JPH06282435A - Knowledge base storage method - Google Patents

Abstract

PURPOSE:To provide the knowledge base storage method for structuring a knowledge base by putting condition rules together in a matrix as to an expert system which puts a question from a computer and infers a conclusion from an operator's answer to it. CONSTITUTION:An answer is made on a question classification screen for a question at the trunk part of tree structure and a next question classification screen or a question screen are obtained in response to the answer. When a next screen is the question classification screen, the operation is repeated until the question screen is obtained. Then when the question screen is obtained, an answer to the question screen is made (ST2). Condition rules are stored in matrix structure in advance (ST1) and the conclusion is inferred (ST4) from the condition rules of this matrix structure and the result of the answer to at least one question screen.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an expert system for diagnosing diseases and providing accurate guidance, and more particularly to a knowledge base storage method for inquiring from a computer and inferring a conclusion from an operator's response to the question.

[0002]

2. Description of the Related Art With the development of computer systems, it is possible to draw conclusions on behalf of an expert such as interviews for illnesses, career guidance in the education of children, and various kinds of guidance, even if there is no expert in the field. Expert systems have been developed and put into practical use in various fields.

Conventionally, in this expert system, a question is prepared in advance, an operator or the like answers the sequentially prepared questions, and a conclusion is obtained from the answered result and output to the operator. For example, in the course guidance in the education of children, the following procedure was used to obtain the course conclusion.

First, when an expert system for career guidance is set up, a question is asked as to whether it is a university entrance exam or a high school entrance exam. For example, if you answer that you are going to a university exam, you will be asked what profession you want to work in the future. In this inquiry, many items are displayed, for example, doctors, school teachers, researchers, ...
Select individuals, eg researchers. There are various types of researchers, so in the next item, questions that further limit the content will be asked. In this way, the operator will answer the questions because the future goals are successively asked.

Then, the person is asked about his current performance. The operator inputs all the results such as school and test of deviation value. Then, the operator will answer further questions about the current study method and the like.

In this way, the operator responds to the inquiry from the expert system in sequence, and the expert system obtains a conclusion from the response and outputs it to the operator.

The expert system as described above can configure the system when the answer to the question and the conclusion are clearly known in advance, and the program is created after knowing the flow.

[0008]

In the above-mentioned conventional expert system, all the questions are prepared in advance, and the conclusion is determined by the response to them, so the problem that the conclusion cannot be obtained unless all the questions are answered. Had.
Moreover, when trying to add a new conclusion, there was a problem that the program of the expert system had to be changed.

It is an object of the present invention to provide a knowledge base storage method for constructing a knowledge base by matrixing requirement rules in an expert system.

[0010]

FIG. 0 is a block diagram showing the principle of the present invention. The present invention relates to an expert system for obtaining a suitable conclusion from a response to each question in a tree structure. A question classification screen showing the classification of the following questions is used to display a question on the trunk side of the tree and a question screen for which no questions exist thereafter. Constructs a query for the last branch of the tree. For example, this question screen is managed by being classified into a Y / N question record that indicates a two-choice selection screen represented by YES and NO, and an n-choice question record that selects one of n items. To be done.

The knowledge base stores the responses to one or more question screens as a requirement rule in a matrix structure corresponding to the conclusion to be obtained (ST1). Reply from the question classification screen which is the question of the trunk part of the tree structure,
In response to the reply, the following question classification screen and question screen are obtained. When the next screen is the question classification screen, it is repeated until it becomes the question screen. When the question screen is displayed, the question screen is answered (ST2). Further, even if it is a tree structure, it jumps to another branch, returns to the trunk direction, responds to another question classification screen, and replies to another question screen again (ST3).

On the other hand, a conclusion is obtained by inferring from the result of replying to the at least one question screen and the requirement rule of the matrix structure stored in the previously obtained knowledge base (ST4).

This conclusion also comprises a conclusion classification screen and a conclusion screen divided by the split screen, and a conclusion detail screen showing the details thereof, and the operator is given an item of the conclusion division screen for dividing the conclusion after the conclusion is obtained. The details not instructed at the time of instructing the question are displayed.

Since the question is divided into a question classification screen and a question screen, and the question screen is divided into a Y / N question record and an n-choice question record for management, the requirements necessary for obtaining a conclusion are obtained. Rules can be matrixed, and expert systems can be freely created and modified.

[0015]

The present invention will be described in detail below with reference to the drawings. FIG. 2 is a system configuration diagram of an embodiment of the present invention. The embodiment of the present invention shown in FIG. 2 comprises a work operation terminal 1 and a host computer 2.

