JPH04213113A - Expert system - Google Patents

Abstract

PURPOSE:To obtain a batch question type expert system by which it is not necessary to input unnecessary information, and by which the burden of an inputting operation by a user can be reduced by collectively inputting information necessary to an inference before starting the inference and operating the inference for solving a problem based on a truth or a rule 10 a knowledge base by an inference engine while referring to this input information. CONSTITUTION:This system is equipped with an attribute relation extracting part 3 which extracts the relation between each attribute by retrieving a rule stored in a knowledge base 2, and prepares a question data in which the other attributes or attribute values which are not needed to be inputted are discribed when the prescribed attribute or attribute value is inputted. And also, this system is equipped with a question answer controlling part 4 which extracts the attributes or the attribute values which are not needed to be inputted by referring to the above mentioned question data, when the information concerned with the attribute value is inputted at the time of the batch input of the above mentioned information.

Description

[Detailed description of the invention]

0001

[Industrial Application Field] The present invention inputs information necessary for inference all at once before starting inference, and while referring to this input information, an inference engine solves a problem based on facts and rules in a knowledge base. Concerning a bulk question type expert system that performs inference.

0002

2. Description of the Related Art In expert systems, when information for inference is insufficient in the working memory, the user is asked about the missing information to obtain the necessary information. There are two types of inquiry methods for this information: sequential inquiry type and batch inquiry type.

[0003] The sequential question type is a method in which an inquiry is made to the user each time a lack of information occurs during inference. This method is realized by starting a daemon that asks the user questions if the working memory is searched and information necessary for inference is insufficient when determining rule conditions.

[0004] The batch question type is a method in which the user is asked to input all information deemed necessary for the inference before the inference is started, and no inquiries are made to the user during the inference.

[0005]

[Problem to be Solved by the Invention] However, in the case of the sequential question type, the inference cannot proceed unless the user inputs information, so even if the inference takes a considerable amount of time, the user is always asked questions from the system. The problem was that they had to be on standby in preparation for.

On the other hand, in the case of the batch question type, the user does not have to wait during the inference, but since it is necessary to input all the information considered necessary before starting the inference, it is not used in the actual inference. There was a problem that even information was required to be input, making the input operation cumbersome for the user. Therefore, it is necessary to reduce the burden on the user as much as possible when inputting information.

[0007] The present invention has been made based on the above-mentioned circumstances, and an object of the present invention is to provide a batch question type expert system that eliminates the need to input unnecessary information and reduces the burden of input operations on the user. .

[0008]

[Means for Solving the Problems] FIG. 1 shows a principle embodiment of the expert system of the present invention. The present invention inputs information necessary for inference all at once before starting inference, and uses inference engine 1 to create knowledge base 2 while referring to this input information.
Knowledge base 2
Attribute relationships that search rules stored in , extract relationships between each attribute, and create question data that describes other attributes or attribute values that do not need to be entered when a given attribute or attribute value is entered. The present invention includes an extraction unit 3 and a question answering control unit 4 that refers to the question data and extracts attributes or attribute values that do not require input when information regarding attribute values is input during the batch input of the information. This is a characteristic feature.

[0009]

[Operation] At the stage when facts and rules for problem solving are stored in the knowledge base 2, the attribute relationship extraction unit 3 is activated. The attribute relationship extraction unit 3 searches all rules stored in the knowledge base 2, extracts relationships between attributes, and collects attribute relationship data as illustrated in FIG. 2, for example. Note that the attribute relationship data in FIG. It means "is".

[0010] The attribute relationship extracting unit 3 extracts data for each corresponding attribute based on the collected attribute relationship data, for example, as shown in FIG.
Create question data as shown below. Figure 3(a) is
This is question data related to attribute name = "Attribute A", and this "
There are three possible values for "attribute A": (a1, a2, a3), and "when attribute value a1 is input, there is no need to input attribute B", and "attribute value a2 is input". , it means that there is no need to input attribute B. In addition, in Figure 3(b), attribute name = “Attribute B
", and the possible values of this "attribute B" are (b1, b2), and "If attribute value a1 is input for attribute A, no input is required for this attribute B.""Thereis" and "For attribute A, attribute value a
When 2 is input, it means that there is no need to input for this attribute B.

The expert system of the present invention can be used as an inference system after creating question data as shown in FIG. 3 as described above.

[0012] Prior to the start of inference, when the user inputs information necessary for inference all at once, this input information is stored in working memory 5 through question answering control section 4. The question answering control unit 4 searches the question data created by the attribute relationship extraction unit 3 every time information regarding attributes or attribute values is input by the user, and determines whether input is required due to the input attribute or attribute value. Check if there are any other attributes or attribute values.

If there are other attributes or attribute values that do not require input, the question answering control section 4 extracts them and displays them to the user in a desired manner. Therefore, the user inputs the information necessary for inference all at once.
Information that is no longer required to be input at that point can be known in real time, and it is possible to eliminate the input of wasteful information that is not necessary for inference, which has been a problem in the past.

[0014] Note that the attribute or attribute value that no longer requires input may be displayed to the user to simply alert the user, or the user may be prompted to input the attribute or attribute value that no longer requires input. It is sufficient to adopt any processing form according to the specifications of the system, such as not accepting .

[0015]

[Examples] Examples of the present invention will be described below. First, as a first embodiment of the present invention, an example will be described in which when the attribute value of a certain attribute is input, there is another "attribute" that does not require input.

Before giving a detailed explanation of the embodiment of the expert system of the present invention, the overall processing flow in the expert system of the present invention will be briefly explained with reference to the flowchart of FIG.

In the case of the expert system of the present invention, at the stage when the knowledge base 2 is created, the attribute relationship extraction unit 3
operates to extract relationships between attributes of all rules stored in the knowledge base 2. Then, the attribute relationship extraction unit 3 extracts the extracted attribute relationship from the attribute relationship shown in FIG. 3(a).
) Create question data such as (b) (step S1
01).

When the question answering process for receiving information necessary for inference is started, the question answering control unit 4 displays a question screen based on the created question data and receives only the information necessary for inference from the user. Receive (step S102),
Write this information into the working memory 5 (step S
103). In this way, when all the information necessary for inference is input by the user, the inference engine 1 is activated and inference for problem solving is started (step S104).

Next, the question data creation process in the attribute relationship extraction section 3, which is one of the features of the present invention, will be explained with reference to the flowcharts of FIGS. 5 and 6. In addition,
As a prerequisite for creating the question data in Figures 3(a) and (b), the knowledge base 2 contains the following rules (rule 1).
) to (rule 4) is stored together with other rule sets that are not specified. Each rule is expressed by a "condition part" and a "consequence part".

[Rule set α] (Rule 1) Conditional part: Value of target attribute A is a1 Consequence part: Execute technique W1 (Rule 2) Conditional part: Value of target attribute A is a2 Consequence part: Technique Execute W2 (Rule 3) Condition part: Value of target attribute A is a3 and target attribute B
The value of is b1 Consequence part: Execute technique W3 (Rule 4) Condition part: The value of target attribute A is a3 and target attribute B
The value of is b2 Consequence: Execute technique W4

In the above rule, it can be seen that the value of attribute B is required only when the value of attribute A is a3. That is, attribute B can be said to be unnecessary information when the value of attribute a is a1 or a2. Therefore, the attribute relationship extraction unit 3 extracts such relationships between attributes, and creates question data shown in FIGS. 3(a) and 3(b) based on the extracted relationships between attributes.

The processing operation will be explained in detail below with reference to FIGS. 5 and 6. In step S201, if all rule sets in the knowledge base 2 have been examined, the process advances to step S208; otherwise, the process advances to step S202. Note that, as described above, it is assumed that the knowledge base 2 stores the [rule set α] and a plurality of other rule sets that are not specified.

[0023] In step S202, knowledge base 2
One of the rule sets, for example [rule set α], is extracted. Then, in step S203, all attributes of the condition part of each rule of this [rule set α] are extracted and recorded as an attribute list (ATTR-LIST). In the case of [rule set α], all attributes extracted from the condition part of each rule are attribute A and attribute B, and these attributes A and attribute B are stored in the attribute list of [rule set α]. Ru.

In step S204, if all the rules are checked, the process goes to step S201; if not, the process goes to step S205, where one rule in [rule set α] is checked.
Take out one. Then, in step S206, it is checked whether or not there is an attribute in the attribute list that is not described in the condition section of the extracted rule.

For example, (rule 1) of [rule set α]
Taking the case of (Rule 1) as an example, the attribute of the condition part is A
It is. In addition, the attribute list includes attribute A as described above.
and attribute B are described. Therefore, (Rule 1)
"Attribute B" is detected as an attribute that is not described in the condition part. Similarly, "attribute B" is detected in (Rule 2) as well. On the other hand, in (Rule 3) and (Rule 4), since both attribute A and attribute B are described in the condition part, there is no detected attribute. If there is an attribute that is not described in the condition part of the extracted rule, the process proceeds to step S2.
The process advances to step S207, and if there is no such information, the process advances to step S204.

In step S207, all attributes not described in the condition part of the detected rule, attributes and their values described in the condition part of the rule at that time are recorded in the result data. For example, the above (Rule 1
), attribute B is the attribute name that is not written in the condition part of the rule, and attribute A and attribute value a1 are the attributes and their values that are written in the condition part of the rule at that time as the result data. recorded.

After recording the resulting data, the process returns to step S204 and steps S205 to S207 are performed.
The following process is executed for all rules in [rule set α].

The process performed on the above [rule set α] is similarly executed on other rule sets stored in the knowledge base 2 (step S201), and the above process is performed on all the rule sets. When finished, step S
Proceed to 208.

In step S208, it is checked whether there is any resultant data, that is, whether there is an attribute in the attribute list that is not described in the condition part of the rule, and if there is not, the process is terminated. If not, the process advances to step S209.

[0030] In step S209, if all the resultant data has been examined, the process ends; otherwise, the process advances to step S210. In step S210, one piece of resultant data is extracted, and in step S211, it is checked whether the extracted data have a common attribute. If there is a common attribute, the process advances to step S212;
If not, the process returns to step S209.

In step S212, the undescribed attributes, common attributes, and attribute values of the common attributes are recorded in the attribute relationship data, and the process returns to step S209. Then, the processes of steps S210 to S212 are repeated, and the process ends when all data has been examined.

As described above, the attribute relationship extraction unit 3
The relationship between the attributes of the rules stored in the knowledge base 2 is checked to obtain attribute relationship data. for example,[
rule set α], attribute relationship data as shown in FIG. 2 is obtained by the above processing.

The attribute relationship extraction unit 3 then creates question data shown in FIGS. 3(a) and 3(b) based on the attribute relationship data shown in FIG. In the question data, attribute names and their attribute values are recorded. Then, it records what attribute value of that attribute is and what other attributes are not required to be input. Also,
If the attribute does not require input due to the attribute value of another attribute, this is also recorded. In this way, Figure 3(a)
) Question data as shown in (b) is created.

Next, the question answering process in the question answering control unit 4, which is another feature of the present invention, will be explained with reference to the flowchart shown in FIG.

The question answering control unit 4 displays a question screen as shown in FIG. 8 based on the question data of FIGS. 3(a) and 3(b) obtained as described above, and provides information necessary for inference to the user. , and stores it in the working memory 5.

First, in step S301, a question screen based on the current settings is displayed. An example of this question screen is shown in FIG. The question screen in FIG. 8 is a setting screen based on the above-mentioned [rule set α]. Note that FIG. 8 is displayed as including not only the rules regarding the attributes A and B described above but also the rules regarding the attribute C.

In step S302, all attribute items are set as P, attribute items for which attribute values have been input are set as Q, and attribute items that do not require input are set as R.

In step S303, if the user instructs to end the input, the process advances to step S314; otherwise, the process advances to step S304.

In step S304, it is assumed that the user inputs the attribute value a1 of attribute A.
In 305, it is checked whether another attribute value a2 or a3 has already been set for the input attribute A. If either attribute value a2 or a3 has already been set for attribute A, the process advances to step S306;
If it has not been set, the process advances to step S309.

In step S306, it is checked whether the already set attribute value a2 or a3 makes it unnecessary to input another attribute B or C. If it is unnecessary, the process advances to step S307; otherwise, the process advances to step S310. For example, if the attribute value a2 has already been input, this attribute value a2 makes it unnecessary to input the other attribute B, as shown in FIG. 3A, so the process proceeds to step S307. Furthermore, if the attribute value a3 has already been input, the process proceeds to step S310, since this attribute value a3 does not require any other attributes to be input, as is clear from FIG. 3(a).

In step S307, the display of attribute B, which does not require input due to attribute value a2 that has already been input, is canceled, and the state is returned to the state before the value of attribute A was input. Then, in step S308, this canceled attribute B is deleted from the attribute items R that do not require input.

On the other hand, in step S305, attribute A
If it is determined that another attribute value a2 or a3 is not set for , the input attribute A is added to the input attribute item Q in step S309.

In step S310, it is determined whether or not there is an attribute that does not require input depending on the input attribute value.
a) Search and examine the question data in (b). If there is an attribute that does not require input, the process advances to step S311; otherwise, the process advances to step S313. In this example, since attribute value a1 is input, it can be seen that attribute B does not need to be input based on the question data of FIG. 3(a). Therefore, processing proceeds to step 311.

In step S311, attribute B, which does not require input, is displayed based on the input attribute value a1. Furthermore, in step S312, this attribute B is added to the attribute item R that does not require input, and then, in step S314, the value of attribute A is set to the input value a1.

[0045] The above process is repeatedly executed, and step S3
14, the value obtained by subtracting the inputted attribute items Q and the attribute items R that do not require input from all the attribute items P is 0 (
nil), it is assumed that all the information necessary for inference has been input, and the question answering ends.

FIG. 8 is an example of a question screen when a1 is input as the attribute value. In this display example, the value a1 is changed to a predetermined color indicating that it is an input setting value, and further,
Due to this attribute value a1, the item of attribute B that does not require input is changed to a predetermined color indicating that input is not required. Therefore, by inputting information while looking at this display screen, the user can input only the information necessary for inference.

Next, as a second embodiment of the present invention, an example will be described in which there is an "attribute value" that does not require input when a certain attribute value is input.

[0048] As a prerequisite for this second embodiment, the following [rule set β],
It is assumed that three rule sets, [rule set γ] and [rule set δ], are stored together with other rule sets that are not specified. In the result part of the rules of [Rule Set β] and [Rule Set γ], an instruction as to which rule set the process should move to next is written.

[Rule set β] (Rule 5) Condition part: The value of the target attribute K is k1 Consequence part: Execute technique W5 and go to rule set γ (Rule 6
) Condition part: The value of the target attribute k is k2 Consequence part: Execute technique W6 and go to the rule set δ [rule set γ] (Rule 7) Condition part: The value of the target attribute M is m1 and the target attribute N
The value of is n1 Consequence part: Execute technique W7 (Rule 8) Condition part: The value of the target attribute M is m1 and the target attribute N
The value of is n2 Consequence part: Execute technique W8 [rule set δ] (Rule 9) Condition part: Value of target attribute M is m1 Consequence part: Execute technique W9 (Rule 10) Condition part: Target attribute The value of M is m2 Consequence: Execute technique W10

In the above rule set, when (rule 5) is selected in [rule set β], the following [
In the rule set γ], there is no rule for the case where the value of attribute M is m2. That is, when the value of attribute K is set to k1, it is meaningless to specify m2 for attribute M. Therefore, the attribute relationship extraction unit 3 extracts such relationships between attributes, and creates question data as shown in FIGS. 11(a) and 11(b) based on the extracted relationships between attributes.

The processing operation will be explained below with reference to the flowchart shown in FIG. In step S401, if all rule sets in the knowledge base 2 have been examined, the process advances to step S410; otherwise, the process advances to step S402.

In step S402, one rule set is extracted. In step S403, it is determined whether all the rules in the extracted rule set have been examined. If all the rules have been examined, the process returns to step S401; otherwise, the process proceeds to step S404.

In step S404, one rule is extracted from the extracted rule set. Next, in step S405, the condition part of the rule is extracted and recorded in the result data.

In step S406, it is checked whether there is a rule set to be activated next. If there is a rule set to be activated next, the process advances to step S407; otherwise, the process returns to step S403.

In step S407, the next set of rules to be activated is extracted. In step S408, if all the values for the attributes described in the condition part of the extracted rule set have been described, the process returns to step S406; otherwise, the process proceeds to step S409.

In step S409, undescribed attributes and attribute values are recorded, and the process returns to step S406.

In step S410, the resulting data is examined, and if there are attributes and attribute values that are not commonly described, they are recorded as final result data, and the process ends. As a result, attribute relationship data as shown in FIG. 10 is obtained.

[0058] The process of FIG. 9 described above is performed as [rule set β],
FIG. 10 shows an example of attribute relationship data extracted when applied to [rule set γ] and [rule set δ]. This FIG. 10 shows that when the value of attribute K is k1, it is not necessary to input the value m2 of attribute M. The attribute relationship extraction unit 3 extracts the attributes in FIGS. 11(a) and 11(b) from the attribute relationship data in FIG.
) Create the question data shown below.

Note that the question answering process for receiving information necessary for inference from the user is the same as in the first embodiment described above. FIG. 12 shows an example of a question screen based on the question data shown in FIGS. 11(a) and 11(b).

The expert system of the present invention is implemented as described above, but generally when inference is made to solve a problem, there may be cases where it is desired to perform inference again by changing set attribute values. Therefore, in order to cope with such a request, a correction function is added to the question answering control section 4 as an auxiliary function for canceling the settings of attributes or attribute values that are set not to be input according to the question data. This is desirable. Therefore, an example of a correction function for such a purpose will be shown below.

FIG. 13 is a flowchart of the operation of the question answering control unit 4 with the above-mentioned correction function added, and FIG. 14 shows an example of a display screen during this correction processing. The display screen of FIG. 14 is an example of a screen when the setting of attribute B that was set to not require input on the above-described question screen of FIG. 8 is canceled and the value of attribute B is changed to be inputtable again.

In step S501 of FIG. 13, the attribute input by the user as the one to be canceled is assumed to be B, and its value is assumed to be b2.

[0063] In step S502, the question answering control unit 4 refers to the question data (FIG. 3(b)) regarding the attribute B whose settings are to be canceled, and determines the attribute and its value that cause attribute B to be no longer input. Extract all pairs. Figure 3
In the case of (b), the value a1 of attribute A and the value a2 of attribute A are extracted. Then, in step S503, if all the sets are checked, the process advances to step S507; otherwise, the process advances to step S504.

In step S504, one of the extracted sets, for example, the value a1 of attribute A, is extracted, and in step S505, the attribute and attribute value currently set on the screen are changed to the extracted attribute A and its value a1. Check to see if they match. If the attributes set on the screen match the attribute values, the process advances to step S506; otherwise, the process returns to step S503. In the example of FIG. 14, since a1 of A is set on the screen, the process advances to step S506.

In step S506, the attribute A and its value a1 are recorded as data to be canceled, and then the process returns to step S503. Similarly, for the next attribute A and value a2 extracted from the question data, step S5
After performing the processes from 04 to S506, the process advances to step S507. Note that attribute A and its value a2 do not match the currently set attribute A and its value a1, and therefore are not recorded as data to be canceled.

In step S507, if there is data to be canceled, the process advances to step S508; otherwise, the process advances to step S511. In this example, step S50
Proceed to step 8. Then, in step S508, attribute B
The value a1 of attribute A, which is data set to require no input, is displayed, and in step S509, the user is asked whether or not to cancel the setting of this data. In FIG. 14, this is displayed as a warning message inside a rectangle at the top of the screen.

When a command to cancel the setting data is given, the value a1 of the attribute A that has already been set is changed in step S510.
is canceled, and instead, the value b2 of attribute B is set as new setting data in step S511. With the above steps, the correction processing operation is completed.

[0068] When the setting data is changed as described above,
It is necessary to perform re-inference based on this changed data. Therefore, it is desirable to devise ways to perform re-inference as efficiently as possible in such cases. In Figure 15,
An example of an expert system that satisfies these requirements is shown below.

The expert system shown in FIG. 15 is the same as the expert system shown in FIG. 1 by adding a function for recording an inference history to the inference engine 1, setting a history data area in the working memory 5, and further adding a function for recording an inference history to the inference engine 1. A backtrack processing unit 6 is added that returns the inference process to the changed position of the setting data during inference and starts re-inference.

The overall processing operation of the expert system shown in FIG. 15 will be briefly explained. The inference engine 1 records the reference status of user input data in the working memory 5 as history data when executing inference. When starting re-inference, the backtrack processing unit 6 activates the question answering control unit 4 and allows the user to change data settings. After this data change, the backtrack processing unit 6 compares the previous setting data and the new setting data, and extracts the difference as changed data. Then, based on this changed data, the inference is returned by referring to the history data, and when the point related to the changed data is returned, the inference engine 1 is started again from that position and the inference is restarted.

A flowchart of the processing operation of the inference engine 1 is shown in FIG.

In step S601, a table is first created to record the number of times the setting data input by the user is referenced. An example of this table is shown in FIG.

[0073] In step S602, it is determined whether the inference has ended. If it is in the final state, the inference is successful and ends. Otherwise, the process advances to step S603.

In step S603, it is checked whether all rules have been executed. If it has already been executed, the process will fail and end. Otherwise, the process advances to step S604.

In step S604, one rule is extracted, and in rule 605, the condition part of that rule is extracted.

In step S606, it is checked whether all conditions have been determined. If all conditions have been determined, the process advances to step S611; otherwise, the process advances to step S607.

In step S607, the attribute values in the working memory 5 are referenced for each condition. Then, in step S608, the referenced value is compared with the value in the condition section to check whether the condition holds true.

In step S609, the referenced attribute name, its value, and the determination result of the condition are recorded in the table shown in FIG. When the attribute is set by the user, the reference count table is updated and the reference count is also recorded.

In step S610, if the condition is met, the process returns to step S606; otherwise, the process returns to step S603.

After executing the consequent part of the rule in step S611, the process returns to step S602. Then, the inference process is repeatedly executed until the inference ends successfully or ends in failure.

In the inference process, the history data recorded by the inference engine 1 is as shown in FIG. Here, the attributes and their values referenced when determining the condition part,
The determination results of the conditions, the type of reference data indicating whether or not it is user setting data, and if it is user setting data, the number of times it was referenced are recorded.

FIG. 18 shows a flowchart of the processing operation of the backtrack processing section 6 in FIG. 15.

[0083] In step S701, the current user setting data is retrieved from the working memory 5 and saved.

[0084] In step S702, the question answering control unit 4 is activated and accepts data setting changes from the user. Then, in step S703, the saved previous setting data and this new setting data are compared, and the difference is recorded as changed data.

[0085] In step S704, it is checked whether or not there is a changed part of the data. If there is a changed part, the process advances to step S705; otherwise, the process ends.

In step S705, it is checked whether all the history has been returned. If you have returned everything, step S
If not, the process advances to step S706.

In step S706, one history regarding a certain rule is extracted from the history data shown in FIG. Then, in step S707, it is determined whether or not the attribute part whose setting has been changed is referenced during inference, and whether or not the reference number is the first time. If so, the process advances to step S708; otherwise, the process returns to step S705.

[0088] In step S708, the referenced data is deleted. Then, in step S709,
If the changed data becomes 0 (nil), the backtrack process ends and the process advances to step S710. Otherwise, the process returns to step S705.

In step S710, the inference engine 1 is restarted, and re-inference is started from the inference position where the changed data is 0 (nil). This eliminates the need to restart the inference from the beginning, making it possible to perform efficient re-inference.

Note that the functions of the attribute relationship extraction section 3, question answering control section 4, and backtracking section 6 described in the first and second embodiments may be configured by either software or hardware. It is something.

[0091]

[Effect of the invention] As is clear from the above,
When using the expert system of the present invention, the user only has to input the necessary information, and there is no need to input unnecessary information when inputting information all at once, which is an excellent feature that reduces the burden of information input work on the user. be effective.

[Brief explanation of the drawing]

FIG. 1 is a diagram showing a principle embodiment of an expert system of the present invention.

FIG. 2 is a diagram showing an example of attribute relationship data extracted by the attribute relationship extraction unit 3 in the first embodiment.

FIG. 3 is a diagram showing an example of question data created by the attribute relationship extraction unit 3 in the first embodiment.

FIG. 4 is a flowchart showing the entire processing of the expert system of the present invention.

FIG. 5 is a flowchart of the first half of the operation of the attribute relationship extraction unit 3 in the first embodiment.

FIG. 6 is a flowchart of the second half following FIG. 5;

FIG. 7 is a flowchart of the operation of the question answering control section 4 in the first embodiment.

FIG. 8 is a diagram showing a display example of a question screen in the first embodiment.

FIG. 9 is a flowchart of the operation of the attribute relationship extraction unit 3 in the second embodiment.

FIG. 10 is a diagram showing an example of attribute relationship data extracted by the attribute relationship extraction unit 3 in the second embodiment.

FIG. 11 is a diagram showing an example of question data created by the attribute relationship extraction unit 3 in the second embodiment.

FIG. 12 is a diagram showing a display example of a question screen in the second embodiment.

FIG. 13 is a flowchart of the operation of the question answering control unit 4 with a correction function added.

FIG. 14 is a diagram showing a display example of a question screen with a correction function.

FIG. 15 is a diagram illustrating an example of an expert system of the present invention to which a re-inference function is added.

16 is a flowchart of the operation of the inference engine 1 in FIG. 15. FIG.

17 is a diagram showing an example of history data recorded by the inference engine 1 in FIG. 15. FIG.

18 is a flowchart of the operation of the backtrack processing unit 6 in FIG. 15. FIG.

[Explanation of symbols]

1 Inference engine 2 Knowledge base 3 Attribute relationship extraction part 4 Question answering control section 5 Working memory 6 Backtrack processing section

Claims (1)

[Claims] [Claim 1] A batch question type in which information necessary for inference is input all at once before inference starts, and an inference engine performs problem-solving inference based on facts and rules in a knowledge base while referring to this input information. In the expert system, the relationship between each attribute is extracted by searching the rules stored in the knowledge base, and when a given attribute or attribute value is input, other attributes or attribute values that do not need to be input are described. an attribute relationship extraction unit that creates question data; and a question answering control unit that refers to the question data and extracts attributes or attribute values that do not require input when information regarding attribute values is input during bulk input of the information. An expert system characterized by: