JP3350047B2 - Question generation method and system - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial Application Field] The present invention has a question generation unit which can avoid generation of useless questions to a user and can verify the truth of a conclusion with a small number of questions. The present invention relates to a method and apparatus for generating a question for estimation, design support, a production system, and the like.

[Prior Art] A conventional question generation method of this type of system uses a so-called shim (CIM: Computer Integrated Manufacturing) in which the design, production, sales, delivery, etc. of a product order are consistently performed by computer management. ) Is performed in a system or the like. In this case, in general, the rule-based knowledge of "IF ... THEN ..." is stored in a knowledge base, and from this knowledge, a backward inference method is used to determine whether the conclusion is true or false. We verified using. As another method, for example, as described in JP-A-61-206037, the content of a question is stored as object data, and the range is narrowed down sequentially based on the parent-child relationship between these object data. And a method of generating the same.

[Problem to be Solved by the Invention] A specific example of the question generation method according to the related art will be described below.
Generate a question as shown. This figure shows a graph of a question for verifying the authenticity of the case of the motor, assuming that the value of the "enclosure structure" is "fully enclosed outer sector" (1802). For example, a set of rules indicating basic causal relationships such as "IF (type) is (safety-enhanced explosion-proof type) and (rating) is (continuous) THEN (jacket structure) is (fully enclosed outer fan-shaped)" is registered in advance. (1803) From this rule, it is true that the “enclosure structure” is “fully enclosed outer fan” and “
Since "5" needs to be "continuous", the truth of each condition is verified next. If the truth of these conditions has already been stored as data, verification is performed and the process is terminated. When generating a question, search for other rules that conclude this condition according to the order of the registered rules and the order of the conditional clauses in the rules, and if found, verify the truth of the conditions of the rule . This is repeated one after another to follow the graph of FIG. In this example, Q1: Is "Type 1804" an "Expanded Safety Type"? Q2: Is "Installation location 1807" a "dangerous location"? Q3: Is "Atmosphere 1808""In explosive gas"? Are generated in the order of

However, in this inference method, since questions are generated in the order of registration of rules, even if the question before determining the "enclosure structure" is already known to be not "continuous rating", Q4: Is "type""low noise type"? (1
806) should be asked immediately, but the above Q1-Q
Generate 3 useless questions. Therefore, there has been a problem that the number of questions before the conclusion is verified becomes extremely large. In the above case, a few conditions were described for simplicity, but when it is actually applied to the above CIM etc., it generally covers several to several hundred items, and the target product and its model are several hundred to The useless questions were usually enormous, as they often exceeded thousands.

In the question generation by the inference method described in the above-mentioned Japanese Patent Application Laid-Open No. 61-206037, the content of the question is stored in advance as object data, and the parent-child relationship between the object data is traced to each object data. In this case, the contents described in the necessary conditions, which are attributes possessed, are sequentially asked. However, in this method, the causal relationship must be described in the form of the object data and its parent-child relationship, which is difficult to describe depending on the structure of the object, and as the number of object data increases, As in the case described above, there is a problem that the structure is complicated.

In addition, there is a problem that a question to be used to verify the conclusion requires a special structure for generating the question. On the other hand, according to the present invention, the rule for generating a question is, for example, if
Can be easily generated from -then inference rules.

SUMMARY OF THE INVENTION An object of the present invention is to provide a question generation method capable of solving the above-mentioned problems of the related art and reducing the number of questions by avoiding generation of useless questions before verifying the truth of the conclusion. is there.

[Means for Solving the Problems] In order to achieve the above object, the present invention provides a computer system capable of verifying the truth of a conclusion that is satisfied under certain conditions by an internal processing thereof. A method and system for acquiring at least one question candidate using terms related to a conclusion, comprising: (a) when the acquired question is a plurality of question candidates, a determination rule for a plurality of questions that are mutually exclusive (B) excluding at least one question candidate that conflicts with one of the generated plurality of question candidates based on the semantic determination rules stored in advance in the storage unit. Is a plurality of question candidates, and when at least one question candidate is denied, a question stored in advance in the storage means including a rule for excluding at least one other question candidate from question candidates. Based on the formation rules,
Extracting question candidates other than at least one question candidate that conflicts with each other; (c) when the acquired question candidates are still plural,
A high-priority question is selected based on a question selection rule stored in advance in a storage unit including a rule for determining a priority of a plurality of questions. In other words, the question generation method according to the present invention has a question generation knowledge base, and based on this, generates a question that is considered to be most appropriate for verifying the truth of the conclusion for the current situation. Question generation means, question and the like display means, user and the like input means,
And processing selecting means for controlling these.

Further, the question generation knowledge base is based on a question generation rule which is a rule for enumerating a question candidate considered to be meaningful, a rule that one of the mutually contradictory question candidates is false if one is true, and the like. Semantic determination rules, question selection rules that are rules for selecting high-priority questions from remaining question candidates excluding the above-mentioned question candidates, question / answer data storing information on each question candidate, and answer It has a frame base and an inference control rule which is a rule for controlling the operation of the various rule groups. Further, the question generating means includes:
A question candidate graph generator for generating a graph of meaningful question candidates using the above-described question generation rule, and if one of a set of disjointed question candidates is true according to the semantic determination rule, the other is false and the question candidate Exclusion question judging unit to delete from the graph, thereby deleting question candidates, and selecting a high-priority question from the remaining question candidates by a question selection rule, a basic question input in advance It has a question generation rule creating unit that automatically creates a question generation rule from an inference rule representing a causal relationship.

This makes it possible to further use the question generation means and the question generation knowledge base to select a model that matches the content requested by the customer from among products having various variations and options. In order to narrow down the candidate models in the case, a model selection method having a function of issuing a question regarding an effective required specification item or a model specification item is possible.

[Operation] In the above-described question generation method, the following processing is performed by a processing device such as a workstation.

First, when the process selecting means is activated and the inference control rules in the question generation knowledge base are executed, the question candidate graph generation unit is activated. The question candidate graph generation unit determines a conclusion to be verified based on this, and in order to verify this conclusion, a question generation rule in the question generation knowledge base and a question / answer in the question generation knowledge base also Generate a meaningful question candidate graph to ask a question while collating with the answer data in the data frame base. When all executable question generation rules are collated, the process selection means is activated again, and the inference control rule is executed, whereby the disputed question determination unit is activated. The disputed question determination unit compares the answer data in the question / answer data frame base with the condition part of the semantic determination rule, and deletes the answer data and the disputed question candidate from the question candidate graph. When all executable semantic determination rules are collated, the process selecting means is activated again to execute the inference control rule, thereby activating the priority question selecting unit. The priority question selecting unit determines the question to be generated this time from the question candidate graph remaining without being deleted while checking the answer data in the question / answer data frame base with the condition part of the question selection rule. . When the question is determined in this way, the content of the question is displayed on the display means. On the other hand, when the user inputs an answer using the input means, the answer data is stored in the question / answer data frame base. Here, the question candidate graph generation unit is activated again by the processing selection unit. The question candidate graph generation unit reduces the question candidate graph based on the logic constituting the condition part and the rejection while comparing the added answer data with the condition part of the question generation rule. When all of the executable question generation rules are collated, the above-described processing selecting means is activated, and then the exclusive question determination section is activated. Here, the answer data and the inconsistent question candidates are deleted from the reduced question candidate graph. Next, the priority question selection unit is activated again by the process selection means, and a high priority question is generated from the narrowed question candidates.
By repeating the series of steps described above, narrowing down the meaningful question candidates according to the user's response progress, and determining the questions in priority order from among them, avoid unnecessary questions and reduce the number of questions to determine the truth of the conclusion. Can be verified.
The question generation rule is automatically generated by the question generation rule creation unit based on basic causal knowledge input in advance. Therefore, the user does not need to input all the rules required for the question, and even if the number of rules input in advance increases, the rules to be generated are the same, and the structure is not complicated.

Embodiment One embodiment of the present invention will be described below with reference to FIGS.

FIG. 1 is a schematic functional configuration diagram showing one embodiment of a question generation method for realizing a question generation method according to the present invention. In FIG. 1, reference numeral 101 denotes a question generation knowledge base for storing data of various rules and frames used for generating a question, and includes 102 to 106 shown below. 102 is an inference control rule for controlling the operation of the question generation process, 103 is a question generation rule for generating a meaningful question candidate graph, and 104 is a false if a plurality of disjoint question candidates are true. 105 is a semantic determination rule composed of rules, etc., 105 is a question selection rule for determining a high-priority question candidate from a question candidate graph, 106
Is a question / answer data frame base that stores generated data on each question candidate and the user's answer.

107 is a question generation means for determining a question to be generated using the question generation knowledge base 101, and 108 to 111 shown below.
It consists of. 108 is a question candidate graph generation unit that generates a question candidate graph using the question generation rule 103 and the question / answer data frame base 106, and 109 is an exhaustion candidate using the semantic determination rule 104 and the question / answer data frame base 106. Exclusion question determination unit that deletes the unquestionable question candidate, 110 selects the question selection rule from the generated question candidate graph.
A priority question selection unit that determines a question using 105 is a question generation rule generation unit that generates a frame of the question generation rule 103 and the question candidate from a rule describing a basic causal relationship input in advance. is there. 112 is a display means for displaying the above-mentioned various questions, 113 is an input means for inputting a user's answer or a basic inference rule, and 114 is a process selection means for selecting a process when generating a question using the inference control rule 102 It is.

FIG. 2 shows an operation flow of the question generating means shown in FIG. FIG. 3 is a graph showing a process of generating a question by the present question generation method. FIG. 4 shows a list of frame configurations in the question / answer data frame base 106. Fifth
The figure shows the attributes and values of various frames in FIG. FIG. 6 is a graph explanatory diagram showing a method of generating a question generation graph. Hereinafter, the process of generating a question using FIGS. 2 to 6 will be described in detail with reference to the flowchart of FIG. In order to make the comparison with the conventional example clear,
The same shall apply to the selection of the motor model as in Fig. 18.

When the question generating means shown in FIG. 1 is activated, the processing selecting means 114 refers to the inference control rule 102 in the question generating knowledge base, and first controls the question candidate graph generating section 108 in the question generating means 107. Move. The selection status of each process in this case is set in the procedure control frame 401 in FIG. The question candidate graph generator 108 refers to the selection state frame 402 of FIG. 4 in the question / answer data frame base 106,
The “type” of 502 in FIG. 5, which is the attribute in the frame,
Reference is made as to whether each value of “enclosure structure”, “mounting frame shape”, “output”, “number of poles”, and “rating” is “undecided” or “decided”.

Among them, the attribute whose value 503 is “undecided” is the item of the conclusion to be verified by the question. (Step 201) In this case, the "type", "envelope structure", and "output" are verified in order. As shown in FIG. 4 404, six values of “envelope structure” are registered, such as “fully enclosed outer sector”, and the authenticity of these values is verified. In the present embodiment, a case will be described in which it is verified that the value of the item “envelope structure” is “fully enclosed outer sector”.

When the conclusion to be verified is determined, the question candidate graph generator
108 starts matching the condition part of the question generation rule 103 with the attribute and value of each question candidate frame in the question / answer frame base 106. The attribute content of each frame is configured as shown in FIG. 5, for example, in the case of the “installation location is an urban area” (403) frame. That is, “question name” 508 representing the question item name, “answer name” 509 representing the value of the question item, “value” 510 for asking the user and setting the answer, and whether the 403 frame was selected as a question candidate "Question candidate" that represents 51
1. In the question candidate graph, a "success rule" 512 indicating the presence or absence of a rule following this question candidate is provided, and a "question priority" 513 for setting a priority between question candidates.

The question generation rule 103 generates a question candidate graph with the contents shown in FIG. 6 at the time of collation with the frame in the question / answer frame base 106 (step 202). In FIG. 6, reference numeral 601 in 600 of case a indicates verification [A] which is a node indicating a conclusion to be verified, 602 indicates a question [B] which is a node indicating a question item required for verification, and 603 is the same as 602. Is a question [C] which is a node indicating the question item. Now [A]
[B] and Question [C] are both likely to be <Affirmative> when it is desired to verify, that is, a question candidate frame
When the attribute “value” 510 of 403 is not “negative”, [B] and [C] are question candidates (case a: 60).
0). Also, as in case b (607), when question [B] is already "No" or question [B] is <unknown> and there is no subsequent question generation rule, that is, the attribute of question candidate frame 403 When the “subsequent rule” is “absent”, questions [C] and question candidates subsequent to [C] are not required to be asked.
According to the above-described procedure, all question candidates meaningful to ask a question from the answer situation of the user are listed as a question candidate graph.

In FIG. 3, reference numeral 301 denotes a question candidate graph generated by the question candidate graph generation means 108.

Next, more unnecessary question candidates are deleted from the question candidate graph. That is, when there are no more question generation rules that can be executed, control is transferred to the processing selection means 114 to determine whether there is one or more question candidates. If not, the process is terminated because there are no more questions that can be generated. If there is, processing selection means
114 refers to the inference control rule 102 again, and
Transfer control to 109. Exclusivity question determination unit 109 performs semantic determination rule
4 is compared with the question candidate frame in the question / answer data frame base 106, and if one of the set of disagreeable questions is true, the other is false (step 203).

For example, as shown in FIG.
Is "in the air"? If the answer 702 to (701) is "yes", is the "ambient", which is a question candidate that dismisses it "is in explosive gas"? (703), the value (704) of the attribute “question candidate” is set to “impossible” and deleted from the question candidate graph. As described above, questions that are useless in terms of the meaning of the questions can be eliminated. The 705 demon is an automatic update operation function in which when these values are entered, the slots in other frames related thereto are updated accordingly.

When all of the semantic determination rules 104 are executed and there are no more semantic determination rules that can be executed any more, control is transferred again to the processing selecting means 114, and the processing selecting means 114
Start 0. The priority question selection section 110 is based on the above question selection rule.
105 is compared with the question candidate frame in the question / answer data frame base 106.

Here, for example, when it is <affirmative> that the “installation place” is “city area”, as the question selection rule 105, “IF“ installation place is urban area ”is <affirmative> THEN“ is type low noise type? ” If the rule "question priority <presence>" is registered, the value of the attribute "question priority" of the frame "type is low noise type" 405 (FIG. 4) in which this rule is executed is <presence> Is set to The resulting question is determined (step 204). In this example, the question priority values are <Yes> and <No>,
It can also be represented by a value between 0 and 1.0.

When the question is determined in this way, control is transferred to the process selecting means 114, and the process selecting means 114 asks the display means 112 whether the question "type" is "low noise type". Is displayed (step 205). The user inputs an answer to this question through the input means 113 (step 206).

The process selecting means 114 sets the answer data to the attribute "value" of the corresponding frame in the question / answer frame base 106, in this case, the frame "type is low noise type" 405 (step 207).

With this answer, the verification is completed. If the verification is successful, the process ends. If the verification is not yet completed, the process returns to step 208 again.

Next, the process proceeds to step 202 again based on the data newly added by the answer 207, and further generates a question candidate graph. Here, as the answer data increases, the number of question candidates generated this time decreases from all times, and the question candidate graph is reduced. Further, in the next steps 203 and 204, a question having a higher priority is selected from the question candidates and a question is generated as in the first time.

In order to make the essential points of the embodiment described above more apparent, a description will be given with reference to FIG. In the figure, 301 is a question candidate graph, 302 to 310 indicate each question candidate, 311 is a meaning determination rule (corresponding to the above 104), 312 is a question generation rule (corresponding to 103), and 313 is a question selection rule (105 Equivalent).

First, in the process of generating questions, is the 302 “installation location” an “urban area”? Is <Affirmation> 303 “Atmosphere” Is “Atmosphere”? It is assumed that <Yes> has already been answered. Next, when the question generation rule 312 enumerates question candidates for verifying that the 304 “envelope structure” is “fully enclosed outer fan shape” (304), 303 “atmosphere” is “atmosphere”. Because of the semantic determination rule 311, 305 “atmosphere” is “negative” for “in explosive gas”, and the question generation rule 312 excludes all questions in the range indicated by 306 from question candidates and should be generated Question candidates are narrowed down. Furthermore, according to question selection rule 313 from the remaining question candidates, is 310 “type” “low noise type”? Is selected. As described above, it is possible to eliminate useless questions and perform a question order according to the answer situation.

FIG. 8 shows an embodiment of the above-mentioned priority selection rule 313 and the question generation by the embodiment. The inference rule 801 indicating a basic causal relationship that has been input in advance is made up of rules as shown in the figure. The question generation rule 103 and the question candidate frame 802 are automatically generated by the question generation rule creating unit 111 in FIG. Is shown. As described above, according to the present embodiment, it is not necessary for the user to create a question generation rule, and an efficient question can be automatically generated by inputting fragmentary causal relationships one after another.

Next, an example of a model selection device using this question generation method will be described.

In this embodiment, the present question generation method is applied to a model selection device for selecting a model that meets a customer's request from among products having various variations and options. That is, by providing the question generation means of the present invention, even if there are omissions in the customer's required specifications inputted, it is possible to ask a question regarding required required specifications and model specifications and select a model in a short time. It is.

FIG. 9 is a schematic functional configuration diagram showing an estimation support method having a model selection device according to the present embodiment. In Figure 9,
901 is a means for inputting a model specification in which a customer's required specification or a target product is, for example, a classification item of a motor, 902 is a model specification inference means for inferring a value of a model specification whose value is undetermined from the input specification, 903 is A model specification inference knowledge base that stores inference rules required when performing model specification inference. The 904 is used to determine whether an input specification value is inconsistent with another specification value already input. A specification value inconsistency determination knowledge base that stores necessary knowledge, and 905 is model specification inference means 9
Model specification inference result storing means for storing the inference result of 02, 906 is a search condition determining means for determining a search condition when searching the model database from the inference result, and 907 is a candidate model for executing a model database search. Search means, 90
Reference numeral 8 denotes a model database for storing the relationship between models and model specifications described in a product catalog or the like, and 909 denotes a candidate model output means for displaying the model selected as a result of the search and the model specifications.

The following steps 910 and later are necessary when selecting processing based on the candidate model selection result, so that appropriate model selection can be performed even when the number of candidate models is 0 or when the number of candidate models output is large. It is for becoming. 910 is a process selection means for selecting a process when the number of candidate models is 0, 1 and 2 or more, and 911 is for explaining the reason for the contradiction when the number of candidate models is 0 and prompting for correction. Means for explaining the reason for contradiction, 912 is a delivery date / price calculation that calculates a delivery date / price based on the model specification of the selected model when the number of candidate models is one or when one is selected from two or more. Means, 913 is a question generation means for generating a question regarding required specifications required to select one of the candidate models by model specification inference and model database search when there are two or more candidate models, and 914 is a question generation means 915 is a selection result registration means for registering information on the selected model specification, delivery date, price, and customer name in a customer database, 916 is an estimate output means, 917
Is a customer database.

FIG. 10 is a diagram of an example of a hardware configuration showing a use form of the apparatus shown in FIG. 9 when applied to a quotation business at a special agent. A workstation is installed at the special agent, and a host computer is arranged at the factory. And a communication line connecting them. In the following description, a case where the target product is a motor will be described as an example. In FIG. 10, reference numerals 1001 to 1007 denote devices installed in special dealers, 1001 denotes a communication control device,
1002 is a workstation system device, 1003 is a display device, 10047 is a keyboard, 1005 is a mouse, 100
Reference numeral 6 denotes a printer for printing a quote, and reference numeral 1007 denotes an external storage device for storing information on customers and the like. The dealer and the factory are connected by a communication line as shown in the figure.
Devices up to 14 are installed in factories. 1009 is a communication control device, 1010 is a host computer, 1011 is a display terminal, 1012 is a keyboard, 1013 is a printer that prints order information and the like sent from a special agent via a communication line,
Reference numeral 1014 denotes an external storage device for storing price and delivery date information. In this configuration, when an order is confirmed based on the result of model selection and delivery date / price estimation at a dealer, the order information is immediately sent to the factory online, and the order information sent to the factory is Printed on a printer or arranged for design.

FIG. 11 is a flowchart showing a processing procedure when selecting and estimating a model of a target product motor of the apparatus shown in FIG.

In FIG. 11, first, a required specification, a model specification, and a customer name are input in a required specification / model specification input unit 901 in FIG. 9 (step 1101). Here, conventionally, when selecting a model of a motor of a target product, a product configuration table indicating a relationship between a model and a model specification described in a product catalog has been referred to. However, in the product catalog, the classification of model items is based on the model specification, which is a technical term of the motor, so it is strictly divided and maintenance is easy, but it is difficult for customers to understand, and which model is suitable for customers Cannot be immediately determined. In the present invention, terms that are easy for the customer to understand and at the same time can infer the model specifications are defined as the required specifications, and two types of terms, the required specifications and the model specifications, can be input, and the inference is made based on these. In this way, an appropriate motor is selected.

As the required specifications, the ambient temperature, etc., for the intended use, which is easy for the customer to understand, are used. For the model specifications, the type, output, number of poles, mounting method that indicates how to mount the motor, and the structure of the housing are described. The structure to be used is used.

When the input is completed in step 1101, the model specification inference means 90
2, the model specification inference rules of the model specification inference knowledge base 903 are activated, and the inference of uninput model specifications is started by forward inference from the input required specifications and model specifications (step 1102). Here, when there is a model specification inference rule as shown in FIG. 12, the rule a) satisfies the condition because “installation place” is “constant temperature room”, and concludes that “motor temperature rise caution” is 1.0. With a certainty factor. The certainty factor is a numerical value indicating the likelihood that the conclusion is true, and here, it is assumed that the certainty is given in advance by the rule conditions and the conclusion. Next, from the conclusion of a), the rule of b) is also activated in order to meet the condition. As a result, it is obtained that the “type” is “heat-resistant”, and the certainty factor is 0.6. In addition, the rule of c) is established because the “use” is “for grinding machine”, the confidence is 0.7 with the “type” for “corrosion prevention”, and the “covering structure” is “ With the totally enclosed fan shape,
0.8.

The result of the model specification inference by the model specification inference means 902 is stored in a table format as shown in FIG. Next, the search condition determining means 906 determines the conditions for searching the model database 908 from the following as follows. First
Paying attention to the column of certainty in FIG. 3, the item of “type” has two candidates of “for anticorrosion” and “for heat resistance”. The same applies to other items. For the model specification items that have already been input, the values are used. As described above, the search conditions determined by the search condition determining means 906 are obtained as shown in FIG. (Step 1103). Next, candidate model search means 907
Is used to search the model database 908 (step 1104).
FIG. 15 shows a table configuration of a database 908 of this kind. Here, when a search is performed under the search conditions shown in FIG. 14, one combination of model specification items as shown by 1501 in FIG. 15 is obtained. In this case, the candidate model output means 909 displays the corresponding candidate model list screen as shown in FIG.

In this example, only one candidate model is output, so this model specification is selected (steps 1105 and 1106). Next, the delivery date / price for this model specification is calculated by the delivery date / price calculation means 912 (step 1107). For details, refer to the related earlier application, Japanese Patent Application No. 2-96 by the same applicant.
As described in No. 559.

The above is the basic flow of model selection and estimation in FIG. However, whether or not the model selection is completed at once depends on input data of required specifications and model specifications. Therefore, if there is a contradiction in the combination of the input required specifications and the model specification values, the number of candidate models becomes zero, or if the input required specifications and the model specification items are insufficient, the number of candidate models becomes Two or more. If the number of candidate models is large, selection is difficult unless the input data is corrected and the process is started again, or the required specifications and model specifications are not additionally input. Therefore, in this embodiment, as a method for dealing with these problems, the inconsistency explanation means 911 and the question generation means 913 are used.
Is provided. Next, the processing function of this will be described.

First, when an apparent inconsistency is found in the data input first as the required specification, the input data is checked whether there is any inconsistency using the model specification inference means 902 and the inconsistency determination knowledge base 904 between the specification values. (Step 111
1) If there is any inconsistency in the input data, an input correction message is displayed (step 1110), thereby prompting re-input (step 1110). Next, the candidate model search means 907 searches the model database 908 based on the model specification inference result of the model specification inference means 902, but there is a case where the number of candidate models is 0. For example, "type" is "waterproof" and " Enclosure structure "
When the model database 908 is searched with the search condition of the search condition determining means 906 that is “completely closed” and “rated” is “continuous”, “completely closed” and “continuous” are obtained from the combination shown in the table configuration 1502 in FIG. There are no candidate models because there is no combination of. Then, the reason for the contradiction is searched by the contradiction reason explanation means 911 (step 1112), the searched reason for the contradiction is displayed (step 1113), and the user is prompted to correct it. It can be seen that it is sufficient to change the required specifications, so that the candidate model can be selected by correcting it and executing the inference and search again.

Next, if there are two or more candidate models from the search results due to insufficient input of required specifications and model specifications,
A question is generated by the question generating means 913 and the question generating knowledge base 914 (step 1114), and based on the generated question, an answer to the content of the question is input (1101). 1102-1104). For example, in the case where the "enclosure structure" is "fully closed" and the "rating" is "30 minutes", when the model database is searched after the model specification inference, the candidate model is 4 based on the combination of 1502 in FIG. Output. At this time, the candidate model output means 909 selects either one of the four candidate models, or further narrows the candidate model by a question because the candidate models are not sufficiently narrowed down. Can be.

Here, if the user selects the question generation,
Since only four poles are different from the combination of 1502 in FIG. 15, four candidate models are output. Therefore, a question for specifying the pole number is generated by the question generating means 913 as follows. Here, when there is a question generation inference rule as shown in FIG. 17 as knowledge of the question generation method in the question generation knowledge base 914, it is first assumed that the number of poles is 2 and the backward inference is performed. Verify by That is, according to the rule in FIG. 17A, it can be seen that the frequency should be 50 Hz and the rotation speed should be 3000 rpm for the number of poles to be two. So, if both of these are still undecided, first ask, "Is the frequency 50Hz?"
Since this rule does not hold, another rule whose result is "the number of poles is 2" is searched and the same is repeated. If you answered yes in the previous question, you will be asked another condition, "Is the rotation speed 3000 rpm?" If "yes" is answered, "the number of poles is determined to be 2". If you do not know again, this time the rule with the conclusion that the rotation speed is 3000 rpm is searched. For example, in the case of the rule b) in FIG. 17, questions such as "Is the output 5.5 kw?" And "Is the full load torque 1.8 kgf?" Are asked in order.
Thus, the candidate models are narrowed down. If “the number of poles is 2” is not established, candidate models are narrowed down in the same manner assuming “the number of poles is 4”, “the number of poles is 6”,.

Note that the present invention is not limited to the above-described embodiment. For example, instead of a method for explaining the reason for inconsistency, it is also possible to automatically correct input data of required specifications and model specifications according to inconsistency resolution rules. Prior application, Japanese patent application Hei 2-96
As described in 559.

In addition, instead of displaying a question and obtaining an answer, instead of displaying the question and obtaining an answer, instead of using the method of FIG. As in the above-mentioned prior application, the method of obtaining the value of the item (1) is also possible.

[Effects of the Invention] According to the present invention, in an interactive system, unnecessary questions can be avoided even if generated based on data answered by a user, and important questions can be preferentially asked. You can verify the conclusion in a short time. Also, the interactivity of the model selection device using this question generation method can be improved,
Efficient estimation support can be implemented.

When the present invention is applied to a so-called CIM or the like, the above-mentioned useless questions are automatically omitted from the products, which usually have hundreds, thousands, or more, models, number, and hundreds of questions. And the practical effect is remarkable.

[Brief description of the drawings]

FIG. 1 is a functional block diagram showing an embodiment of a question generation device having a question generation method according to the present invention, FIG. 2 is an operation flow diagram of the question generation device of FIG. 1, and FIG. FIG. 4 is a list of frames based on the question / answer frame of the present invention, FIG. 5 is an explanatory diagram of the contents of various frames of the present invention, FIG. 6 is an explanatory diagram of a method of generating a question generating graph of the present invention, FIG. 7 is an explanatory diagram of the operation of the meaning determination rule of the present invention,
FIG. 8 is a diagram showing a data created by the question generating rule creating section of the present invention;
FIG. 9 is a functional configuration diagram of a model selection device having a question generating means of the present invention, FIG. 10 is a hardware configuration diagram showing a use form of the device of FIG. 9, and FIG. 11 is a flowchart of a processing procedure of FIG. , FIG. 12 is an explanatory diagram of a rule example in the model specification inference means of FIG. 9, FIG. 13 is a configuration diagram of a table for storing model specification inference results of FIG. 9, and FIG. FIG. 15 is a table configuration diagram of the model database of FIG. 9, FIG. 16 is an explanatory diagram of the applicable model candidate list screen of FIG. 9, and FIG. 17 is a query generation of FIG. FIG. 18 is an explanatory diagram of the inference rules in the means, and FIG. 18 is an explanatory diagram of a question generation method using backward inference. 101 …… Question generation knowledge base, 102 …… Inference control rules, 10
3 ... Question generation rule, 104 ... Semantic judgment rule, 105 ... Question selection rule, 106 ... Question / answer data frame base, 107 ... Question generation means, 108 ... Question candidate graph generation unit, 109 ... Exclusion question determination unit, 110 ... Priority question selection unit,
111 ... Question generation rule creating unit, 112 ... Display means, 113 ...
... input means, 114 ... processing selection means.

Continuation of the front page (72) Inventor Nobuyuki Doi 7-1-1, Higashi-Narashino, Narashino-shi, Chiba Pref. Inside the Narashino Plant of Hitachi, Ltd. (56) References “Kashima Institute of Construction Technology Annual Report” No. 37 (1989) 272-282, Proc. Of the 41st (1990) National Convention, Information Processing Society of Japan, 1990-8-15, p. 2-29-30 (58) Fields surveyed (Int. Cl. ⁷ , DB name) G06F 9/44

Claims (9)

(57) [Claims] 1. A computer system for verifying the truth of a conclusion satisfied under a certain condition, wherein a computer system receives at least one condition and an input of a conclusion satisfied under the condition. Automatically generating at least one question candidate graph for verifying the truth of the conclusion based on the input condition and conclusion; accepting an answer to at least one question candidate in the automatically generated question candidate graph (A) excluding at least one question candidate mutually exclusive from the question candidate graph from the answer and the semantic determination rule that is a determination rule of a plurality of mutually exclusive questions; and (b) When the answer and the question candidate connected as a node are set in a context, if at least one question candidate is denied, another Excluding from the question candidate graph at least one other question candidate linked to the question candidate for which the answer has been obtained, from a question generation rule that is a rule for excluding at least one question candidate. (C) After any of the above steps (a) and (b), if there are still a plurality of question candidates in the question candidate graph, the rule is to determine the priority of the plurality of questions. Selecting a high priority question based on a question selection rule. 2. When the generated question is still a plurality of question candidates, accepting an answer to at least the highest priority question candidate among the plurality of question candidates; Performing one of the steps (a) and (b) and the step (c) using the answer to the at least highest priority question candidate in place of the answer to the candidate. The method of claim 1. 3. The method according to claim 1, further comprising the step of automatically generating said question generation rules based on inference rules stored in advance. 4. A database of at least one of a number of products.
Receiving, in a method of selecting a product type, a computer system for searching for one product, input of a target product and a required specification relating to use of the target product; A step of inferring a required unentered technical specification candidate based on a specification inference rule which is a rule for inferring the specification; if the inferred specification candidate has a plurality of specification candidates, the inferred specification candidate is Searching for candidate models from a model database stored in advance; if there are a plurality of searched candidate models, accepting input of model specifications in the model database; authenticity of the model specifications Automatically generating at least one question candidate graph that verifies: for at least one question candidate in the automatically generated question candidate graph Receiving an input of an answer; and (a) excluding at least one question candidate mutually exclusive from the question candidate graph from the answer and a meaning determination rule that is a determination rule for a plurality of mutually exclusive questions. And (b) a question generation rule that defines the context of the answer and the question candidates connected as nodes and excludes at least one other question candidate connected to the question when at least one question candidate is denied. Removing from the question candidate graph at least one other question candidate linked to the question candidate for which the answer has been obtained from the rule; and (c) any of the above (a) and (b). After this stage, if the number of question candidates in the question candidate graph is still more than one, the higher the priority based on the question selection rule that is the rule for determining the priority of the plurality of questions. Model Selection characterized by having a; selecting a question. 5. When the generated question is still a plurality of question candidates, accepting an answer to at least the highest priority question candidate among the plurality of question candidates; Performing one of the steps (a) and (b) and the step (c) using the answer to the at least highest priority question candidate in place of the answer to the candidate. The model selection method according to claim 4. 6. A step of generating a new question based on the question generation knowledge stored in advance when the required specifications regarding use of the product are insufficient as a result of searching for candidate models in the searching step; 5. The method according to claim 4, further comprising: obtaining an answer to a specific question; and further determining a search condition based on the answer. 7. A database of at least one of a number of products.
A computer system having a function of searching for two products, a means for receiving an input of a target product and a required specification relating to use of the target product; inferring a specification relating to a technical attribute of the product based on the required specification Means for inferring a required uninputted technical specification candidate based on a specification inference rule, which is a rule to be executed; if the inferred specification candidate has a plurality of specification candidates, based on the inferred specification candidate, Means for searching for a candidate model from a model database stored in advance; means for receiving input of a model specification in the model database when there are a plurality of searched candidate models; verifying the authenticity of the model specification Means for automatically generating at least one question candidate graph; receiving an answer to at least one question candidate in the automatically generated question candidate graph; (A) means for excluding at least one mutually exclusive question candidate from the question candidate graph from the answer and the semantic determination rule that is a determination rule for a plurality of mutually exclusive questions; b) From the above-mentioned answer and a question generation rule which is a rule for defining the context of a question candidate connected as a node and excluding at least one other question candidate connected to the question when at least one question candidate is denied, Means for excluding from the question candidate graph at least one other question candidate that is linked to the question candidate for which the answer was obtained; and (c) any one of the above steps (a) and (b). Later, when there are still a plurality of question candidates in the question candidate graph, a higher priority question is selected based on a question selection rule which is a rule for determining the priority of the plurality of questions. Model selection system; and a; unit. 8. A means for accepting an answer to at least the highest priority question candidate among the plurality of question candidates when the generated question is still a plurality of question candidates; Control means for executing any of the means (a) and (b) and the means (c) using the answer to the at least highest priority question candidate in place of the answer to the candidate. The model selection system according to claim 7, wherein: 9. A computer system for retrieving the product is located in a sales office, a database relating to at least the price or delivery date of the product is located in a factory, and data relating to the price or delivery date is transmitted from the sales office via a communication line. 8. The system selection system according to claim 7, further comprising means for searching a database in said factory via a computer.