WO2000026779A1

WO2000026779A1 - Case-based reasoning method and system thereof, and readable recording medium with built-in program for executing case-based reasoning

Info

Publication number: WO2000026779A1
Application number: PCT/JP1999/006085
Authority: WO
Inventors: Toshio Hamada
Original assignee: Shinko Research Co., Ltd.
Priority date: 1998-11-02
Filing date: 1999-11-02
Publication date: 2000-05-11
Also published as: JP2000137615A; AU6488199A

Abstract

When a phrase is input via an input unit (41), a phrase case retrieving unit (21) retrieves cases including the phrase based on a case file (51), a scoring unit (22) scores the cases up to a maximum score of 40 according to phase-included matching levels, and a rearranging unit (23) rearranges the scored cases in a descending score order for displaying on a monitor (42) along with an associated question. When answering to a question displayed from a different view point, a re-scoring unit (25) recalculates and increases or decreases scores according to the answer result. If a maximum score of a certain value of 80 is reached by the re-scoring, an evaluation judging unit (26) recommends a case with the maximum score.

Description

Description Case-based reasoning method, system and readable recording medium with built-in program for executing case-based reasoning

The present invention relates to a case-based reasoning method, a system, and a readable recording medium storing a program for executing case-based reasoning for an input phrase. Background art

Conventionally, in a case-based inference system that has a plurality of similar cases in the case base, case-based inference is performed on the input case, and information is provided on a set of matching cases for the problem template. An inference engine that seeks advice from a group of operating cases has been proposed.

(Japanese Unexamined Patent Publication No. Hei 6—508702). In other words, the case-based inference system described in Japanese Patent Application Laid-Open No. 6-5080702 executes case-based inference to extract a case that matches the customer description input and displayed on the display panel. Then, if the match is high, an advisory message is given to the customer, and in response to the advisory, “application 6001 is a question-answer pair 60 0 Step 2 of case matching can be performed using 8 (the upper right column on page 8 of the gazette), and the above case-based inference is repeatedly executed to extract the best match case. Things.

Further, as an inference system based on case-based inference, those described in JP-A-3-26601 and JP-A-3-148727 have been proposed. .

The case-based inference system described in Japanese Unexamined Patent Application Publication No. 6-50807002 repeatedly executes the case-based inference of “Step 202” in response to a response to an advisory message. Although it is possible to extract the best match case for the description in (1), the execution of case-based inference is only performed by repeatedly executing case-based inference for the description and advice from the customer. Extracted in every The matching cases obtained are inconsistent, and the same case-based inference is performed, which naturally limits the extraction of the best matching case. In addition, the inference systems described in JP-A-3-26631 and JP-A-3-148727 devise a case from an input, and devise the similarity at the time of extraction. It does not relate to case-based reasoning techniques.

The present invention has been made in view of the above, and in a case-based reasoning, by adopting a procedure based on a different viewpoint, it is possible to extract a case that is more matched to a word input by a user, or It is an object of the present invention to provide a case-based reasoning method, a system, and a readable recording medium storing a program for executing the case-based reasoning, which can support the evaluation of the reasoning result. Disclosure of the invention

The case-based reasoning method according to claim 1 performs a case search between at least one or more input words and a case file and, based on the degree of ^^ with the words, generates a case Executing the scoring and reporting the result; reading and reporting the question associated with the searched case from the question file; and performing the search on each of the searched cases according to the answer to the question. And performing a re-scoring on the data.

The case-based inference system according to claim 12 further includes a case file, a question file, and a file in which the question and the case are associated with each other, and further includes a display unit and a case file for the input phrase. Case searching means for performing a case search on the basis of: a scoring means for scoring each searched case in accordance with the degree of matching with the phrase; and displaying means for associating the searched case with a score. First display control means for outputting to the display means a question associated with the searched case, and a second display control means for outputting to the display means a response to each of the displayed questions. Rescoring means for rescoring a score for a case. A readable recording medium storing a program for executing the case-based inference according to claim 17, wherein at least one or more of the inputted URLs is searched for a case with a case file, and the degree of matching with the above is determined. And then scoring each case found The result is reported, the question associated with the searched case is read out from the question file and reported, and the re-scoring is executed for each searched case according to the answer result to the question.

According to these, at least from users using Man-Machine Yuichi face etc.

One or more words and phrases, including sentences, are input as an assignment, and a case search is performed on the input words using a case file as the first stage. Scoring is performed on each case according to the degree of matching of the searched case with the above-mentioned phrase, for example, according to the number of words included in the case where there are a plurality of cases. The result is reported to a display device or the like. Then, the question associated with the searched case is read from the question file and notified to a display device or the like. When the user answers the question, rescoring is performed on each of the searched cases according to the answer result. And

Provide the user with a recommendation that the case with the highest score is the one with the highest likelihood for the task.

The case-based reasoning method described in claim 10 searches and displays a question including at least one or more input words in a question file, and displays an example associated with the displayed question from the case file. It includes a step of reading and displaying, and a step of executing re-scoring on each of the displayed cases according to the result of answering the displayed question.

Further, the case-based reasoning system according to claim 13 includes a case file, a question file, and a file in which the question and the case are associated with each other, a display means, and a question for searching for a question including the input word in the question file. Searching means; means for reading a case associated with the searched question from the case file; first display control means for outputting the searched question to the display means together with the read case; and the displayed question Means for re-scoring each of the displayed cases according to the result of the response to

In addition, the readable recording medium having a program for executing the case-based inference described in claim 26 searches and displays a question including at least one or more input words in a question file and displays the question. A case associated with the displayed question is read from the case file and displayed, and re-scoring is executed for each of the cases being redisplayed in accordance with the result of answer to the displayed question. According to these, questions containing ^ entered are displayed together with examples associated with these questions, and the user can respond to these displayed questions, so that the user is familiar with the field. In some cases, the likelihood of reaching a more probable case more quickly increases. BRIEF DESCRIPTION OF THE FIGURES

FIG. 1 is a block diagram showing an embodiment of a case-based reasoning system.

FIG. 2 is an explanatory diagram showing an outline of a procedure for creating a phrase case index file from a case file using the database construction unit.

Fig. 3 is a drawing illustrating the contents and structure of each file. (A) is a case file,

(b) shows a question file, and (c) shows a question case index file.

FIG. 4 is a diagram showing a display example of a monitor screen.

FIG. 5 is a diagram showing a screen showing a data input state.

FIG. 6 is a diagram showing a display screen at the time when the first search step is completed.

FIG. 7 is a diagram showing a screen showing a state in which a question on the first line or a question on a desired line in the question display column 422 is selected.

FIG. 8 is a diagram showing a screen showing an example of answers to a question.

FIG. 9 is a diagram showing a display screen showing an example of the inference result.

FIG. 10 is a diagram showing a display screen in a state where the menu “case information” is designated.

FIG. 11 is a diagram showing a display screen in a state where the menu “case reference diagram” is designated. FIG. 12 is a flowchart showing the procedure of the case-based inference method.

FIG. 13 is an explanatory diagram showing the content of “phrase search” in step S7.

FIG. 14 is an explanatory diagram showing the contents of “re-scoring” and “score order rearrangement display” in steps S 19 and S 21.

FIG. 15 is a diagram showing a display screen after a question search.

FIG. 16 is a flowchart showing a second embodiment of the present inference system.

FIG. 17 is a flowchart showing a third embodiment of the present inference system.

Figure 18 is a chart showing the results of question analysis, (a) is a hierarchical chart, (b) is a list chart,

(c) is an inclusion diagram, and (d) is an inclusion list chart. Figure 19 is a hierarchical chart showing the relationship between other cases and questions. BEST MODE FOR CARRYING OUT THE INVENTION

FIG. 1 is a block diagram showing an embodiment of a case-based inference system. In the figure, the system includes a processor 10 for executing case-based reasoning and the like, a case-based reasoning engine 20, and a man-machine interface 40 and a case base 50.

The man-machine interface section 40 includes an input section 41 and a monitor 42. The input unit 41 employs a keyboard, such as a keyboard used in a normal computer, in which keys capable of inputting characters, numerical values, and the like, and keys for instructing specific functions are arranged as necessary. The monitor 42 employs a display device that includes at least character display and is capable of displaying at least characters. As long as the information can be exchanged between the user and the user, the man-machine interface section 40 may employ a mode other than key input, such as voice recognition means, as the input section 41. For the purpose of presentation, the monitor 42 is excellent in that it can simultaneously recognize a large amount of information. However, the present invention is not limited to this, and voice guidance may be performed using voice synthesis means.

The case base 50 includes a case file 51, a phrase case index file 52, a question file 53, and a question case index file 54.

For example, in the case where this system is adopted for quality control support, the case file 51 contains examples of various claims as past results (including those that can be predicted if necessary) It is also stored in association with the detailed diagram of the countermeasures and the case number as needed. In addition, the case file 51 includes, as well as the detailed explanation of the case, the case or the measure, and a detailed diagram of the case, as well as, for example, as shown in FIG. The case number (stored in the identification number field) and the question number (stored in the question field) for identifying one or more questions associated with each case are stored in a table format. It is preferable that the question numbers associated with each case number be stored in a hierarchically organized manner as shown in the question field. For example, in case numbers 5 and 7, question number 2 is positioned at the top, while in case number 10 the number is several from the top. In this embodiment, examples are those that are necessary or effective for receiving quality control support from the system, such as the details of abnormalities in the above-mentioned metal material manufacturing process and the symptoms caused by the abnormalities. (A question can be asked.) Refers to a form corresponding to the expression content, which is mainly expressed in sentences. On the other hand, words and idioms are mainly used to search for cases. For example, "compressor", "pump", "crack", "oil leak", "vibration", etc., but other negative words such as "do not start" from the viewpoint of symptom expression can be included. . In addition, the term "measures" may include general conclusions including causes.

I ^ The case index file 52 is content input from the user, and relates the fg ^ J that the user expects to include in the case to be searched to the case stored in the case file 51. In addition, each phrase (stored in the inquiry field) is stored in the table format with the case number (stored in the case number field) of the corresponding case. And the case may be stored in association with each other via an index formula or the like. This has an advantage that the storage capacity of the case can be reduced.

FIG. 2 is an explanatory diagram showing an outline of a procedure for creating a phrase case index file 52 from the case file 51 using the database construction unit 20a. The database construction unit 20a creates a phrase case index file 52 from the case contents and the case number field of the case file 51. That is, the database construction unit 2◦a extracts from the contents of the case what is expected as the user ig ^ J and creates a case number including each piece in a table format. In the example of Fig. 2, the phrase case index file 52 indicates that the phrase “compressor” is included in case numbers 3, 5, and 7, and 1 ^ 0 that is “pump” is the case numbers 5 and 7. , 10.

As shown in Fig. 3 (b), the question file 53 asks the user a question sentence (stored in the question sentence field) that can be answered alternatively, for example, "Yes, No". It is stored in association with the number (stored in the question number field). Further, in the question file 53, in addition to the above recorded contents, weight values for determining whether the score is to be increased or decreased according to the answer result of each question are stored as attribute values. The weight value may be a coefficient by which the score is multiplied, or a value by which addition or subtraction is performed.

As shown in Fig. 3 (c), the question case index file 54 has a table format in which the question numbers (stored in the question number field) correspond to the case numbers (stored in the related case number field). This is what we are building in the formula.

An example of the recording mode of the case file 51, the question file 53, and the question case bow i-file 54 will be described in detail with reference to the explanatory diagram of FIG. 3, for example, a part of the case file 51, Focusing on case number 5, case number 7, and case number 10 in the case number field, they all include question number 2 (the question corresponding to case number 7 is one of question number 2). Case numbers 5 and 10 include more than one.) In the question file 53, a question sentence corresponding to each question number is stored such that a question sentence "Is the model A?" Then, as the question case index file 54 stores related case numbers 5, 7, and 10 based on question number 2, the related case numbers for each question number are written in the arch I format. I'm caught.

The processor 10 includes a case-based reasoning engine 20 including a function unit for executing a search step as a first stage and a function unit for executing an inquiry step as a second stage. It has an I / F 31 and a display control section 32 as an interface between the interface section 40 and the interface section 40.

The data extraction unit 11 extracts words (in text code) for inference from one or more input sentences (contents of the task including the expected ones to be included in the case to be searched) input from the input unit 41. It is for extraction. From the input unit 41, for example, for the input statement such as "The pump is leaking oil." Or "The motor does not start.", "Pump", "Oil leak", "Mooring", "Starting" The phrase "no" is extracted from the sentence, and the phrase (term, idiom, negative form predicate) intended by the user is extracted from the sentence by using grammatical processing. I have. The input sentence is prepared so that a plurality of lines can be displayed on the monitor 42 (see 421 in FIG. 4). If there are multiple input sentences, each sentence is broken down by a line break for each sentence. It is identifiable, or the break of each sentence can be determined by the period. If it is known that a part or all of the input content from the input unit 41 is a word (for example, a line break without a punctuation in the line), the word Can access the case file 51 directly through the data extraction unit 11.

The operation judging unit 12 is for judging whether or not execution of a specific function has been instructed, especially for other than text input. Specific functions are described later on the monitor screen shown in Fig. 4. However, for example, input determination, recovery determination, and the like as conditions for shifting to the next process, and an instruction to perform a logical sum search or a logical product search, and the like.

The case-based inference engine 20 includes a phrase case search processing unit 21, a scoring unit 22 and a rearrangement processing unit 23 as functional units that execute the search step as the first stage. It includes a question case search processing unit 24, a re-scoring unit 25, and an evaluation judgment unit 26 as functional units that execute the inquiry step.

The detail display processing unit 27 is a functional unit for displaying the detailed contents of the best search case on the monitor 42.The analysis processing unit 28 analyzes the relationship between questions and cases, and the questions to create a case. When examining the validity of the included questions (expression method, number, order, etc.), make a table in an appropriate form and display it. Also, analyze the relationship between measures and cases, and measures This is a functional unit that also examines the validity of the measures included in, and displays them. The details will be described later. The case correction processing unit 29 is a functional unit for making a correction to the searched case.

The case search processing unit 21 uses the case file 51 and the phrase case index file 52 to search for relevant cases from the words extracted or passed through the data extraction unit 11. This is to search by inference. The phrase case search processing unit 21 reads a case including, for example, from the phrase case crossbow I file 52 by the case number, and reads the case contents of the case number from the case file 51 together with its countermeasures. . As will be described later, when multiple words are input, processing can be performed with each inference method depending on whether inference by logical OR or logical inference is instructed. It has been done.

The scoring unit 22 assigns a score according to the degree of matching (degree of matching) between the phrase and the case according to a preset method. In the first step, the degree of matching is 1 (%) (Hereinafter referred to as score 1), preferably a predetermined value of score 50 or less is the maximum value

(In the present embodiment, a score of 40) is assigned. By setting the likelihood of the search step in the first stage to a score of preferably 50 or less, the reliability of the first stage is ensured, and the significance of the second inquiry step is increased, and the certainty is higher. The case can be finally inferred.

In one embodiment of the scoring method, a score of 40 is set for a certain number of input words and for cases including all inquiries, and the number of included ^ According to the above, scores lower than the maximum value are sequentially calculated and set according to a preset calculation method (including a tabulated mode). For example, when performing inference on two terms, a score of 40 is given to a case containing two phrases, and a score of 20 is given to a case containing one phrase. Become like Note that all the searched cases may be displayed on the monitor 42 as search results.However, in consideration of the fact that the number of searched cases is rather complicated, such as when a large number of searched cases, For example, only those that clear the score of 10 or more may be the results of the first search step.

The sorting unit 23 sorts the searched cases in the order of the scores assigned to each case, and places them in the case display column of the monitor 42 (see 4 23 in FIG. 4) from the first row to the lower row. Are displayed in order.

The question case search processing unit 24 is a case extracted in the first search step. The question case index file 5 is used to guide the question sentences individually associated with the cases having the highest scores to the monitor 4 2. The question number assigned to the relevant case is read from 4 and the question sentence of the question number is read from the question file 53. Usually, there are multiple cases extracted, but one or more question sentences associated with each case are read from the question file 53, and are displayed in the question display column of the monitor 42 in the order in which the cases are displayed line by line. Displayed correspondingly. If there is a common question number for the retrieved cases, the display of the second and subsequent questions is omitted to reduce the complexity of the display.

The re-scoring unit 25 recalculates the score as the certainty of the case according to the answer of “Yes” or “No” to the question sentence corresponding to the case. As described above, the re-scoring calculation method uses the weighting values according to the answers “Y es” and “No” to the question, as described above. The current score given is increased according to the weight (match weight) set for “Y es” (for example, in the case of a coefficient, by the ratio) (the calculation result exceeds 100) In this case, the score may be set to 100 or 9), and for the answer of "No", the current score assigned to the corresponding case is set for "No" According to the weight (mismatch weight), it is set to be lowered by (for example, by a ratio) (may be the same as the weight value set for “Y es”). Resco Alling may be addition / subtraction instead of ratio calculation, or another method may be adopted. For cases that are not associated with a question (ie, cases that do not have the question), the weight value as the absolute weight is set corresponding to “Yes” or “No”. In this way, the score is increased / decreased even in the case not related to the question.

In the second inquiry step, the evaluation determination unit 26 determines whether the highest score has reached a preset value, for example, a value of 80, out of the scores obtained by the re-scoring at each re-scoring. Judgment is made, and when the highest score exceeds the value 80, the inference procedure is automatically terminated.

The detail display processing section 27 displays the countermeasures corresponding to the searched cases in the countermeasure display column (see 42.4 in Fig. 4) on the monitor 42 screen. By specifying a specific or arbitrary measure, a more detailed measure content and a detailed measure reference diagram are read from the case file 51 by specifying the display line, and the specified area of the monitor 42 (for example, a new A window is opened on the screen) and the details of the countermeasures are shown to the user. The detailed display processing unit 27 is also driven in response to an instruction to display case details and a case reference diagram.

FIG. 4 is a diagram showing a display example of the screen of the monitor 42. As shown in the figure, at the top of the monitor screen 420, there are multiple menus such as “Search”, “Clear”, “Response Evaluation”,.

“Case information”, “Case reference diagram”, “Measures details”, “Measures reference diagram” and “Exact search” are prepared in the form of icons, and the power of the input unit 41 in this menu display area is provided. The cursor is moved and clicked (or double-clicked as appropriate) using the sol indicating member 4 1 1 (for example, a mouse), and the function of the display menu is selectively executed. In the lower part of these display areas, the entered text is displayed sequentially for a specified number of lines. The “detailed explanation (keyword)” column 4 21, the question display column 4 2 2, and the case display column 4 2 3 And a countermeasure display column 4 2 4 is provided. Each column 4 2 1 to 4 2 4 is not limited to the number of lines shown in FIG. 2, and a larger number of lines is prepared. In addition, it is possible to scroll display in the line direction,

The content of (line) can be input or displayed.

The menu "Search" is a search by OR for multiple input URLs. Step: coarse search). The menu “Clear” instructs the deletion of the input contents for search displayed in the “Detailed explanation (keyword)” column 4 2 1 and the selective deletion of the contents of other columns. Things. The menu “Response evaluation” is for instructing the evaluation of the case, that is, re-scoring, after inputting one or more answers to the displayed question (determining the answer). In addition, it is also possible to selectively designate a mode in which evaluation is performed sequentially for each answer, and in this case, the menu “answer evaluation” is not required.

The menu “case information” is for instructing the display of “detailed explanation” for the case indicated as shown in FIG. The menu “Case Reference Diagram” is used to instruct a new window を to open and display a diagram (reference diagram), etc. that was captured as an image to show more details of the case as shown in Fig. 11 It is. The menu “Exact search” is used to specify a logical AND search (first search step: coarse search) for multiple input words.

Subsequently, the monitor screen 420 shown in FIGS. 5 to 11 will be described.

Fig. 5 shows a screen that shows the data entry status. In this display example, the word "scratch" appears on the first line of the "detailed explanation (keyword)" column 4 21 and the word "break" appears on the second line. Is displayed.

Fig. 6 shows the display screen at the end of the first search step, and the results of the search are displayed in the question display column 42, the case display column 42, and the countermeasure display column 42. I have. In the first line of the case display column 4 2 3, the score 40 is displayed, and the case number C 6 is displayed as `` crack during cold working, '' and in the second line, the score 40, and the case number C 9 is "Cracks during cold forging" are displayed. The first line of the countermeasure display column 4 2 4 contains “Thorough inspection of slabs” on the first line, and the second line “Alumina inclusions under the skin” It does not remain. "

Also, the question display column 4 2 2 contains at least the questions associated with the case on the first line.

"Is the Q13 steel grade S40C?" Is displayed. The first line has its display changed (black and white highlighted) and the question has been selected to answer (ie,

"Not answered").

FIG. 7 is a screen showing a state in which a question on the first line or a question on a desired line in the question display column 422 is selected, and a window screen 422 a for receiving an answer is displayed. Now, “Y es” is selected for the force indicating member 4 1 1. In this state, when the user clicks on the force indicating member 4 1 1, the user has answered “Yes” to the selected question.

Fig. 8 is a screen showing an example of answers to a question. The display form of the answered question is different from the unanswered display form. Although not shown in the figure, the displayed lines that have been answered are moved to the bottom line of the question column, making it easier to select and answer unanswered questions. For unanswered questions, the re-scoring process changes the display order of the questions in accordance with the sorting of cases.

FIG. 9 is a display screen showing an example of the inference result. The case “C 6 Cracking during cold working” in the first line of the case display column 4 23 has a score of 100. In this case, in the first search step, the score was 40 as shown in Fig. 6. Similarly, in the first line of the countermeasure display column 424, the score corresponding to the first line of the above-mentioned case, “A5 Thorough inspection of steel slabs” has a score of 100. In addition, as a result of the question being answered to Q13, Q14, and Q15 (Q is the meaning of the question) shown in Fig. 6 and moving to the lower row, in this display example, the next row is displayed. Q26 which was in the top row is displayed.

Fig. 10 shows the display screen when the menu “Case information” is specified, in which information on cases and countermeasures is confirmed. When the menu “Case Information” is instructed, a case number and title column are displayed, and in this column, the “crack during cold working” that was on the first line of the case is extracted as case number “6”. ing.

The content to be displayed may be made instructable sequentially from the first line of the case, or the user may desire (regardless of, but not limited to, a score of 80 or more, which is a predetermined value) regardless of the line position. It is also possible to choose one.

Immediately below the title, the details of the case are read from the case file 51, for example, “A crack occurred during cold forging (screw rolling). The cause is the cause of rolling cracks due to crimping scratches. "At the same time, the answer history to the question is displayed together with the question number, making it possible to confirm the answer to the question, and taking measures to respond to the case" Thorough use of steel slabs "is displayed with measure number 5.

Fig. 11 shows the display screen when the menu “Case Reference Diagram” is specified. A new window is opened to show detailed information on cases and countermeasures in images such as charts. FIG. 12 is a flowchart showing the procedure of the case-based reasoning method. This flow is roughly divided into a search step (steps S1 to S11) as the first stage and an inquiry step (steps S13 to S23) as the second stage.

First, the user manages the data input from the input unit 41 by the user, and determines whether or not the input has been completed by confirming the instruction to the menu “search” or the menu “strict search”, and Judgment is made in step 1 and if there is no instruction, the input permission state is maintained (No in step S1 and step S3). The input contents are displayed in the “detailed explanation (keyword)” column 421 as shown in FIG. For example, if there is an instruction of the menu "search", then the extraction process of the input data is executed by the data extraction unit 11 (step S5). If it is determined that the word is a term or the like and the extraction process is unnecessary, step S5 is skipped.

When the phrase extraction process is completed or skipped for all input data, the case-based inference engine 20 executes the search process for the case including the input ^ as the first search step (step S 7). Next, scoring (score calculation and assignment) is performed on the cases searched in the search process according to a calculation method set in advance according to the degree of inclusion of the word (match degree) (step S9). ), The cases are sorted in descending order of the score, and the result is led to the monitor 42.

2 2, Case display column 4 2 3 and countermeasure display column 4 2 4 are displayed respectively (Step S 1

1).

Next, one or more desired question sentences associated with the case displayed in conjunction with the display of the searched case in the case display column 4 Is selected by the cursor indicating member 4 11 (step S 13). Then, each answer from one or more selected questions, that is, as shown in FIG. 7, either “Yes” or “No” is specified (step S15), and further, When the menu "Response evaluation" is instructed, a question search consisting of re-scoring (step S19), re-scoring processing (step S19), and sorting display in score order (step S21) is determined as the answer. Be executed. Then, it is determined (evaluated) whether or not the highest score has reached the value 80 (step S23), and as shown in FIG. Exit. On the other hand, if the highest score does not reach the value 80 even after re-scoring, it is determined whether the user has performed an answer operation for the next (unanswered) question or entered new data (step S25). If the answer operation has been executed using the one-sol instruction member 4 11 1 and the menu “answer evaluation”, steps S 13 to S 23 are repeated.

Conversely, if the user determines that it is unlikely that a higher score will be obtained even after continuing the interview step, the user can accept input (including deletion) of a new fg ^ J, and in this state, -When the input is completed, return to step S1 and delete unnecessary items including new words or the previous input ^ using the "Clear" menu as appropriate. This flow is continued from step S1. Even before the highest score reaches the value of 80, the user may forcibly end the procedure by the operation of the user and maintain the search result before the end.

Although not shown in Fig. 12, after this flow is completed, the menu "Case Information" is specified and the details of the case are displayed together with countermeasures as shown in Fig. 10, and furthermore, the menu is displayed. By specifying the new “Case Reference Diagram”, a detailed reference diagram of the case can be displayed on the display screen as shown in Fig. 11.

FIG. 13 is an explanatory diagram showing the content of “phrase search” in step S7, and corresponds to the example in FIG. In the figure, for example, assuming that “compressor” and “pump” are extracted by the data extraction unit 11, case numbers 3, 5, and 5. Search for case numbers 5, 7, and 30 that contain 7 and the phrase "pump", and extract about {^}. That is, case numbers 5 and 7 are recognized as cases that include all search groups, and case numbers 3 and 30 are recognized as cases that include only one search term. The former case numbers 5 and 7 are given a score of 40, for example, and the latter case numbers 3 and 30 are given a score of 20 (scored).

FIG. 14 is an explanatory diagram showing the contents of “re-scoring” and “score-sorted display” in steps S 19 and S 21, corresponding to the examples of FIGS. 3 and 13. I have. The rescoring is processed using the case file 51 and the question file 53. In the question file 53, a weight value (match or mismatch) is set for the associated case number as described above for each question number, while the absolute value is also set for cases other than the associated case. The up and down weights are set using the weights (not shown in Fig. 14). Therefore, according to the result of answering the question associated with the case displayed in the case display column 4 2 3, the current score of the case was increased or decreased, and the current score was set for other cases. The current score is moved up or down according to the weight values at the time of up, down or up. Note that cases in which the current score is reduced to a value equal to or less than a predetermined threshold may be deleted in consideration of display capacity and the like. Absence weights for cases that do not have a answered question need not be uniform, and are based on the relationship between the hierarchically structured question and the case, taking into account the similarity of the relevant case or question. It may be systematized and decided.

Then, when the score is changed by the re-scoring, the cases are sorted in the order of the highest score, and the questions are also sorted in the order of the cases, and the result is displayed on the monitor 42.

Next, a second embodiment of the present system will be described. The system configuration in the second embodiment is configured by removing the phrase case index file 52, the phrase case search processing unit 21, and the score link "processing unit 22" from the block diagram shown in FIG.

FIG. 16 is a flowchart showing a second embodiment of the inference system. Steps S31 to S35 are the same as steps S1 to S5 in FIG.

When a word or phrase is input, all question sentences including the word and phrase are read from the question file 53 (question sentence field), and the corresponding case is stored in the case file 51 via the question case index file 54. It is extracted by using it, and the question, case, and countermeasure are displayed correspondingly on the monitor 42 (step S37). At this stage, only the question is extracted, and no scoring process is particularly performed on the case read as the corresponding. For example, a uniform score is set as the initial value. Alternatively, it may be in an unassigned state.

Next, a search process is executed for the question (step S39). In other words, from the displayed question, you can input a search keyword into the display section of the question, which is organized in advance from the displayed question, or the words included in the question sentence. Only the questions that satisfy the search conditions are extracted and displayed in the question display column (step S41), and the questions that do not satisfy the search conditions are deleted. At this time, only the cases and countermeasures corresponding to the questions that satisfy the conditions will be displayed in the respective display columns. Then the top quality Select a desired or desired question (step S43), answer with "Yes", "No", and indicate the menu "evaluation of answer" (steps S45, S47) According to the answer result, as in the first embodiment, re-scoring is performed (step S49), and the cases are sorted in the order of the score, that is, the questions are rearranged in the corresponding form, and the results are displayed on the monitor 42. Is displayed (step S51).

In step S41, each question may be displayed in a manner that can be distinguished according to whether the search condition is satisfied (not shown in FIG. 15 described later). FIG. 15 is a diagram showing an example of a screen after a question search. At a slightly central position of the monitor screen 420, a menu “Question jump” or the like is provided so that only a question that satisfies the search condition can be answered. The menu “Question Jump” designates, in response to a question that satisfies the search conditions only, each time the question is answered and the corresponding line can be answered sequentially (from the top line). As shown in this figure, the selection of a question allows the user to answer a desired question (within a range that satisfies the search conditions).

Subsequently, it is evaluated whether or not the score exceeds a predetermined value 80 (step S53). If the score is exceeded, the flow is terminated as a search end, and if not, the flow returns to step S31. 1§ ^ Data re-entry is resumed.

Steps S61 to S71 are the same as steps S1 to S11 in FIG. 12, and steps S73 to S87 are steps S39 to S5 in FIG. The description is omitted because it is the same as 3. In step S89, when the maximum score of the case does not exceed the value 80 in step S87, the question search in step S73 and the re-input of the data in step S61 are selectively permitted. It is for the purpose.

Next, the configuration and operation of the analysis processing unit 28 shown in FIG. 1 will be described. The analysis processing unit 28 operates by instructing a menu “question analysis” separately prepared in the form of an icon using a specific key or a force indicating member 4 1 1 of the input unit 4 1. An analysis that associates the questions stored in the index file 54 with related cases and reads them (by question number and case number) and performs the process of hierarchically sorting them as shown in Fig. 18 (a). A function section and a tabulation function section for displaying the function section on the screen 420 of the monitor 42 in the format shown in FIG. Question file 53 is shown in Fig. 3 (a), for example, to enable this question analysis. As described above, the data is stored in the question field in a form in which a hierarchical relationship is added between the questions in advance, and each question is included in the hierarchical relationship together with an example included in the question. Process and display. The display format of the relationship between a case and a question may be such that a list of questions included in each case is displayed based on the case, as shown in Fig. 18 (b). In this case, the case file 51 may be used.

Further, the analysis processing unit 28 includes an analysis function unit that displays a question included in the case shown in FIGS. 18 (b) to 18 (c) in an inclusion diagram so that the inclusion relationship can be easily recognized. It also has a tabulation function that is displayed together with the inclusion diagram or individually as an inclusion list as shown in (d) of the figure.

By displaying hierarchical relations and inclusion relations collectively or individually (including non-collection relations) in the form shown in Fig. 18 (a) to (d) on the monitor screen 420, preferable hierarchical relations and inclusion relations Analysis can be performed to obtain the information, and the user can be urged to make appropriate corrections (addition, change, movement, deletion, etc.) Specifically, the correction is to add, change, move, or delete corresponding cases and questions for each file 5 :! In addition, it is possible to grasp or propose problems of the entire case.

Furthermore, the analysis processing unit 28 also enables the following analysis in a mode in which the countermeasure file 55 and the countermeasure case index file 56 are provided as the case base 50. The countermeasure file 55 is required when the countermeasure content is not stored in the case file 51, and stores the countermeasure number and countermeasure content in correspondence. The measure case index file 56 stores the measure number and the case number associated with the measure in a contrasting manner. In this way, by having a file that stores the countermeasure content separately from the case file 51, it is possible to reduce the total storage capacity when multiple cases are associated with one countermeasure. There are advantages. Then, the analysis processing unit 28 performs the same processing as in the case of the above-described questions and the relationship between the questions and the cases with respect to the measures and the relation between the measures and the cases to thereby determine whether there is a preferable hierarchical relationship. It can assist in the analysis of the presence or absence of inclusion relations, and in understanding or proposing problems in the entire case. If the system is equipped with an output means such as a printer, the charts of FIGS. 18 (a) to 18 (d) may be output as a hard copy.

Note that the present invention is also applicable to the following modifications. (1) In the present embodiment, the description has been made by taking as an example the support for quality control abnormalities.However, the present invention is not limited to such a field, and is applicable to all inspection services. Various tasks such as process design work, planning work (production planning, process design, scheduling, inventory planning), failure diagnosis work, material design work, operation, operation work, education, and other highly specialized intellectual work. It can be applied to the support system (expert system). In addition, it can be applied to examinations in lending and leasing operations.

(2) In the present embodiment, as a re-scoring method, weighting coefficients for up and down are provided in the case associated with the question in accordance with the answer result to the question, but only the sign of the same coefficient in both the up and down is inverted. May be set.

(3) In the present embodiment, the answer to the question is either binary as the re-scoring method. However, the multiple-choice method, numerical value (physical quantity, etc.), or even the answer in text It is possible to apply. In these cases, the rescoring method should be adopted in a format that can handle the contents of the answer.

(4) In the first to third embodiments, the termination condition of the inference processing is that the case with the highest score has reached the value of 80, but the present invention is not limited to the value of 80, but the value of 8 It may be set to a value higher than 0 or lower. Also, this value does not need to be fixed, and may be appropriately set according to the situation, or may be set appropriately according to the type of business.

(5) Also, in the first to third embodiments, the inference processing is terminated when the case with the highest score reaches a predetermined value (for example, a value of 80). However, the present invention is not limited to this. The user may be notified that the predetermined value has been reached (or without any special notification), and the end of the inference processing may be left to the user.

(6) In addition to the above-mentioned financing business and leasing business, the present invention can also be applied to various proposals, evaluation of request examples (projects), settlement, and support for judgment at the time of examination.

The aspect described in the present embodiment is a support system in which even a novice can derive a response equivalent to that of an expert for a generated case, but the present invention is not limited to this. It can be applied suitably to clarify the criteria of professional judgment required at the time of examination. In other words, it is common practice to prepare an examination manual, etc. in advance, bind the book, and then conduct an examination or assessment while referring to this book in the actual examination. It is hard to say that it is suitable for all projects in the manual, and there are many cases in which it is no longer used.

Therefore, the following is a description of a method for flexibly obtaining one or more questions from the case to be examined to its countermeasures (conclusion) and systematizing the cases, questions and countermeasures.

FIG. 19 is a diagram corresponding to FIG. 18 and shows an example of a hierarchical structure in the course of systematization. Explaining the relationship between each case and the question, the question extracts factors that are important for the examination (affecting the conclusion) from the terms in the case content, and converts the question format including the factors or the case into another. It is registered in association with the case in the form of a self-assertion sentence that distinguishes it from the case.

In the figure, it is assumed that cases C A1 to C A6 have already been systematized, including their countermeasures. Case CA1 and Case CA2 are registered in association with questions Ql, Q2, Q3, and Q4, Case CA3 is registered in association with questions Q5, Q6, and Q4, and Case C A4 and Case CA5 are questions Q7 and It can be seen that case C A6 is registered in association with questions Q10, Q8, Q9, and Q4. Now, assuming that CA7 is proposed as a new case, keywords (terms) are preferably automatically extracted from the contents (texts, etc.) of the input case CA7. Since the existing questions Q1 to Q9 are displayed on the screen 420 of the monitor 42, an attempt is made to derive a conclusion by answering a desired one from among them, and the suitability is examined from the viewpoint of an expert. In Figure 19, it can be seen that Case CA7 did not draw appropriate conclusions from the existing questions Q1 to Q9, and required additional questions Q11 and Q12 in addition to questions Q9 and Q4.

In this way, when judging a new case, we first performed a search within the existing questions, and when we judged that answering the required questions gave us a conclusion that was considered appropriate, we adopted it. If a question is registered in connection with this case, but it cannot be said that a proper conclusion has been obtained, one or more new questions may be added to determine whether a proper conclusion has been obtained. Judgment is made, and if obtained, additional questions are also linked and registered. Do this until you have determined that you have reached the appropriate conclusion and create new questions. If necessary, the systematization can be checked by printing out in the form shown in Fig. 19, etc., or by making appropriate corrections, etc., so that a more suitable systematization can be realized. . Follow For example, when the result of a case is approved by the boss, by presenting the contents of this question search and similar cases together, it becomes more objective and easy to judge the approval.

In the system shown in Fig. 19, the relationship between the case 1 and each of the other cases is numerically expressed as the degree of coincidence (degree of approximation) according to the number of common questions. The registration is made in association with the registration file. For example, case CA1 and case CA2 share all of questions Q1 to Q4, and if this is a degree of coincidence of "4", case CA1 and case CA3 share only question Q4 Therefore, the matching degree may be set to "1". Alternatively, when numerically calculating (100 x number of common questions / total number of questions), the former should be set to 100 (= 100 x 8/8), and the latter should be set to 2 9 (= 100 x 2/7), or a numerical value may be set based on the branching distance at which stage in the hierarchy the branch was taken. In any case, if each case is numerically represented as a degree of coincidence with each other, similar cases can be searched for, and the versatility is high. The above quantification can be calculated by automatic calculation if the above-mentioned various calculation formulas (or tables recording the calculation results) are provided in advance, so that the degree of coincidence follows as questions are added. The advantage is that it can be changed dynamically.

Hereinafter, an example of a question using FIG. 19 will be described by taking a loan operation as an example.

Case No. 1: Title “The credit limit of 00 Co., Ltd. is 00 million yen”

"Established: Showa 00 0 February 0 Capital 00 million yen

Industry: 00 Manufacturing

Debt excess, sales channels stable. It is in the capital business of 00 Shoji with handling of 00 Shoji 00%. As a financial improvement, the company sold its headquarters building to reduce debt.

The extraordinary loss of unrealized loss due to the sale of the head office real estate is ¥ 0 billion, which is in excess of debt. "

Is entered and displayed on the monitor 42, and the terms “excess debt” and “sales channel” are extracted from the above contents, and questions related to this and common questions (for example, in FIG. 19, question Q 4 Or questions included in the same fork, etc.) are displayed on the monitor 42. Further, a question sentence including the extracted term is extracted and displayed. Here, the following three questions are selected and answered.

"Question questions Question number 1: "Is debt overrun?"

Question category: Y e s / N o

Question number 2: "Is ordinary profit in the black?"

Question category: Y e s / N o

Question number 3: "Is the sales channel stable?"

Question classification: Y e s / N o "

When you finish answering the above question, the score is calculated and the result is displayed.

"Countermeasures for cases

Countermeasure number 1: Title "Credit amount 00 million yen"

This indicates that the result of this question search showed that the result of the loan 0 K was obtained.

In the above, if the answer to each question is Yes, the score is calculated as a predetermined high value, for example, 10 points are added as relative points, and if No, the value is lower than the high value of the above H if constant. For example, two points are added as relative points. Note that a positive value (for example, 10 points) may be added by Y es and a negative value (for example, -2 points) may be added by No. Then, scoring is performed in consideration of the relation with the relative scores of other questions. The scoring may be calculated as 100 out of 100 points by multiplying the ratio of the total score of the relative points of Y es in each question to the actually acquired score by the value 100. If the score obtained by answering the required question exceeds a predetermined value, a positive result is obtained as described above. Alternatively, the operator may judge from the score.

As described above, according to the inventions described in claims 1, 12, and 17, inference is performed from different viewpoints of word search and question answering by questioning, so that a case with higher certainty can be extracted. And provide a highly workable and versatile system.

In the case-based reasoning method described in claim 2, the notification is displayed on display means. Further, in a readable recording medium having a program for executing the case-based inference described in claim 18, the notification is displayed on a display means. As a result, a large amount of information can be recognized at the same time, and the user's operation work is easy and efficient.

The case-based reasoning method described in claim 3 is that measures corresponding to each searched case are also reported. Further, the case-based inference system according to claim 15 is characterized in that: The display control means displays the countermeasure corresponding to each case on the display means. Further, the readable recording medium incorporating the program for executing the case-based reasoning described in claim 19 is configured to notify a countermeasure corresponding to each searched case together. As a result, not only high-probability cases can be recommended, but also direct measures as actions to be taken are presented, and the function as support is effectively demonstrated.

In the case-based reasoning method described in claim 4, words for search are extracted from sentences input from input means. A readable recording medium incorporating a program for executing the case-based reasoning described in claim 20 is configured to extract a query for search from a sentence input from input means. As a result, it is not necessary to be aware of the input terms individually, and the burden on the user is reduced.

The case-based reasoning method according to claim 5, wherein the question can be selectively answered. A readable recording medium incorporating a program for executing the case-based inference according to claim 21 can selectively answer the question. As a result, the degree of freedom to appropriately answer the questions expected by the user is increased.

In the case-based reasoning method described in claim 6, the display of the cases is rearranged in the order of the score every time the re-scoring is performed. Further, the readable recording medium incorporating the program for executing the case-based reasoning described in claim 22 is arranged so that the display of the cases is rearranged in a scoring order every time rescoring is performed. These enable inference work without overlooking highly probable cases. Further, in a display mode in which the display means has a small number of display lines and needs to be scroll-displayed, the highest score is always displayed, which is very useful.

In the case-based reasoning method described in claim 7, the questions associated with the displayed case are displayed in correspondence with the display order of the cases. Also, the readable recording medium incorporating the program for executing the case-based reasoning described in claim 23 is configured to display the questions associated with the displayed case in the order in which the cases are displayed. is there. As a result, the questions are displayed in the order corresponding to the case, so that the questions can be displayed more accurately.

The case-based inference method according to claim 8, wherein the highest score is a predetermined value by re-scoring. Is reached, the inference processing ends. Further, the readable recording medium incorporating the program for executing the example-based inference described in claim 24 is configured to terminate the inference processing when the highest score reaches a predetermined value by re-scoring. It was done. These enable faster inference work.

A case-based reasoning method according to claim 9, wherein the case and a question associated with the case are tabulated and displayed in a form that can be analyzed. A readable recording medium incorporating a program for executing the case-based reasoning according to claim 25, wherein the case and a question associated with the case are tabulated and displayed in a form that can be analyzed. It is. From these, the relationship between questions and cases and questions is analyzed, and the validity of the questions included in the case (expression method, number, order, etc.), and a study to obtain hints for improvements to the field work from the entire case. Work becomes easy.

The case-based reasoning method described in claim 10 searches for and displays a question including at least one or more input words in a question file, and displays, from the case file, an example associated with the displayed question. It includes a step of reading and displaying, and a step of executing re-scoring on each of the displayed cases according to the result of answering the displayed question.

In addition, the readable recording medium having a program for executing the case-based reasoning described in claim 26 searches and displays at least a question including at least one input phrase in a question file, and displays the question. A case associated with the displayed question is read from the case file and displayed, and re-scoring is executed for each of the cases being redisplayed in accordance with the result of answer to the displayed question.

According to these, the questions including the entered are displayed together with the examples associated with these questions, and the user can answer these displayed questions. The likelihood of reaching a more reliable case more quickly, such as when the user is familiar with the field, is improved.

In the case-based inference method according to claim 11, prior to the step of executing the re-scoring, a search is performed by setting a condition for the displayed question, and a question satisfying the search condition is obtained. It is provided with a step of re-displaying with a related case. The case-based inference system according to claim 14, wherein a condition is set for the question displayed by the first display control means, a search is performed, and a question satisfying the search condition is associated. It also has a means for redisplay. In addition, the readable recording medium incorporating the program for executing the case pace inference described in claim 27 allows the displayed question to be set and searched before executing the re-coring, The question that satisfies the search condition is redisplayed together with the related case. As a result, when the number of questions is large, questions can be narrowed down under appropriate desired conditions, so that operational efficiency and workability are improved. The case-based inference method according to claim 16, wherein the degree of coincidence between each case and each other case is set based on the hierarchical situation of the case and the question. According to this, a case similar to a certain case can be easily extracted, which is extremely useful for referring to the content of a similar case.

Industrial applicability

As described above, the present invention provides quality management, inspection services in general, sales services, estimation design services, process design services, planning services (production planning, process design, scheduling, inventory planning), failure diagnosis services, and material design services. It is useful as a support system (expert system) for obtaining professional judgment in various tasks, such as general operations, operations, education, and other highly specialized intellectual services. It is also suitable for screening in leasing operations.

Claims

The scope of the claims

1. At least one or more input words are searched for a case with the case file, and according to the degree of matching with the word, scoring is performed on the searched cases, and Notifying the result and reading the question associated with the searched case from the question file, and performing re-scoring on each of the searched cases according to the answer result to the question A case-based reasoning method characterized by having the following.

2. The case-based reasoning method according to claim 1, wherein the notification is displayed on display means.

3. The case-based reasoning method according to claim 1 or 2, wherein a countermeasure corresponding to each searched case is reported together.

4. The case-based reasoning method according to claim 1 or 2, wherein the word for search is extracted from a sentence input from an input means.

5. The case-based reasoning method according to claim 1 or 2, wherein the question can be selectively answered.

6. The case-based reasoning method according to claim 2, wherein the display of the cases is rearranged in the order of the score every time the re-scoring is performed.

7. The case-based reasoning method according to claim 6, wherein the questions associated with the displayed cases are displayed in correspondence with the display order of the cases.

8. The case-based inference method according to claim 1, wherein the inference processing is terminated when the highest score reaches a predetermined value by re-scoring.

9. The case-based reasoning method according to any one of claims 2 to 8, wherein the case and a question associated with the case are tabulated and displayed in a form that can be analyzed.

10. Searching and displaying a question containing at least one or more input words in the question file, reading the case associated with the displayed question from the case file, and displaying the question; And performing a re-scoring on each of the displayed cases according to a response result to the case-based inference method.

1 1. Prior to the step of executing the re-scoring, a step is provided for performing a search on the displayed question by setting conditions and re-displaying the question that satisfies the search condition together with related examples. 10. The case-based reasoning method according to claim 10, wherein:

1 2. A case file, a question file, and a file in which a question and a case are associated with each other. A display means, a case search means for performing a case search between the input ^ J and a case file, and a search. Scoring means for scoring each searched case in accordance with the degree of matching with the word, and first display control means for outputting the searched case and the score in correspondence to the display means. Second display control means for outputting a question associated with the searched case to the display means, and re-scoring for re-scoring a score for each searched case according to an answer to the displayed question. Case-based reasoning system characterized by having

13 3. Include a case file, a question file, and a file that associates the question with the case, display means, a question search method that searches the question including the input ^ J in the question file, and a Means for reading an associated case from the case file; first display control means for outputting the searched question to the display means together with the read case; and displaying the question in response to a result of answer to the displayed question. Means for rescoring each case.

1 4. means for performing a search by setting a condition for the question displayed by the first display control means, and redisplaying the question satisfying the search condition together with the associated case. 13. The case-based reasoning system according to claim 13.

15. The case-based inference system according to any one of claims 12 to 14, wherein the first display control means displays a countermeasure corresponding to each case on the display means. Am.

16. The case according to any one of claims 1 to 15, wherein the degree of coincidence between each case and each of the other cases is set based on the hierarchical situation of the case and the question. Case-based reasoning system.

1 7. At least one or more input words are searched for a case with the case file, and according to the degree of fg ^ I, scoring is performed on each searched case. Notify the results and ask questions related to the searched cases from the question file. A readable recording medium incorporating a program for executing case-based reasoning for reading out and informing and performing re-scoring on each of the searched cases according to the answer to the question

18. A readable word 3 ^ medium having a program for executing the case-based reasoning according to claim 17, wherein the notification is displayed on a display means.

19. A readable recording medium incorporating a program for executing the case-based reasoning according to claim 17 or 18, wherein a measure corresponding to each searched case is reported together.

20. A readable record with a built-in program for executing case-based reasoning according to claim 17 or 18, wherein a word for search is extracted from a sentence inputted from an input means. Medium.

21. A readable recording medium incorporating a program for executing case-based reasoning according to claim 17 or 18, wherein the question can be selectively answered.

22. A readable recording medium incorporating a program for executing case-based inference according to claim 18, characterized in that the display of cases is sorted in the order of scores each time rescoring is performed.

23. A program for executing the case-based reasoning described in claim 22, wherein the questions associated with the displayed cases are displayed in correspondence with the order of displaying the cases. A readable recording medium.

23. The readable recording medium according to claim 22, wherein the inference processing is terminated when the highest score reaches a predetermined value by the re-scoring.

25. The case-based inference according to any one of claims 18 to 24, wherein the case and the question associated with the case are tabulated and displayed in a form that can be analyzed. A readable recording medium with a built-in program.

2 6. Search and display questions containing at least one or more entered words in the question file, read the case associated with the displayed question from the case file, display it, and answer the displayed question. A readable recording medium containing a program for executing a case-based reasoning for executing a re-scoring on each of the cases displayed according to a result.

2 7. Set the conditions for the displayed question before performing re-scoring. 26. The readable recording medium according to claim 25, further comprising: causing a search to be performed to re-display a question satisfying the search condition together with a related case.