The business operation terminal 1 has a processor (not shown), and performs processing as an operation terminal under the control of the processor. The business operation terminal 1 includes an MMI control unit 1-
1, a window control table creation unit 1-2, a communication control unit 1-3, a window control table group file 1-4, and a dialogue input data file 1-5. In the embodiment of the present invention shown in FIG. 2, the question screen has a multi-stage tree structure as shown in the window configuration diagram of FIG. The windows or screens that make up the tree can be divided into a question classification screen and a question screen. The question classification screen is the white frame in FIG. 3, and the question screen is the shaded area. That is, in the embodiment of the present invention, the question classification screen constitutes the upper part of the tree structure, and the question screen constitutes the final branch portion.

In order to set a screen having such a tree structure relationship, the window control table creating section 1-2
Then, when the screen is generated, the question classification record, the YES / NO question record (hereinafter, Y
/ N question record) and n-choice question records. Then, the window control table is created from the record. The window creation table created by the window control table creation unit 1-2 is the window control table group file 1
-4 is stored in the file.

In the embodiment of the present invention, the information on the window screen of the tree structure is temporarily stored in the window control table group file 1-4, and after being confirmed, transferred to the host computer 2. Further, the matrix table is also stored in the host computer 2 and is read and used by the file server when necessary. The present invention is not limited to this, and a window control table or a matrix table may be stored in the business operation terminal 1, and the matrix table and the information to be input may be transferred to the inference server when necessary for inference.

Further, the conclusion has a one-layer tree structure and is divided into a conclusion classification record, a conclusion record, and a conclusion detail record as in the question screen. The window control table creation unit 1-2 inputs these records and also converts them into a table.

The MMI controller 1-1 is connected to the window control table group file 1-4. MMI control unit 1
Reference numeral -1 is a human interface type input / output unit, which reads one of the window control tables stored in the window control table group file 1-4 and displays it on a display device (not shown). Also, the input of a designated key or the like corresponding to the screen displayed on the display device is accepted.

The MMI control section 1-1 is further connected to the dialogue input data file 1-5, displays the screen based on the window information stored in the window control table group file 1-4, and correspondingly displays it. The input information is stored in the dialogue input data file 1-5. The screen displayed to the operator when creating various records and the like in the window control table creation unit 1-2 is also M
It is created and displayed by the MI control unit 1-1.

FIG. 4 is an explanatory diagram of automatic generation of the window control table. In the figure, the left half shows the operation screen and the right half shows the internal processing. By operating the window control table creating unit 1-2, the knowledge base construction support tool operates.

In the embodiment of the present invention, the knowledge base can be constructed after the system has been introduced to the end user, and thus such a knowledge base construction support tree is supported.

First, the operator assembles a target record by instructing the target item of each operation menu on the screen.
For example, the first item "update question text, update conditional statement, update requirement (rule), update conclusion" is displayed in one screen. Here, for example, the operator selects the update of the question text by clicking the mouse. For updating the question text, there are classification items or question texts, which are displayed. Similarly, select the question text by clicking the mouse.

By this selection, the window control table creating section 1-2 automatically gives a window control ID and assembles the id. That is, the id is stored in the specific area of the target record here. Since the update of the question text is selected, the classification screens are sequentially displayed to the point where the question text can be specified, and the instruction is prompted. Therefore, for example, it is displayed whether the question sentence, that is, the consultation content is the A situation, the B situation, or the C situation. Then, the operator designates the target item.
Select the items linked to the classification items you want to add on this screen by clicking the mouse. Figure 3 above
As shown in, the question is at the bottom of the final tool, so such a selection is made. For example, when the C situation is clicked, the contents of the C situation are displayed.

For example, "who is C mainly, about the state of C, about the environment of C" is displayed. Here, when "about the environment of C" is instructed, the tit is assembled for the first time. This is because there is no question text below this. As will be described later, "tit" is the content of the display title. Further, for example, when the environment of C is not registered, the title is assembled by registering here.

Then, an inquiry is made as to whether the question text is a Y / N selection question text or an n-choice selection question text, and the question type is selected by mouse clicking. By this selection, the SYS, msg, pb, etc., which will be described later, are assembled. If the user indicates that the question text is a Y / N selection question sentence, "About environment of C" is displayed in the title provided at the upper part of the screen, and the display data of the remaining YES / NO items is requested. . Here, the operator gives an affirmative or negative instruction, and inputs information corresponding to the instruction using the keyboard. By operating as described above, tree-structured screen data can be registered.

In the above description, the case of selecting and inputting the Y / N-selected question has been described, but the same applies to the case of inputting an n-choice question sentence, question classification, and further conclusion classification and conclusion. Incidentally, these operations will be described later in detail.

FIG. 5 is an explanatory diagram of items constituting the window control table. The window control ID is the ID of the window control table and is linked to each table. The id is composed of a screen format, a start X coordinate, and a start Y coordinate, and is an instruction to display the designated screen format at the start XY coordinate. The tit is composed of "display title contents" and is an instruction to display "display title contents" in the screen title bar. The upper part of the screen operation in FIG. 4 described above is the content of this title. msg is the number of displayed messages and "message content", and is an instruction to display "message content" on the message display line. sys is composed of a system control ID and "system control content", and is an instruction to perform screen control according to this system control ID. c
d is the number of Y / N selection items and "item contents", and Y
/ N is an instruction to control the window of the screen that requires selection (check box). yn is composed of "message body at the time of affirmation" and "message body at the time of denial", and is an instruction to display the information on the screen when the Y / N selection is affirmative or negative. lb is an instruction for controlling the window of the screen displaying the n-choice selection, which consists of the number of lines to be selected, the item determination flag, and the "jump destination window control ID". pb is the number of push buttons, the initial active button,
Consists of "button contents" and destination window control ID,
A push button is an instruction to control the required window of the screen.

The window control table consists of the above items. In the above-described embodiment of the present invention, when creating the window control table, a record is created by the first instruction operation. FIG. 6 is an explanatory diagram of the window control table. When generating this window control table, first, the record RX in the upper part of FIG. 6 is created by the operation of the operator in FIG. 3 described above. Then, when the table conversion is executed at the end, it is automatically created as each table including the menu label name, the menu classification, and the like. Although the creation of the question screen has been described above, the classification of questions, the classification of conclusions, the creation of conclusions, etc. are also created in substantially the same manner.

Using the window control table created by the above-mentioned operation, the MMI control section 1-1 creates a screen by the following operation and displays a question to the operator. FIG. 7 is a flowchart of the window display process. In step S1, the MMI control unit 1-1 operates and “Msg Close Dlg function issuance (screen deletion),
Jump destination window control table interpretation, Msg Crea
t Dlg function issue (screen generation) ”is executed. Dlg is an abbreviation for a dialog box, which means a window that constitutes one screen,
One table structure constitutes one dial box. "Close" and "Creat" indicate that the screen is erased (CLOSE) or displayed (CREAT). Therefore, depending on the information that can be expressed in msg, the MMI control unit 1-1 performs some action (selection).
If there is, the screen currently displayed is deleted (CLOS
E) Then, a series of operations such as interpreting the window control table, searching for a screen to be displayed next and displaying it (CREAT) are performed.

After the execution of step S1 described above, the process waits for an instruction to display the jump destination screen in step S2. Then, when the operator clicks the item on the display screen with the mouse in step S3, the instruction is added to the MMI control unit 1-1, the above-described step S1 is executed, and the same operation is repeated. By repeating this, a tree-structured screen corresponding to the screen is displayed, and by responding to the sequentially displayed questions, a response to the question sentence is obtained.

Returning to FIG. 2, the description will be continued. The inference server is provided in the host computer 2, and includes a knowledge compiler unit 2-1, an inference function unit 2-2, a knowledge base management function unit 2-3, a knowledge base file 2-4, and a knowledge editor unit 2.
-5.

Window control table creation section 1-2 and M
The MI control unit 1-1 is connected to the communication control unit 1-3, and the input information to the inference server, that is, the input data when requesting inference, the knowledge information, that is, the knowledge base, and the window control information are It is transmitted via the communication control unit 1-3. Further, for example, the obtained conclusion is the communication control unit 1-
3 is added to the MMI control unit 1-1 via 3 and displayed to the operator.

In the embodiment of the present invention, the question item is set in advance, the program is made in accordance with the question item, and the conclusion is obtained from the answer to the question, but the conclusion is made from the knowledge base by the inference server in the host computer 2. You have to edit that knowledge base. The knowledge editor section 2-5 edits the knowledge base.

FIG. 8 is an explanatory diagram of automatic knowledge base generation. The automatic generation of this knowledge base is also started from the knowledge base construction support tool. In the automatic generation of the window control table shown in FIG. 4, the update of the question sentence was selected, but the update of the requirement rule is selected to operate the knowledge base management function unit 2-3.

When the requirement rule update is selected with a mouse click, a request for inputting the conclusion and the answer to the question which is the requirement for the conclusion is added from the knowledge editor unit 2-5, and the operator clarifies the relationship between the conclusion and the question. .

First, an instruction for selecting a conclusion is added from the knowledge editor section 2-5 via the communication control section 1-3, and the conclusion A and the conclusion B for updating the requirement are displayed. When adding a conclusion, the operator points the area where the conclusion is not displayed by mouse clicking. When adding or changing a rule for the conclusion A, the conclusion A is designated by clicking the mouse. At this time, if nothing is stored in the conclusion A, the required question sentences are sequentially designated by mouse clicking, the screen is displayed, and the answer value is selected. For example, in question a, “yes”, in question b, the answer is n, and answer 2 is selected. By operating in this way, the requirement rule is instructed and a matrix table of questions and conclusions is created inside the knowledge editor unit 2-5.

Incidentally, in order to create the matrix table, YES /
A NO question record, an n-choice question record, and a conclusion record are required, and therefore these records are each managed in one dimension. As described above, the conclusion name of the requirement rule generated from those records is called by the conclusion record, and the question item necessary for deriving the conclusion is selected from the question record and the answer is input.

Window control table creation unit 1-2 and M
A communication control unit 1-3 is connected to the MI control unit 1-1, and the input information to the inference server, that is, the input data when requesting inference, the previously input knowledge information, and the window control information are It is transmitted via the communication control unit 1-3. Further, for example, the obtained conclusion is added to the MMI control unit 1-1 via the communication control unit 1-3 and displayed to the operator.

FIG. 9 is a matrix table chart, and the requirement rule in which the answer to the question and the conclusion are linked to the question sentence is expressed in the matrix table of FIG. For example, the titles of the question sentences are age, sex, question A, and question B, and the conclusion names are conclusion x and conclusion y.

After the input of the requirement rule described above is completed, the knowledge compiler section 2-1 executes the knowledge base to generate a knowledge base from the knowledge rule. This knowledge compiler section 2-1
By executing the above, the knowledge base is generated from the data obtained by dividing the matrix table into the vertical axis and the horizontal axis. Further, it is stored in the knowledge base file 2-4.

FIG. 10 is an explanatory diagram of knowledge base generation for generating a knowledge base from knowledge rules. The knowledge rule associates each conclusion with which question on the horizontal axis is the conclusion x on the vertical axis and what the answer is. In the knowledge rule,
The vertical axis is XXXXXXXXX-conclusion x (ID-result),
The horizontal axis is age-num-gender-sym-question at. f-
Question b-sel (question sentence data type), data is XXX
It is composed of XXXXXX-65-1-T-2 (ID-answer value). Then, by the knowledge compiler section 2-1 shown in FIG.
A knowledge base of the structure is generated as shown in the lower part of FIG. This compilation is made to correspond to the knowledge base in the inference server used, and this compilation allows the end user to freely generate a question sentence and further a knowledge base after the system is delivered.

A final knowledge database is created by the knowledge compiler section 2-1 and the knowledge base file 2-
Stored in 4. Then, the operator operates the MMI control unit 1
When an answer to a specific question is input by -1, and a record of the answer is sent to the inference server, the knowledge base management function unit 2-3 activates the inference function unit 2-2 and requests a conclusion for it. The conclusion obtained by the inference function unit 2-2 is output from the knowledge base management function unit 2-3 to the MMI control unit 1-1 via the communication control unit 1-3, and the conclusion is displayed on the display device.

With the above operation, in the embodiment of the present invention, the business operation terminal 1 is operated to input the question screen and the conclusion information, and notify the reasoning server in the host computer 2 as a knowledge base. As a result, it is possible to obtain a conclusion for the answer that each operator subsequently inputs via the MMI control unit 1-1.

The operation of the present invention will be described in detail below. FIG. 11 is a flowchart of the question classification screen generation, FIG.
2 is a flow chart of question screen generation. The flowcharts of FIGS. 11 and 12 show the operation of the window control table creation unit 1-2 described above. When the classification item and the question text are selected from the update of the question text, the processing of FIGS. 11 and 12 is executed.

FIG. 11 shows the processing when a classification item is instructed. When the classification screen generation starts to be executed, the question classification screen is displayed in step S14. This question classification screen is a display of each situation such as A situation, B situation, and C situation.

Then, when the operator selects one item on the screen in step S15, the instructed item, that is, the position when the item is clicked, is determined in step S16. When the instruction is blank, additional input of a new item is prompted in step S4, and the input result is added in step S5. That is, a question classification sentence is newly added to the question classification record. This comes after the C situation, so it is the situation entered.

On the other hand, when the area in which the item exists is instructed, for example, the situation A is instructed, the item of the question classification code is searched in step S6, and the displayed screen is erased in step S7. Then, it is determined in step S8 whether the menu label name exists. For example, when the A situation is selected, the menu label name of the next situation is NSGT0002.
However, since there is a question about the configuration on the lower side, it is valid in this determination. Therefore, in order to display the next lower item, the question classification record of the menu label name is searched in step S9. Then, the screen is assembled and the process is repeated from step S14. At this time, the contents of the menu label name MSGT0002 are displayed.

It is assumed that such instructions are sequentially given and finally is selected. At this time, it is determined in step S8 that there is no menu label name. That is, in the state of D, nothing is written in that state, and a new question classification record is added in step S11. Question classification record MS
GT002 is MSGT0002 "SUB""Asituation"
"MSGT0001""Aboutconfiguration""Dstate""M
SGT003 ”. In addition, a menu name automatic numbering of MSGT003 is performed and a classification record is created. It should be noted that here, the basic record has been configured, and has not been completed yet. Then, in step S12, a process of writing the menu label name into the newly added question classification record is performed. Ie D state and MS
GT0002 is written by being delimited by delimiters (inserted as indicated by arrows SY1 and SY2).

Subsequently, in step S13, the automatically assigned menu label name is written in the original question classification record.
This automatically assigned menu label name is MSGT000
3, the menu label name MSGT0003 is stored in the final position of the question classification record MSGT0002. Then, the screen assembly processing step S10 is executed. At this time, in step S13, the item names inside the D state are sequentially determined, and when the next round comes, the lower order items in the D state are further determined in step S13. For example, the item input in the D state described above may be sequentially input, and the series of steps S1 to S13 described above may be performed.
It may be configured such that they can be configured one by one.

FIG. 13 is a structural diagram of a question classification record. By the above-mentioned operation, the “menu label name”, “menu classification”, “title”, “previous menu label name”, “question classification sentence 1”, “next menu label name” ... Are repeated in sequence.
Here, the menu label name is MSGTnnnn (nnnn
Is a number), which is automatically added by the question screen generation process. Menu classification is "SUB"
This is fixed and becomes "SUB" in the case of a question classification record. The title is the item name clicked and selected on the question classification screen and is the question classification sentence of the question classification record. The previous menu label name is MSGTnnnn, which is the menu label linked before the record above. The question classification sentence is the question classification sentence finally added to the screen. The next menu label name is the menu label name of the next succeeding record, and is MSGTnnnnn, QUINNNnn, or QUNSnnnn depending on the type of record. Here, MSGTnnnnn represents the menu label name of the classification record, and
nnn represents the menu label name of the Y / N question sentence record, and QYNSnnn represents the menu label name of the n-choice question sentence record.

When the question text is updated and the question text is instructed in the basic menu, the question screen generation flowchart of FIG. 12 is executed. In the embodiment of the present invention, the question sentence is always placed at the bottom of the tree structure.

First, in step S21, a question classification screen is displayed. When this process is executed for the first time, the record with the menu label name MSGT00001 is displayed.
When the operator clicks and selects the target item in step S22, the item of the question classification record is searched in step S23. Then, the displayed screen is deleted in step S24. Then, in step S25, it is determined whether or not the next menu label name is present or blank. If it is present, the question classification record of the next menu label name is searched in step S26, the screen is assembled in step S27, and the step is repeated. Repeat from S21. By repeating this, the items on the question classification screen are sequentially designated.

Sequentially these steps S21 to S
By executing 27, for example, the menu label name MSGT0
001, MSGT0002. Further, if the D state is selected and becomes blank, it is determined that it is blank in step S25, and it is determined that the next sentence is a question sentence. That is, the D state is the final question. At this time, in step S28, Y
/ Displays an instruction to select N questions or n alternative questions. Then, when the operator's generation screen instruction is n choices, that is, an n choice question record is created in step S30 in the question about "configuration". The menu label name is automatically numbered, and for example, the menu label name is QUNS0001.

Then, in step S31, the process of writing the previous menu label name into the newly added record is performed, and then in step S32, the automatically assigned menu label is written into the record in the original question classification. These two steps link the records before and after. Succeedingly, in step S33, the selection sentence is input, and in step S34, the entire process of writing it into the target record is completed.

When it is determined in step S29 that the generated screen is the Y / N question text, for example, when the C situation is selected and there is no solution for the C situation, and the Y / N question text is input. (In the case of), a new Y is obtained in step S35.
/ N Add question record. This menu label name is determined by automatic numbering, and is assigned, for example, QUIY0001. Then, in step S36, the previous menu label name is written in the newly added record, and then step S
In step 37, the process of writing the automatically assigned menu label name in one question classification record is executed. These two steps link the records before and after.

Next, the question explanation is input in step S3.
8. The input information is written in the record in step S39, and the entire process is completed. On the other hand, in step S25, the next menu label name is QUINnnnn or QUN.
If it is Snnnn, the next record is a question sentence, so the corresponding record of the next menu label name is searched for in step S40, and the screen is assembled in step S41.
Then, the question screen is updated in step S42. In the question screen generation, an existing question record is not instructed except when newly created, and when the question text is instructed in this way, it is determined that the question record is updated and the update process is executed. With the above configuration, YES / NO question records and n-choice question records are created.

FIG. 14 is a structural diagram of a YES / NO question record, and FIG. 15 is a structural diagram of an n-choice question record. Y
The ES / NO question record includes “menu label name”, “menu classification”, “title”, “previous menu label name”, “question sentence 1”, “YES explanation sentence”, and “NO
It consists of repetition of the number of "Explanation" items. In the embodiment of the present invention, not only one YES / NO but also YES or NO can be instructed over a plurality of items. By repeating this plurality of sentences, YES is sequentially obtained.
/ NO is created, and some questions are requested at the time of question execution. Here, the menu label name is QUIYN
nnnn (nnnn is a number), menu classification is tf
(Fixed). The title is the item name selected in the previous question classification screen, that is, the question classification name of the previous question classification record, the previous menu label name is MOSGTnnnn, the question text is the additionally entered question text, and the YES explanation text is the question text Yes. The explanatory text when selected and the NO explanatory text are the explanatory text when No is selected in the question text.

On the other hand, the n-choice question record includes "menu label name", "menu classification", "title", "previous menu label name", "selection statement 1", "selection statement 2" ...
As in, the selection sentence for the number of items continues at the end. Here, the menu label name is QUNSnnnn, and the menu classification is S
It is fixed by EL. Also, the title is the item name clicked and selected on the previous question classification screen, and the previous menu label name is MSG
Tnnnn and the selected sentence are the additionally input selected sentence. The Y / N question record and the n-choice question record having the above-described configurations are created by the processing of FIGS. 11 and 12 described above. Then, a tree-structured question screen can be constructed by creating a window control table from these question records.

FIG. 16 is an explanatory diagram of creating a matrix table. The Y / N question record YN1 and the n-choice question record NX1 are related on the matrix table MXT. The question a corresponds to the question a of YN1 of the Y / N question record, and the question b corresponds to the content of NX1 of the n-choice question record. For example, the question a is a YES / NO question, so the answer is y, and the question b is an n-choice choice, so the answer is 1. In addition, the number of the result of the alternative n indicates the number of the selected item. Related to this is the conclusion record PX1.

Unlike the question screen, the conclusion screen and the conclusion classification screen in the embodiment of the present invention are expanded as a tree structure of only one layer to automatically generate a record. FIG. 17 is a block diagram of the conclusion screen. The conclusion classification screen, the conclusion screen shown in FIG.
The conclusion detail screen is represented by a conclusion classification record, a conclusion record, and a conclusion detail record. FIG. 18 is a flowchart of the conclusion screen generation. The conclusion screen generation process starts to be executed when it is instructed to update the conclusion. First, step S61
The screen is assembled according to the conclusion classification record. The conclusion classification record indicates, for example, whether the conclusion is a national system, a Tokyo system, or a ward system when drawing a conclusion. This classification record is composed of a plurality of conclusion classification names n1, n2, ... As shown in FIG.

Subsequently, the conclusion classification screen is displayed in step S62. When the operator clicks and selects the target item in step S63, the click position is determined in step S64. When a blank is designated, for example, as shown in FIG.
When the blank portion of the system is designated in the system fixed system classification BX in the middle, an additional item is input in step S65. For example, the ward system. And step S66
Write the conclusion classification name in the conclusion classification record. Then, the process ends. By this processing, the system of the ward is registered in the conclusion classification record. On the other hand, when the existing item is designated, it is determined that the item exists in step S64, and the item classification name item search of the conclusion record is executed in step S67. Then, the displayed screen is erased in step S68, and the conclusion screen is assembled in step S69. For example, if you select () the national system,
Gymnasium, parts, etc. are displayed. Then, the operator selects one of these items in step S70. This selection is determined in step S71, and when an item exists, the conclusion screen is displayed in step S72 and the operator changes the content.

On the other hand, if blank is selected, a new conclusion record is added in step S73. Conclusion records are "INF0001", "gymnasium", "national system"
, “INF002”, “parts”, and “country system” already exist, and the following record INFO003 is created by the allocation number of this process. Then, the conclusion classification name is written in the newly added record in step S74, and then the conclusion name is input in step S75.

The conclusion name is, for example, "referee",
Further input of the conclusion details is performed in step S76, and the input content is written in the conclusion detail record in step S77. For example, the items of the national system include a national gymnasium, national parts (volleyball or net), and referees (referees when performing volleyball). In this example, when an organization wants to play a volleyball game, they can see the conclusions as described above in the result of inference to the question. You can also see the metropolitan system rather than the national system. In the embodiment of the present invention, the above-mentioned display can be made because the conclusion is one layer and the details of the item are visible.

In this way, the conclusion classification record shown in FIG. 19, the conclusion record shown in FIG. 20, and the conclusion detail record shown in FIG. 21 can be obtained by the conclusion screen generation process. The conclusion classification record has a plurality of conclusion classification names.

The conclusion record is composed of "file name", "conclusion name", and "conclusion classification name". Here, the file name is INFnnnnn, which is the file name included in the conclusion details. The conclusion name is the additionally entered conclusion name, and the conclusion classification name is the item name clicked and selected on the conclusion classification screen, that is, the conclusion classification name of the conclusion classification record.

Conclusion The structure of the detail record is "file name"
And the "conclusion details". File name is INTFnnn
n, the detailed statement of conclusion is the input detailed statement of conclusion. FIG.
Return to and continue the explanation. The conclusion a in the matrix MTX is indicated by the conclusion KX1, which gives the classification name. Further, the conclusion and its details can be obtained from the conclusion classification name. In the past, only the conclusion was obtained, whereas in the embodiment of the present invention, the details can be seen separately for each item from the conclusion obtained.

To generate the above-mentioned matrix table MXT, a matrix generation process MXS is executed. First, a conclusion is selected in step S81, and then step S8.
In step 2, it is judged whether the requirement rules have been entered. When the requirement rule is not completed (not yet), the question classification is selected in step S83, and the answer is selected in step S84. For example, if the result is the conclusion x, the question a is selected and the answer Y is used as the requirement input. Further step S82
To execute. If the requirement rule is not yet set, the question b is selected in step S83, and the answer 1 is input as a requirement in step S84. If it is limited by these two selections, the requirement rule input is completed and the process ends. As a result, the requirement rule data becomes "conclusion record name", "event value 1", "event value 2", "event value 3", "event value 4".
・ ・ It becomes. For example, the matrix table MXT described above
, Then INFT0001, *, *, y, 1,
*, *, ... Here, * is filled in as a non-input for questions without requirement input. For example, age and gender are not required, so they are marked with *. The requirement rule data is created by the above processing.

FIG. 22 is a flow chart of matrix table generation. Although the matrix table generation is briefly described in FIG. 16 described above, it will be described in more detail in FIG.

When determining the requirement rule for selecting one conclusion, the processing shown in FIG. 22 is started. In step S100, a process of selecting a conclusion such as the country system of INF0001 is executed. Then, it is determined in step S101 whether or not the input of the requirement rule is completed. If it is complete, exit.

If not completed (not yet), step S1
At 02, the question classification screen is displayed. And step S1
The item is clicked and selected in 03, and the linked window control table is interpreted in step S104. Then, in step S105, the displayed screen is erased, and in step S106, the question screen is checked. When the screen is not the question screen (NO), the screen is assembled in step S107 and executed again in step S101. By repeating this sequence, the tree-structured question classification is sequentially performed, and finally, in step S106, the question screen (Y / N question or n
Alternative question) is determined.

When a Y / N question is asked, a screen such as "Can you play baseball?" Is displayed in step S108, where the screens are assembled and displayed. Then, in step S109, the operator clicks and selects the requirement.

Then, in step S110, the instructed requirement is added. When the conclusion is selected in step 100, no conclusion has been set for INF0001 as shown in FIG. 23, but Y / Y is determined in step S109.
When the display selection "Can you play baseball?" On the N question screen is set to Y, for example, the item is set (Y) as shown in FIG. Subsequently, the interpretation of the window control unit immediately above it is executed in step S111. Then, the displayed screen is displayed in step S112, the screen is assembled again, and the process is executed from step S101. On the other hand, when it is determined in step S106 that the question is an alternative choice, step S1
At 13, the screen is displayed and assembled, and the assembly is displayed. If this is for example a configuration, multiple (3 or more), 2 people, 1
Since there is an item of "people", one of them is selected in step S114. As a result, for example, when (3 or more) is selected, the item is the first item, so 1 is set in the "configuration" section, and the requirement rule shown in FIG. 25 is added.

After step S114, the requirement rule is added in step S110, and the above-described operation is repeated. When the input of the requirement rule is completed by repeating the above, the processing is ended. This creates a matrix table.

FIG. 26 is an input flow of screen development and question answer. When the processing is started, the question classification screen is displayed in step S121. For example, the operator designates one item (C situation) of the consultation content in step S122. Then, the linked window control table is interpreted in step S123, and the displayed screen is erased in step S124. Then, in step S1, a question screen is displayed.
If the question screen is not displayed (NO), the screen is assembled in step S126, and the step S120 is executed again.
Do more.

When the question is Y / N, the screen is assembled and displayed in step S127. For example, when the C situation is selected, the C situation is a question "whether baseball can be played". On the other hand, the operator performs step S128.
Answer with. Based on this answer, the answer is added to the question content table in step S129. At this time, the question content table contains "Can you play baseball?" And "Y". Then, the window control table immediately above is interpreted in step S130, the displayed screen is erased in step S131, and the process is repeated from step S121.

On the other hand, step S125 of determining whether the question screen is displayed.
When the n-choice question is selected by, the question is displayed in step S132. For example, when “About configuration” in the situation A is selected, “About configuration” and a plurality of (3) two persons and one item are displayed. Step S13
If any one of them is selected in 3, the question content table is added in step S129. By this addition, a table consisting of "I can play baseball", "Y", "about composition", and "1" is created.

With the above operation, as shown in FIG. 27, "question 1 character string", "question 1 answer", "question 2 character string",
A question content table is created such as "answer to question 2" .... Here, the delimiter is, for example, one or more half-width spaces on the database. In addition, the question n character string is the Japanese character string itself of the question sentence, the answer of the question n is a numerical value, for example, 65, 100, etc. if the age, symbol is male, female, selection is 1, 2, 3. The nth alternative question. Moreover, the truth is Y and N.

By executing the question answer input flow as described above, the question content table shown in FIG. 26 is created. The created question content table is transmitted to the inference server via the communication control unit 1-3 in FIG.

In the above-described embodiment of the present invention, the question classification screen has a tree structure. In the embodiment of the present invention, it is not necessary to respond to questions in a tree-like flow, and a plurality of icons for changing the flow of questions are provided in the lower part of the display screen or the like so that the question order is not such a tree-like flow. By using the plurality of icons, it is possible to move to a question screen or the like next to the upper branch position.

Since the inference server makes a conclusion from the question content table which is not answered in this order,
Operators can jump to unnecessary questions,
The total operation time can be shortened.

Since the matrix is constructed as described above and the knowledge base generation shown in FIG. 10 is created from this matrix and transferred to the inference server for inference, accurate information can be obtained. Further, since the question table is further generated and the inference corresponding to the question table is performed, the conclusion becomes accurate.

[0084]

As described above, according to the present invention, since the question and the conclusion are input from the screen, the question sentence corresponding to the user can be made by the user, and the conclusion can be made similarly. Furthermore, since it is configured so that it can be stored in the knowledge base, it is possible to build an intelligent base for inference.

[Brief description of drawings]

FIG. 1 is a principle block diagram of the present invention.

FIG. 2 is a system configuration diagram of an embodiment of the present invention.

FIG. 3 is a configuration diagram of a window.

FIG. 4 is an explanatory diagram of automatic window control table generation.

FIG. 5 is an explanatory diagram of a window control table.

FIG. 6 is an explanatory diagram of a window control table.

FIG. 7 is a processing flowchart of window display.

FIG. 8 is an explanatory diagram of automatic knowledge base generation.

FIG. 9 is a matrix table chart.

FIG. 10 is an explanatory diagram of knowledge base generation.

FIG. 11 is a question classification screen generation flowchart.

FIG. 12 is a question screen generation flowchart.

FIG. 13 is a structural diagram of a question classification record.

FIG. 14 is a structural diagram of a yes / no question record.

FIG. 15 is a structural diagram of an n-choice question record.

FIG. 16 is an explanatory diagram of creating a matrix table.

FIG. 17 is a configuration diagram of a conclusion screen.

FIG. 18 is a flowchart of conclusion screen generation.

FIG. 19 is a structural diagram of a conclusion classification record.

FIG. 20 is a structural diagram of a conclusion record.

FIG. 21 is a structural diagram of a conclusion detail record.

FIG. 22 is a flowchart of matrix table generation.

FIG. 23 is a matrix table chart.

FIG. 24 is a matrix table chart.

FIG. 25 is a matrix table chart.

FIG. 26 is a drawing development and question answer input flow.

FIG. 27 is a structural diagram of a question content table.

[Explanation of symbols]

 1 Business Operation Terminal 1-1 MMI Control Unit 1-2 Window Control Table Creation Unit 1-3 Communication Control Unit 1-4 Window Control Table File 1-5 Dialog Input Data File 2 Host Computer 2-1 Knowledge Compiler Unit 2-2 Inference function section 2-3 Knowledge base management function section 2-4 Knowledge base file 2-5 Knowledge editor section

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Eiji Adachi 3-3-3 Toyosu, Koto-ku, Tokyo NTT Data Communications Corporation

Claims (3)

[Claims] 1. A question classification screen that represents the classification of the following questions is used to configure a question on the trunk side of the tree, and a question screen on which no questions exist thereafter configure the question at the last branch of the tree. At least one question screen is determined by one or more questions on the classification screen (ST2), and a conclusion is inferred from the operator's response (ST2) to the at least one question screen and the knowledge base (ST1). In the knowledge base system for obtaining (ST4), the knowledge base stores the responses to one or more question screens in advance as a requirement rule in a matrix structure corresponding to the conclusion to be obtained (ST1). Base storage method. 2. The question screen includes a Y / N question record for instructing a two-choice selection screen represented by YES and NO, and an n-choice question record for selecting one of n items. 2. The knowledge base storage method according to claim 1, wherein the knowledge base storage method is classified into and managed. 3. The conclusion comprises a conclusion split screen for splitting the conclusion, a conclusion screen split by the split screen, and a one-stage tree screen consisting of a conclusion detail screen showing the details thereof. 2. The knowledge base storage method according to claim 1, wherein: