a certainty factor CF +. CF + = 1− (E +) (4) The confidence factor CF + obtained by the above equation (4) is 0 ≦ (CF +)
≦ 1, indicating the certainty as information supported by the information theory.

Here, the belonging probability SK + calculated by the belonging probability calculating section 12 and the entropy E + calculated by the above equation (3)
Is shown in FIG. 3A, and the relationship between the belonging probability SK + and the certainty factor CF + obtained by the above equation (4) is shown in FIG. 3B.

As can be seen from FIG. 3A, when SK + is "0.5", the entropy E + (fluctuation) is the highest at "1". In other words, the probability of SK + indicates that it is uncertain whether or not the item belongs to the conclusion item, that is, the probability of information is low. Incidentally, SK + becomes "0.5" when the belonging element condition existence ratio (YJS) and the non-member element condition existence ratio (NJS) both have the same value. At this time, for example, when the new patient has a fever, the probability of the disease A is the same as the probability of not having the disease A. Therefore, the patient has the disease A depending on whether or not the new patient has fever. It is difficult to find out that it is consistent with the above results. Then, as can be seen from FIG. 3B, when SK + is “0.5”, that is, when there is no value as information, the certainty factor CF + becomes “0”. The certainty indicates the certainty of a rule described later, and indicates that the certainty decreases as the value approaches “0”.

For example, when SK + is "0" or "1", the entropy E + (fluctuation) is the lowest at "0". In other words, the probability of SK + indicates that it is certain that the item belongs to the conclusion item, that is, the probability of information is high. By the way, SK + is "1"
Means that the belonging element condition existence rate (YJS) is “1”,
This is the case where the non-belonging element condition existence rate (NJS) is “0”. At this time, for example, if a new patient develops fever, it can be concluded that the disease is always disease A. When SK + is “0”, the reverse is true. Therefore,
As can be seen from FIG. 3A, as SK + approaches “0” from “0.5” or “1” from “0.5”, as described above, the likelihood of information becomes higher. Coincides with becoming. However, it indicates that the probability of not being disease A increases as SK + approaches “0” from “0.5”, and that the probability of disease A increases as “SK +” approaches “1” from “0.5”. It indicates that it will be higher.
Then, as shown in FIG. 3B, SK + is “0.5”.
As the value approaches from “0” or from “0.5” to “1”, the certainty factor CF + approaches from “0” to “1”.

When the new patient has no fever, the belonging probability (SK−) belonging to the disease A is
Similarly to the above-mentioned belonging probability (SK +), entropy E− is obtained by substituting (SK−) in place of (SK +) in equation (3), and (E +) in place of (E +) in equation (4). −)
Is substituted to obtain a certainty factor CF−. The meanings of the obtained entropy E− and certainty factor CF− are the same as in the case of the membership probability (SK +) described above.

It should be noted that the method of obtaining the confidence factors CF + and CF- from the entropies E + and E- is not limited to the above equation (4), but may be obtained by other methods. For example, using the reciprocal of the entropy E +, E−, the certainty C
An equation may be established in which F + and CF− converge between “0” and “1” to calculate the certainty factors.

Next, the rule creator 14 creates the knowledge base 2 based on the affiliation probabilities SK + and SK− obtained by the affiliation probability calculator 12 and the credibility CF + and CF− obtained by the credibility identifier 13. Create a rule to store the relationship between the conclusion item and the condition item. The rule creating unit 14 corresponds to a rule creating unit according to the present invention.

As described above, the belonging probability SK + (or SK−) is “0.5” or “0” or more and “0.
The meaning of each is different depending on whether it is less than "5" or more than "0.5" and equal to or less than "1". Therefore, the rule creation unit 14 creates a rule based on the value of the belonging probability SK + (or SK−) as shown below.

(1) When SK + is "0.5". In this case, no rule is created because the information is not valuable.

(2) SK + exceeds “0.5” and “1”
If: In this case, if the new element satisfies the condition item, it indicates that it belongs to the result item with the obtained certainty factor CF +, so the following rule is created. if (condition item) = + then (conclusion item) = + CF = CF + Here, “+” indicates affirmation.

For example, if the obtained confidence factor CF + is “0.9”, which is applied to FIG. 2 described above, then if (thermal) = + then (disease A) = + CF = 0.9. This is a rule indicating that if a new patient has fever, he / she is a disease A with a certainty of “0.9”.

(3) When SK + is equal to or more than “0” and less than “0.5”. In this case, if the new element satisfies the condition item, it indicates that it does not belong to the result item with the obtained certainty factor CF +, so the following rule is created. if (condition item) = + then (conclusion item) = − CF = CF + Here, “−” indicates negation.

For example, if the confidence factor CF + obtained by applying to FIG. 2 described above is “0.9”, then if (thermal) = + then (disease A) = − CF = 0.9. This is a rule indicating that if a new patient has fever, it is not a disease A with a certainty of “0.9”.

Also, a rule is created for the belonging probability SK− in the same manner as the above-described belonging probability SK +. The rules created are shown below. (1) When SK- is "0.5" No rule is created.

(2) SK− exceeds “0.5” and “1”
If: if (condition item) = − then (conclusion item) = + CF = CF− This means that if the new element does not meet the condition item, the confidence factor belongs to the conclusion item with certainty of CF−. It is a rule shown.

(3) When SK− is equal to or more than “0” and less than “0.5”. if (condition item) = − then (conclusion item) = − CF = CF− This means that if the new element does not meet the condition item, the certainty factor does not belong to the conclusion item with certainty of CF−. It is a rule shown.

The rules created are described in the simplest form without any logical connectors (and, or) internally. This is because, as will be described later, when the inference mechanism 3 infers the same conclusion from two or more condition items, the inference is uniformly performed by rule OR.

In this way, two rules for SK + and SK− (for the case where the condition item is matched and for the case where it is not matched) are created for the predetermined conclusion item and the condition item. However, if the belonging probability is “0.5”, no rule is created. The created rules are stored in the knowledge base 2.

The knowledge base 2 creates and stores the rules for the relationship between a plurality of conclusion items and a plurality of condition items in accordance with the above-described procedure. For example, "Disease A"
In response to this, rules for relations with condition items such as “heat fever”, “pain”, “sex”, and “age” are respectively created. Similarly, for “disease B” and “disease C”, a plurality of Rules relating to the condition items are created, and the rules are stored in the knowledge base 2.

The data input / output device 5 includes a CRT 51 and a K /
B (keyboard) 52 and the like. CRT51
The display of the condition items from the inference mechanism 3 and the inference result are displayed, and the physician reports the diagnosis result of the patient (new patient) to the condition items displayed on the CRT 51 by the K / B5.
2 is input.

The database 4 stores auxiliary data such as condition items to be displayed on the CRT 51 of the data input / output device 5, for example.

The inference mechanism 3 sequentially displays the condition items stored in the database 4 on the CRT 51, and when a diagnosis result (response) corresponding to the condition item is input from the K / B 52, the inference mechanism 3 determines the knowledge base 2 based on the data. While searching for, make inferences and estimate the conclusion (disease). This inference scheme is
It may be backward inference or forward inference. When inferring the same conclusion from a plurality of condition items, the conclusion is inferred by rule OR. In this rule OR, the confidence factors of the condition items a, b,.
If a, CFb,..., CFn, the certainty factor CFx of the conclusion obtained by inference can be obtained by the following equation. CFx = (CFa + CFb + ... + CFn)-(CFa * FCb * ... * CFn) (5)

For example, when the condition item a (for example, heat generation) is satisfied, the certainty factor of the conclusion item A is “0.8”, and when the condition item b (for example, pain) is satisfied, the conclusion item A is satisfied. Is 0.5, CFx = (0.8 + 0.5)-(0.8 × 0.5) = 0.9, which satisfies the condition item a and the condition When the item matches the item b, the certainty factor that is the conclusion item A is “0.9”. That is, as a patient is diagnosed as to whether or not it satisfies many condition items, the patient's disease can be more accurately estimated.

For example, when the condition item a is satisfied, the certainty factor of the conclusion item A is “1”, and the condition item b
When the confidence level of the conclusion item A is "0.1"
, CFx = (1 + 0.1) − (1 × 0.1) = 1, and the confidence factor that is the conclusion item A when the condition item a is satisfied and the condition item b is satisfied is “ 1 ". This is the case where the disease A is always present if the condition item a is satisfied, regardless of the result of the other condition items. It turns out that "1" is obtained.

For example, when an inference mechanism that uses a rule logical product as shown in the following equation to estimate a conclusion item from a plurality of condition items is used, there are the following inconveniences. CFx = (CFa.times.FCb.times..times.CFn) For example, when the certainty factor is substituted into the above condition, CFx = (1.times.0.1) = 0.1. This means that if condition a is met, then the patient
Nevertheless, there is an inconvenience that if one attempts to infer a conclusion from a plurality of conditions, the certainty will decrease. On the other hand, if the inference is performed using the rule logical sum as described above, the condition is not affected by other conditions, and if the condition item a is satisfied, the patient can be inferred to have the disease A.

Further, by inferring with the rule logical sum in this way, as described above, each rule can be described in a simple form without a logical connector or the like internally, and the entanglement between the rules becomes complicated. Since this is not done, there is also an effect that the creation of rules becomes easy.

When the inference mechanism 3 cannot perform inference by performing a rule OR operation, for example, the rule creation unit 1
At 4, the degree of certainty for a plurality of condition items is stored,
As described below, a rule may be created in which each condition is connected by a logical connector "and" and stored in the knowledge base 2. if ((condition item a = +) and (condition item b = +) and ... and (condition item n = +)) then (disease A) = + CF = CFx Note that CFx is obtained by the above equation (5). It is something that can be done.

Next, the operation of the automatic diagnostic apparatus having such a configuration will be described below. First, the rules stored in the knowledge base 2 are as follows. The belonging element condition existence ratio and the non-affiliation element condition existence ratio stored in the storage device 63 are sequentially taken into the knowledge base creation support device 1 via the communication device 11. Rules for the relation between each conclusion item and the condition item are sequentially created and stored in the knowledge base 2.

Next, a new patient is diagnosed based on the rules stored in the knowledge base 2. First, the inference mechanism 3 sequentially stores the condition items in the CRT 5 of the data input / output device 5.
1 is displayed. The physician determines whether or not the patient meets the condition item by the K / B5 of the data input / output device 5.
Input from 2.

For example, the patient's gender, age, etc. are input, and whether or not there is fever, whether or not there is a pain, and various condition items such as diagnosis results are sequentially input according to the display on the CRT 51. . In the inference mechanism 3, the input by the doctor
The result of whether or not the condition item is satisfied is compared with the rules stored in the knowledge base 2 to infer which disease the patient has. At this time, when inferring a disease from a plurality of condition items, the inference is made by the rule OR as described above.

Using this apparatus, a diagnosis was made by inputting the suitability of 16 condition items such as sex, age, and pain for 40 bone patients whose clinical diagnosis had already been made. Table 1 below shows a comparison between the diagnosis result obtained by the apparatus of this embodiment and the clinical diagnosis result.

[0059]

[Table 1]

In the table, "1st" indicates a case in which the diagnostic result obtained first by the apparatus of this embodiment matches the clinical diagnosis result, and "2nd" indicates a case obtained second by the apparatus in this embodiment. Cases in which the results agreed with the clinical diagnosis results, "mi
"ss" is a case in which the diagnosis result obtained by the apparatus of this example does not match the clinical diagnosis result, and "unown" is
The cases in which a diagnosis result was not obtained by the apparatus of this embodiment are shown. The minimum and the maximum of the certainty factor indicate the minimum and the maximum of the certainty factor for each condition item for each disease, and the average is the average value of the certainty factors for all the condition items for each disease.

As is clear from this table, "1st"
However, it can be seen that 26/40 cases and “2nd” could infer the patient's disease with a high hit rate of 8/40 cases.
Also, it can be seen that a high hit rate is shown for a disease having a high degree of certainty.

In this method, a rule for relating a conclusion item and a condition item is created using statistical data obtained from past performance data and a certainty factor supported by information theory. This is because a high knowledge base was created.

When updating the rules of a knowledge base that has already been created, the conventional apparatus has a disadvantage that the operation is extremely complicated. That is, in the conventional device, the knowledge base is constructed by adjusting the entanglement between the rules using the certainty determined arbitrarily by the KE. For example, the certainty of one rule is updated. Even in this case, all rules had to be reviewed in consideration of the effect on other rules, and the certainty of the affected rules had to be adjusted accordingly. Also, the more complicated the knowledge built in the knowledge base is, the more complicated the task becomes. Therefore, the burden on the KE is enormous.

On the other hand, in this embodiment, the rule can be updated automatically even when the rule is updated.
No burden on KE. An example of a rule update procedure according to this embodiment will be described below. First, when new performance data is input to the performance data processing device JDP, the performance data, the belonging element condition existence ratio, and the non-affiliation element condition existence ratio stored in the storage device 63 are updated. Next, each disease (conclusion item) stored in the storage device 63, the belonging element condition existence rate and the non-affiliation element condition existence rate for each condition item are sequentially supplied to the knowledge base creation support apparatus 1.
The plurality of belonging element condition existence rates and the non-affiliation element condition existence rates supplied to the knowledge base creation support device 1 include updated ones. The knowledge base creation support apparatus 1 creates rules based on the sequentially acquired belonging element condition existence rates and non-affiliated element condition existence rates, and updates the entire knowledge base 2. When a part of the result data is updated, a rule that affects the update of the result data may be searched, and only the rule may be updated.
The rules created in the knowledge base 2 are independent for each conclusion item and each condition item, and there is no entanglement with other rules. it can.

In the above embodiment, the belonging element condition existence rate and the non-affiliation element condition existence rate are configured to be fetched via the communication device 11. For example, the knowledge base creation support apparatus 1 It may be configured as shown in FIG. The apparatus shown in FIG. 4 takes in the belonging data and the non-affiliated data from the communication device 11 (or an input device or a flexible disk device, etc.), Based on the data, the data processing unit 6 of the performance data processing device JDP
2, the belonging element condition existence ratio and the non-affiliation element condition existence ratio are calculated, and the calculation results are assigned to the belonging probability calculation unit 12.
It is configured to give to. Even with this configuration, it is possible to create the same rules as in the above-described embodiment. Note that, in the configuration of FIG.
1 (or an input device, a flexible disk device, or the like) and the condition presence ratio calculation unit 15 correspond to a condition presence ratio capturing unit in the present invention.

[0066]

As is clear from the above description, according to the present invention, the knowledge base used for the expert system is automatically created, so that the burden on the KE can be reduced. Further, since the certainty of each rule is determined based on statistical data and information theory, and the rule is stored using the certainty in the knowledge base, a highly reliable knowledge base can be created. .

[Brief description of the drawings]

FIG. 1 is a block diagram showing an overall configuration of an automatic diagnosis device (expert system) including a knowledge base creation support device according to an embodiment of the present invention.

FIG. 2 is a diagram for describing belonging result data, non-belonging result data, belonging element condition existence ratio, and non-belonging element condition existence ratio.

FIG. 3 shows the relationship between the belonging probability and the obtained entropy,
It is a figure which shows the relationship between the belonging probability and the obtained certainty factor.

FIG. 4 is a block diagram showing a configuration of a modification of the embodiment apparatus shown in FIG. 1;

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 ... Knowledge base creation support apparatus 2 ... Knowledge base 3 ... Reasoning mechanism 11 ... Communication device 12 ... Membership probability calculation part 13 ... Confidence identification part 14 ... Rule creation part 15 ... Condition existence rate calculation part EPS ... Automatic diagnostic apparatus JDP ... Actual data processing device YJS… belonging element condition existence rate NJS… non-member element condition existence rate

Claims (1)

(57) [Claims] 1. A knowledge base creation support device used for an expert system, which supports creation of a knowledge base in which a relation between a conclusion item and a condition item is expressed by a predetermined rule and stored, and (a) a predetermined conclusion For a plurality of elements belonging to the item (affiliation element) and a plurality of elements not belonging to the conclusion item (non-affiliation element), the actual data indicating whether the belonging element satisfies a predetermined condition item (affiliation element) Existence rate (affiliation element condition existence rate) of the belonging element that satisfies the condition item in the belonging element, which is obtained based on the actual data
And the existence rate (non-affiliation element) of the non-affiliation element that satisfies the condition item in the non-affiliation element, which is obtained based on the performance data (non-affiliation performance data) as to whether the non-affiliation element satisfies the condition item. (B) a condition element existence rate and a non-affiliation element condition existence rate acquired from the condition existence rate acquisition means. For the propriety, a membership probability calculating means for calculating a membership probability of whether or not the new element belongs to the conclusion item; and (c) an entropy definition formula based on the membership probability calculated by the membership probability calculating means. (D) a confidentiality degree determined by the affiliation probability calculating means and a conviction degree determined by the confidentiality degree determining means. Preparative the basis, the knowledge base for storing, the conclusion items and knowledge base creation support apparatus is characterized in that a rule generation means for generating rules of the relationship between the condition item.