JP2913969B2 - Case search system and search method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an inference system for obtaining a solution to an input problem from case information stored in a database in advance, and searches for a case corresponding to the input problem as a solution candidate. The present invention relates to a case search system, and particularly to a case search system and a search method suitable for efficiently developing and maintaining a system.

[0002]

2. Description of the Related Art In an inference system and a case retrieval system for finding a solution to an input problem from case information stored in a database, it is important how to retrieve case information similar to an input problem. Here, the case information is data composed of data representing a problem and data representing a solution or solution to the problem. One purpose of using case information is that the case search system searches for a solution case (similar example) of a problem similar to the problem currently being solved, and the user refers to the answer or solution there. is there. Another purpose of use is that the inference system further uses the similar example searched by the case search system to correct the similar example and find a solution to the current problem. As described above, a technical problem in using the case information is to develop a system that efficiently searches for a similar example to the current problem from a plurality of accumulated cases.

[0003] Conventional search techniques of similar examples include, for example,
"American Association for Artificial Intelligence Proceedings" (Procee
dings of the American Asso
citation for Artificial In
tellingence (1986) at pages 267-271. There, a search is made for a case example of a dish similar to the current dish with respect to the planning of the dish. For this purpose, the input problem and the case problem are represented by a set of attributes such as a goal to be achieved, constraints to be observed, and a situation at the time of solving the problem. For example, the problem is represented by attributes such as a cooking method to be adopted, ingredients to be used, and mistakes to be avoided. Then, a similar example is selected using the following similarity definition. It should be noted that the magnitude of the similarity used for selecting a similar example may be represented by a distance, and in such a case, the one having a small distance is regarded as having a large similarity. (A) Definition of similarity for each attribute: First, using the attribute hierarchy, the same attribute, then two attributes (equivalent to siblings) whose upper-level attribute is equal, and two-level higher Are defined as having smaller distances in the order of two attributes having the same attribute (i.e., cousin attributes), that is, having a higher similarity. (B) Definition of importance of attributes: Indicates the order of importance between attributes. (C) Definition of similar example: The case most similar to the input problem is a case including the most similar problem with respect to the most important attribute. If there are multiple such cases,
Among them, the case containing the most similar problem with the next most important attribute is the most similar case. Further, when there are a plurality of such cases, the comparison is made on the next most important attribute. When the same comparison is continued until the number of cases becomes one, the last remaining case is the most similar case.

[0004] However, such a similar example definition is not always optimal in all applications other than cooking. For example, consider a problem of disposing devices such as a central processing unit and a console display in a computer room. In such a case, as a similar example, an arrangement example of similar devices in a similar room is searched. Then, for example, the problem is represented by the vertical and horizontal sides of the room and the names of the individual devices to be arranged. In the above prior art, first, it is necessary to define which of the vertical side and the horizontal side of the room is important. Suppose now that "vertical side is more important".
Then, as compared with the input problem, the vertical side is equally horizontal side is entirely different case (P _1), the vertical side is also the transverse edges are also intended to slightly have different cases (P _2), from (c) above , Case (P ₁ ) has similarity, that is, a case with a high degree of similarity. This is, in fact, an inappropriate choice. However, conversely, the same inconvenience arises even if "horizontal side is more important". That is, a case (P ₃ ) in which the horizontal sides are equal but the vertical sides are completely different is given priority over a case (P ₂ ) in which the horizontal and vertical sides are slightly different.

A conventional system for searching for similar examples,
That is, in the case search system, a general-purpose part and a part depending on each field are not clearly separated in the system. For this purpose, different systems are constructed for each field.

As described above, the definition of similarity in the conventional case retrieval system lacks versatility, and furthermore, the general-purpose part and the field-dependent part in the system have not been clearly separated. As a result, when developing a case search system, it is necessary to rebuild the system for each field,
There has been a problem that the burden of system development costs, personnel, and time is large. In addition, there is also a problem that the cost, personnel, period, and the like are heavy when maintaining and optimizing the similar example search function of the existing system.

[0007]

The problem to be solved is that in the conventional case retrieval system, a general-purpose part and a field-dependent part are not clearly separated, and a general-purpose part is used for each field. It is necessary to develop a system including An object of the present invention is to solve these problems of the prior art, and separate a case search procedure into a general part common to all case search operations and a different part for each field. It is an object of the present invention to provide a case search system and a search method thereof that enable case searches corresponding to respective fields only by replacement, and that can greatly reduce the load related to system development and maintenance.

[0008]

In order to achieve the above object, the case search system according to the present invention uses a general framework to describe the contents of the similarity definition for each field in a conventional similar example search. Focus on what can be achieved by doing
(1) A case base storing at least one case information composed of data representing a problem and data representing a solution to the problem in advance, and a case including a problem similar to a problem input via an input device provided. In a case search system having a case search unit for searching information from a case base based on a procedure stored in a storage unit provided in advance, a case stored in the case base in the case search procedure stored in the storage unit A field-dependent procedure storage unit that stores a search procedure for each field that depends on the field of information is provided, and the storage unit stores only a general-purpose search procedure that does not depend on the field of case information that is stored in the case base. Is changed, only the search procedure for each field stored in the field-dependent procedure storage section is changed. (2) In the case search system according to the above (1), the search procedure for each field stored in the field-dependent procedure storage unit is performed by inputting a problem of each case information stored in the case base by the case search unit. It is characterized by comprising a similarity calculation procedure for calculating a similarity to a problem. (3) In the case search system according to any one of the above (1) and (2),
In response to this input problem, a case search unit is provided which makes an inference based on the case information retrieved using the field-dependent procedure storage unit and obtains a solution to the input problem. (4) The case search method of the case search system according to any one of (1) to (3), wherein the input problem and the problem of the case information stored in the case base are classified into a set of attributes or , And includes at least one of a set of attribute pairs, which are pairs of attributes and values of the attributes. The search procedure for each field for the input problem stored in the field-dependent procedure storage unit is a similarity for each attribute, or The case search unit includes at least one of a procedure for calculating a similarity for each attribute and a procedure for calculating a similarity for each attribute pair. The similarity calculation for each part of the input problem and the problem of the case information is performed using the calculation procedure of (1). (5) In the case search method of the case search system according to (4), the similarity of each part between the input problem calculated by the case search unit and the problem of the case information is combined, and the case information for the input problem is combined. The overall similarity calculation procedure for calculating the overall similarity of the problem is stored in the field-dependent procedure storage unit, and the case search unit uses the overall similarity calculation procedure to determine the input problem and the case information problem. Is characterized in that a similarity as a whole is obtained, and a case from a case base for the input problem is selected based on the obtained similarity.

[0009]

In the present invention, the case retrieval section calculates the similarity of each of the input problem and the problem of the case information for each attribute or attribute pair by using the partial similarity calculation procedure, and executes the overall similarity calculation procedure. Then, the similarity of the input problem and the problem of the case as a whole is calculated from the attribute or the similarity of each attribute pair calculated as described above, and the case is searched based on the overall similarity. As described above, the procedure relating to the case retrieval process is divided into a general part, that is, a procedure relating to the operation of the case retrieval unit, and a field-dependent part, namely,
By separating the partial similarity calculation procedure and the overall similarity calculation procedure stored in the field-dependent procedure storage unit into the partial similarity calculation procedure and the overall similarity calculation procedure, the case search unit can be described. A case search system suitable for each field can be constructed by using a general-purpose search procedure relating to the operation of (1).

[0010]

Embodiments of the present invention will be described below in detail with reference to the drawings. FIG. 1 is a block diagram showing a first embodiment of a configuration according to the present invention of a case search system according to the present invention. The case search system according to the present embodiment searches for a plurality of cases similar to an input problem, and outputs a solution to the input problem based on the plurality of cases. Input device 1, a display device 2 for displaying and outputting a processing result of the system, a case searching device 3 for performing a case searching process and a solution determining process for an input problem according to the present invention, case information, etc. Is stored in the storage device 4 for storing the information. The input device 1 inputs case information, a partial similarity calculation procedure and an overall similarity calculation procedure according to the present invention, and a problem, and passes them to the case search device 3. In addition, the output device 2 outputs
Receives and outputs answers to questions. Case search device 3
Is a case search processing unit 5 for performing case search processing of the present invention, an input / output processing unit 6 for performing data input / output processing with an external device such as the input device 1, and a target search operation using a plurality of searched cases. The case use processing unit 7 that determines a solution, and controls the operations of the case search processing unit 5, the input / output processing unit 6, and the case use processing unit 7 to control the case search device 3.
Is composed of a control processing unit 8 for controlling the overall operation of. In addition, the storage device 4 stores in advance a case base 9 storing case information including a problem and its solution, a partial similarity calculation procedure storage unit 10 according to the present invention, and an overall similarity calculation procedure storage unit 11. Become.

The case search system according to the present embodiment is configured by focusing on the fact that conventional similar example searches can be realized by describing the contents of similarity definitions for each field in a general framework. I have. For example, conventional technology,
IPSJ Transactions on Volumes 32, 5 (1991), 6
A similar example search technique in the computer room layout problem discussed on pages 26 to 634, and a case-based processing workshop, Proceedings (1988) (Cas
e-based Reasoning Worksho
p (1988)) on page 21 to page 30 and a similar example search technique in international dispute mediation, as well as the IPSJ National Convention, Vol. 2 (1991), Vol.
Similar example search technology in Fortran program optimization discussed on pages 9 to 62, and 213 of IPSJ Annual Conference, Vol. 2 (1991).
The search for similar examples in the project deadline forecasting discussed on pages 214 through 214, all of which are as follows:
In the general framework of (C), this can be realized by describing the contents of the similarity definition for each field. When the similarity is represented by a distance, the one having a small distance has the large similarity. (A) Expression of input problem and case problem A problem is represented by one or both of a set of attributes and a set of pairs of attributes and their values. (Note that the value of an attribute is referred to as an attribute value, and a pair of an attribute and an attribute value is referred to as an attribute pair.) (B) Definition of Similarity This is represented by the following two types of definitions. (B-1) Partial Similarity Calculation Procedure A procedure for calculating a similarity between two attributes or two attribute pairs, or a definition of the similarity. (B-2) Overall Similarity Calculation Procedure A procedure for calculating the overall similarity from the attributes of the input problem and the case problem, and the similarity of each attribute pair. (C) Similarity Example Retrieval Process (C-1) The attribute between the input problem and the case problem is calculated using the partial similarity calculation procedure, and the similarity for each attribute pair is obtained. (C-2) Using the overall similarity calculation procedure, the overall similarity between the input problem and the case problem is determined from the similarity for each attribute and each attribute pair. (C-3) A case is selected based on the overall similarity described above.

Hereinafter, a specific system configuration and an example of operation will be described using the general framework including (A) to (C). First, pages 267 to 271 of the American Association for Artificial Intelligence Transactions (1986) cited as the prior art apply to the general framework consisting of the above (A) to (C). Show. The similarity definition is described as follows. (B-1) Partial similarity calculation procedure Distances (similarity) between the same attribute are set as “0”, siblings are set to “1”, cousins are set to “2”, etc., using the attribute hierarchy
Is defined. (B-2) Overall Similarity Calculation Procedure N is the maximum value of the distance between the above attributes. In addition, this N
Can be easily determined from the attribute classification hierarchy. Here, the weight of the least important attribute is “1”, the weight of the next least important attribute is “N + 1”, the weight of the next least important attribute is “(N + 1) ² ”, and so on. Assign weights to attributes. Then, the overall similarity is defined as the sum of the weights of the similarity between the attributes. By the above (B-1), a distance equivalent to the definition of the similarity between the attributes described in (a) of the related art can be defined. Also, since the weight of the attribute of (B-2) is larger than the maximum value of the distance between the attributes, one of the two cases “C1” and “C2” is
If the attribute is more similar to the input question,
Even though "C2" is more similar for all less important attributes, "C1" has priority. Thus, similar examples are selected as in (b) and (c) described in the related art.

Next, a similar example search in the computer room layout problem discussed on pages 626 to 634 of the above-mentioned Transactions of the Information Processing Society of Japan, Vol. 32, No. 5, (1991),
The above framework applies. In this case, "room vertical, 1
An attribute pair (attribute = vertical side of the room and its value = 10) such as “0”, “room side, 15”, and CP
The problem is represented by an attribute such as U or CD indicating the device to be arranged. Then, the similarity definition is described as follows. (B-1) Partial Similarity Calculation Procedure The similarity between attributes representing devices is based on the classification hierarchy described in the related art. For example, the distance between the same devices is defined as “0”, and the distance between sibling devices is defined as “1”. The distance between attribute values is the absolute value of the difference between numerical values. (B-2) Overall Similarity Calculation Procedure The sum of the distance weights for each attribute and for each attribute value. By properly adjusting the weight, the problem described in the prior art can be solved. For example, by equalizing the vertical and horizontal weights of a room, cases where the vertical and horizontal sides are slightly different from the input problem, cases where the vertical sides are completely different and the horizontal sides are completely different, or where the horizontal sides are equally vertical Edges can be given priority over any of the completely different cases.

As described above, focusing on the fact that a conventional case search can be realized by describing the contents of the similarity definition for each field in a general framework, the case search system of the present embodiment The problem and the problem of the case are represented by a set of attributes or a set of attribute pairs consisting of attributes and their values, and a procedure for calculating the similarity of each attribute or attribute pair (partial similarity calculation procedure) is input. A procedure (overall similarity calculation procedure) for calculating the overall similarity of the input problem and the case problem from the attribute and the similarity of each attribute pair is input and stored. As described above, the case search procedure is changed from a general procedure related to case search such as the control processing unit 8 or the case search processing unit 5 to a procedure depending on the case base 9 (a partial similarity calculation procedure and an overall similarity calculation procedure). ), The case search system of this embodiment can easily perform system development and maintenance for each field.

The operation of each processing unit in the case search device 3 will be described below. The problem input from the input device 1 is represented by at least one of a set of attributes and a set of attribute pairs. The cases stored in the case base 9 are composed of questions and their answers. The partial similarity calculation procedure stored in the partial similarity calculation procedure storage unit 10 is a procedure for calculating the attributes of the input problem and the case or the distance (similarity) for each attribute pair. The overall similarity calculation procedure stored in the similarity calculation procedure storage unit 11 is a procedure for obtaining the distance (similarity) of the input problem and the case as a whole from the distance of each attribute or attribute pair. The partial similarity calculation procedure and the overall similarity calculation procedure are procedures and reference data described in a programming language, and by execution thereof, an attribute, an attribute pair and a distance as a whole are calculated.

The input / output processing unit 6 inputs the case information, the partial similarity calculating procedure, and the overall similarity calculating procedure from the input device 1, and stores the case base 9 and the partial similarity calculating procedure in the storage device 4, respectively. Unit 10, overall similarity calculation procedure storage unit 11
To be stored. Also, input the question from the input device 1,
The answer is passed to the control processing unit 8, and the answer to the question is received from the control processing unit 8 and output to the display device 2. The control processing unit 8 receives the input problem from the input / output processing unit 6, passes it to the case search processing unit 5, and receives a similar example of the above-described problem from the case search processing unit 5. Then, the input problem and the similar example are transferred to the case use processing unit 7, the answer to the input problem is received from the case use processing unit 7, and the answer is transferred to the input / output processing unit 6. The case search processing unit 5 receives the input problem from the control processing unit 8, takes out the cases one by one from the case base 9, performs the following processing for each case, and calculates the distance from the received input problem. Then, the case where the distance becomes the minimum is passed to the control processing. That is, by executing the partial similarity calculation procedure stored in the partial similarity calculation procedure storage unit 10, the distance for each attribute of the input problem and the case problem is calculated, and the overall similarity calculation procedure storage unit 11 is calculated. Then, the overall distance between the input problem and the case problem is calculated from the similarity for each attribute by executing the overall similarity calculation procedure stored in the. The case use processing unit 7 receives the input problem and the similar example from the control processing unit 8, obtains a solution to the input problem based on the solution of the similar example, and returns the solution to the control processing unit 8. This processing is performed, for example, by using the American Association for Artificial Intelligence.
Transactions (1986), pp. 267-271 (P
rosededs of the American
Association for Artifici
al Intelligence (1986)). In the storage device 4, the case base 9 receives the case information from the input / output processing unit 6, stores the case information, and passes the case information to the same process in response to the reading of the case search processing unit 5. The partial similarity calculation procedure storage unit 10 receives the partial similarity calculation procedure from the input / output processing unit 6, stores the procedure, and passes the procedure to the same processing in response to the reading by the case search processing unit 5. The overall similarity calculation procedure storage unit 11 receives the overall similarity calculation procedure from the input / output processing unit 6, stores the procedure, and passes it to the same processing in response to reading by the case search processing unit 5. As described above, in the case search system according to the present embodiment, the procedure (partial similarity calculation procedure and overall similarity calculation procedure) depending on the case base 9 is separated from the general-purpose search procedure used for the operation of each processing unit. .

Hereinafter, the operation of the case search processing unit 5 according to the present invention will be described in detail with reference to FIG. FIG.
3 is a flowchart showing one embodiment of a processing operation according to the present invention of the case search processing unit in FIG. 1. First, the partial similarity calculation procedure and the overall similarity calculation procedure are read from the partial similarity calculation procedure storage unit 10 and the overall similarity calculation procedure storage unit 11 of FIG. 1, and are converted into an executable form such as a machine language. Are stored in the main memory (step 101).
Then, the first case of the case base 9 in FIG.
A candidate for a similar example is set (step 102). Next, by executing the partial similarity calculation procedure, the distance between the extracted candidate of the similar example, the attribute of the input question, and the attribute pair is calculated (step 103). Then, by executing the overall distance calculation procedure, the distance between the candidate and the input question as a whole is calculated (step 104). It is determined whether or not the processing has been performed on all the cases in the case base 9 in FIG. 1 (step 105). When all the cases have been processed, the remaining candidates are regarded as similar examples and the control of FIG. Process 302
(Step 106). If not,
The following processing is performed. The next case is taken out from the case base 9 of FIG. 1 and is taken as the next candidate (step 107), and a partial similarity calculation procedure is executed to calculate the distance between the next candidate and the attribute of the input question (step 108). Furthermore, the overall similarity calculation procedure is executed to calculate the distance between the next candidate and the input question as a whole (step 109). It is determined whether the total distance between the next candidate and the input question calculated in this way is smaller than the total distance between the previously calculated candidate and the input question (step 110). If it is not smaller, the process returns to step 105. If it is smaller, the next candidate is set as a new candidate (step 111), and the process returns to step 105. In this way, the case search process is performed.

Next, an example of executing a similar example search by the present search processing method will be described with reference to FIGS. As an example, consider a problem of arranging devices in a computer room. The computer room has a rectangular shape and one entrance. In addition, the side with the entrance and the side opposite to it will be called a horizontal side, and the side next to it will be called a vertical side.

FIG. 3 is an explanatory diagram showing an embodiment of the case data stored in the case base in FIG. In the figure, 31 to 33 are case data each representing one case. Here, for the sake of simplicity, the case of three cases is considered, but in a practical system, many cases are used. In addition, the case data 31 to
33 is described in a table format for easy understanding, but is actually described using a programming language such as C language or a knowledge expression language. Case data 31-33
Is the problem of each case, and the right side is the solution corresponding to the problem. For example, the left side of the case data 31 is
In a room where the vertical side is 10 m, the horizontal side is 15 m, and the entrance is 9 m from the end of the horizontal side, a CPU (Central P
processing Unit, central processing unit), CD
(Console Display, console display), MDU (Magnetic DiskUni)
t, magnetic disk drive), MTU (Magnetic)
TapeUnit, magnetic tape unit), CR (Card)
Reader, card reader). The right side is an arrangement result for each device, and is represented by coordinates and directions.

FIG. 4 is an explanatory diagram showing an embodiment of the partial similarity calculating procedure stored in the partial similarity calculating procedure storage section of FIG. In this figure, reference numeral 41 denotes an attribute-similarity calculation procedure representing a similarity calculation procedure for each attribute pair (attribute and its value). For each attribute pair relating to a room, such as the problem (vertical side 11) in FIG. Defines the distance. Reference numeral 42 denotes an attribute value definition that defines a distance between attribute values, that is, numerical values representing a room size in the attribute-similarity calculation procedure 41. Specifically, the absolute value of the difference between the numerical values is set as the distance. 43
Is an attribute-to-distance definition that defines the distance between the input problem and the case for the attribute pair "e". Reference numeral 44 denotes an attribute similarity calculation procedure representing a similarity calculation procedure for each attribute, and defines a distance between an input problem and a case for each device such as a CPU. The attribute similarity calculation procedure 44 uses data relating to the device classification hierarchy shown in FIG. 5 described later. Reference numeral 45 denotes a hierarchy definition that defines an upper node of the device “e”. This hierarchy definition 45
, The upper node of the device “e” is defined as “e” itself or its parent, grandfather,. 46
Is an attribute distance definition that defines the distance between the input problem and the case problem for the device “e”. Here, among the nodes that are the upper nodes of the device “e” and the upper nodes of the device in question, the lowest node is “u”, and the level difference between “e” and “u” is , “E”. Note that the partial similarity calculation procedure, the data described below, and the overall similarity calculation procedure are expressed in Japanese and figures for easy understanding of the description.
It is described in a programming language such as C language or a knowledge expression language.

FIG. 5 is an explanatory diagram showing an embodiment of an attribute classification hierarchy used in the attribute similarity calculation procedure in FIG. An example of an attribute, that is, an example of a device classification hierarchy is shown. Below the device 51 as the highest layer, there are a computing device 52, an input / output device 53, a storage device 54, and an operation device 55. The input / output device 53 has an input device 57 and an output device 58, and the storage device 54 has an MDU 59 and M
The TU 60, the operating device 55 includes a CD 61, and the input device 57 includes an FDD (Flexible Disk D).
live Unit) 62 and CR 63, and LP (Line Printer) for output device 58
64 and LBP (Laser Beam Print)
r, laser beam printer) 65. In this classification hierarchy, for example, an upper node of the FDD 62
DD 62 itself, input device 57, input / output device 53, device 5
1 is represented.

FIG. 6 is an explanatory diagram showing an embodiment of the overall similarity calculation procedure stored in the overall similarity calculation procedure storage unit in FIG. Overall similarity calculation procedure 66 of the present embodiment
1 requires data on attribute weights, and the case search processing unit 5 in FIG. 1 using the overall similarity calculation procedure 66 performs a case search while referring to the attribute weights 67. In this embodiment, the weight of the attribute, that is, the weight of the vertical side is 2
It represents that.

FIG. 7 is an explanatory diagram showing an embodiment of a problem input to the case search system in FIG. The input problem 71 of the present embodiment is as follows: the vertical side is 10 m, the horizontal side is 15 m,
This represents a problem of arranging a CPU, a CD, an MDU, an MTU, an FDD, and an LP in a room where an entrance is located 10 m from an end of the side. Hereinafter, for this input problem 71,
The case search system of FIG. 1 performs the processing operation shown in FIG. 2 using the similarity calculation procedure and data of FIGS.
The operation of searching for a similar example of interest from the cases in FIG. 3 will be described.

The input / output processing unit 6 shown in FIG. 1 inputs an input question 71 shown in FIG. 7 and passes it to the control processing unit 8 shown in FIG. The control processing unit 8 in FIG. 1 passes this input problem to the case search processing unit 5 in FIG. The case search processing unit 5 of FIG. 1 searches for a similar example from the case base 9 of FIG. 1 by the following processing. First, based on the processing of step 101 in FIG. 2, the partial similarity calculation procedure shown in FIG. 4 and the classification hierarchy shown in FIG. 5 are read from the partial similarity calculation procedure storage unit 10 in FIG. Convert and store. Also, the overall similarity calculation procedure 66 and the attribute weight 67 shown in FIG. 6 are read from the overall similarity calculation procedure storage unit 11 in FIG. 1, and are converted into an executable form and stored. Next, step 10 in FIG.
Based on the process 2, the case data 31 is extracted from the cases in FIG. 3 stored in the case base 9 in FIG. 1 and is set as a candidate for a similar example. By executing the partial similarity calculation procedure in FIG. 4 based on the processing in step 103 in FIG. 2,
The attribute between the input problem 71 in FIG. 7 and the case in the case data 31 in FIG. 3 and the distance for each attribute pair are obtained. First, since the first element (vertical side 10) of the input question 71 in FIG. 7 is an attribute pair, a distance is obtained according to the attribute pair similarity calculation procedure 41 in FIG. That is, an attribute pair (vertical side 10) is found from the left side of the case data 31 in FIG. Next, the attribute value (1
0) and the attribute value (10) of the input question 71 in FIG. 7 is obtained. Here, each attribute value is “1”.
0 ", and the absolute value is" 0 ". This value is defined as the distance related to the vertical side. Similarly, the distances regarding the horizontal side and the entrance are “0” and “1”, respectively.

Since the next element CPU in the input problem 71 of FIG. 7 and the case data 31 of FIG. 3 has no value and is not an attribute pair but an attribute, the CPU shown in FIG. Find the distance by referring to the hierarchy. Here, the problem on the left side of the case data 31 in FIG.
The upper nodes for both the attribute of the input question 71 in FIG. 7 and the attribute of the case data 31 in FIG.
It becomes itself. Therefore, the distance for the CPU is “0”
Becomes Similarly, the distance for CD, MDU, and MTU is “0”. However, regarding the LP, since the problem of the case data 31 in FIG. 3 does not include the LP but includes the CR,
In the classification hierarchy of FIG. 5, the upper node for both is the input / output device 53, and the distance is “2”. Also, FD
Regarding D, since the problem of the case data 31 in FIG. 3 does not include FDD but includes CR, in the classification hierarchy in FIG. 5, the upper node for both is the input device 57 and the distance is “1”. By the above operation, the case data 3 in FIG.
In FIG. 7, distances related to the vertical side, the horizontal side, the doorway, the CPU, the CD, the MDU, the MTU, the LP, and the FDD of the input problem 71 in FIG. 7 are “0, 0, 1, 0, 0, 0,
0, 2, 1 ".

Further, by executing the overall similarity calculation procedure 66 with reference to the attribute weights 67 in FIG. 6 based on the processing in step 104 in FIG. An overall distance between the input problem 71 and the problem of the case data 31 in FIG. 3 is obtained. Here, the distance is “5”. That is, the attributes with the distance “1” are the entrance and the FDD, and the weights of the entrance and the FDD are respectively as shown in the attribute weight 67 of FIG.
"2" and "1", and the distance between the two is "(1
× 2) + (1 × 1) = 3 ”, and the distance“ 2 ”
The weight of the attribute LP is “1”, and the sum is
“(1 × 2) + (1 × 1) + (2 × 1) = 5”.

Next, in step 105 of FIG.
“No”, and based on the processing of step 107, FIG.
Is taken out as the next candidate. And
Based on step 108 in FIG. 2, the distance for each attribute of the input question 71 in FIG. 7 and the next candidate (case data 32 in FIG. 3) is obtained by the same processing as in step 103. Here, vertical side, horizontal side, doorway, CPU, CD, MDU, MTU,
The distances for LP and FDD 62 are “1,
1, 2, 0, 0, 0, 0, 0, 0 ", and based on the processing of step 109 of FIG. 2, the input problem 71 of FIG. The distance as a whole of the data 32) is obtained. Here, the distance as a whole is “8” based on the attribute weight 67 in FIG.

In step 110 of FIG. 2, the distance between the input problem 71 of FIG. 7 and the next candidate (case data 32 of FIG. 3) is “8”, and the input problem 71 of FIG. Since the distance of 31) is “5”, the result is “No”, and the process returns to step 105 of FIG. This step 105
Is "No" because the case data 33 of FIG. 3 remains. Then, based on the processing in step 107 in FIG. 2, the case data 33 in FIG. 3 is extracted and set as the next candidate. By the processing based on step 108 in FIG. 2, the vertical side, horizontal side, doorway, CPU, CD, MDU, MTU,
The distance between the input problem 71 of FIG. 7 and the next candidate (the case data 33 of FIG. 3) relating to LP and FDD is “1, 0,
0, 0, 0, 0, 0, 1, 1 ". Further, the distance between the input question 71 in FIG. 7 and the next candidate (the case data 33 in FIG. 3) becomes “4” by the processing in step 109 in FIG. In step 110 of FIG. 2, FIG.
The distance between the input problem 71 and the next candidate (the case data 33 in FIG. 3) is “4”, the distance between the input problem 71 in FIG. 7 and the candidate (the case data 31 in FIG. 3) is “5”, and “Yes” , And the case data 33 in FIG. 3 is set as a new candidate based on the processing in step 111 in FIG. Further, in step 105 of FIG. 2, “Yes” is returned, and the case data 33 of FIG. 3 is returned to the control processing unit 8 of FIG. 1 as a similar example based on the processing of step 106 of FIG.

The control processing unit 8 of FIG. 1 receiving the case data 33 of FIG. 3 from the case search processing unit 5 of FIG.
Is passed to the case use processing unit 7 in FIG. The case use processing unit 7 in FIG. 1 obtains an answer to the input problem 71 in FIG. 7 based on the answer on the right side of the received case data 33 in FIG. 3 and returns the answer to the control processing unit 8 in FIG. . Then, the control processing unit 8 of FIG. 1 outputs the answer to the input problem to the output device 2 of FIG. 1 via the input / output processing unit 6 of FIG.

As described above with reference to FIGS. 1 to 7, the case retrieval system according to the present embodiment employs a partial similarity calculation procedure and an overall similarity calculation procedure that are separate and independent from the general-purpose case retrieval procedure. A similar example of the input problem is searched, and an answer to the input problem is obtained based on the answer in the similar example. As a result, a procedure other than the partial similarity calculation procedure and the overall similarity calculation procedure is provided as a standard system, and the partial similarity calculation procedure and the overall similarity calculation procedure are input according to the field. A similar example search system and an inference system according to the above can be easily developed. That is, a case search system that searches a similar example of an input problem from a case base, or a similar example of an input problem is searched from a case base by a case search system, and the problem is solved based on the similar example. When developing an inference system, the procedure of calculating the similarity (distance) between an attribute and a pair of attributes between a problem and a case, and calculating the overall similarity (distance) from these partial similarities (distance) If data or programs corresponding to the target field are developed (added) only for procedures, a general-purpose case search program can be used for the other parts. This reduces costs, manpower, and time involved in system development. Further, by changing the similarity calculation procedure for each attribute or attribute pair and the overall similarity calculation procedure, the search function can be easily adjusted, thereby facilitating system optimization and maintenance work. Note that, as described in the description of each drawing, the present invention is not limited to the embodiment described with reference to FIGS. For example, regarding a method of calculating a partial similarity, a system that calculates a similarity and performs a case search by using an image, a graphic, and a voice recognition technology for an image or a problem input by voice is also used. Can be applied. In this embodiment, the field-dependent procedure storage unit stores the partial similarity calculation procedure and the overall similarity calculation procedure separately and stores them.
Considering a general procedure of selecting cases in descending order of similarity to an input problem, the procedure of selecting cases in descending order of similarity corresponds to the overall similarity calculation procedure. The procedure can also be a general procedure.

[0031]

According to the present invention, the case search procedure is separated into a general part common to all case search operations and a part different for each field, and only the separated field-dependent parts are replaced. This makes it possible to search for cases corresponding to the fields described above, and to greatly reduce the load involved in system development and maintenance.

[0032]

[Brief description of the drawings]

FIG. 1 is a flowchart illustrating an operation of a case search process according to an embodiment of the present invention.

FIG. 2 is a flowchart showing one embodiment of a processing operation according to the present invention of a case search processing unit in FIG. 1;

FIG. 3 is an explanatory diagram showing one embodiment of case data stored in a case base in FIG. 1;

FIG. 4 is an explanatory diagram showing one embodiment of a partial similarity calculation procedure stored in a partial similarity calculation procedure storage unit of FIG. 1;

FIG. 5 is an explanatory diagram showing an embodiment of an attribute classification hierarchy used in the attribute similarity calculation procedure in FIG. 4;

FIG. 6 is an explanatory diagram showing an embodiment of an overall similarity calculation procedure stored in the overall similarity calculation procedure storage unit in FIG. 1;

FIG. 7 is an explanatory diagram showing one embodiment of a problem input to the case search system in FIG. 1;

[Explanation of symbols]

 REFERENCE SIGNS LIST 1 input device 2 display device 3 case search device 4 storage device 5 case search processing unit 6 input / output processing unit 7 case use processing unit 8 control processing unit 9 case base 10 partial similarity calculation procedure storage unit 11 overall similarity calculation procedure storage Sections 31 to 33 Case data 41 Attribute-to-similarity calculation procedure 42 Attribute value definition 43 Attribute-to-distance definition 44 Attribute similarity calculation procedure 45 Hierarchy definition 46 Attribute distance definition 51 Device 52 Computer 53 Input / output device 54 Storage device 55 Operating device 56 CPU 57 Input device 58 Output device 59 MDU 60 MTU 61 CD 62 FDD 63 CR 64 LP 65 LBP 66 Overall similarity calculation procedure 67 Attribute weight 71 Input problem

────────────────────────────────────────────────── ─── Continuation of the front page (56) References JP-A-63-54635 (JP, A) JP-A-63-138423 (JP, A) JP-A-1-288927 (JP, A) Kiyoshi Todo, Shunji Matsumoto , Tomoaki Sato, "Partitioning / reuse environment of knowledge system based on development methodology", Journal of the Japanese Society for Artificial Intelligence vol. 5, no. 2, pp. 69-75 (March 1, 1990) Hiroshi Yoshiura, "A Case-Based Reasoning Method Using Partially Similar Examples Based on Partitioning of Problem and Case", Transactions of Information Processing Society of Japan, vol. 32, no. 5, pp. 626-634 (May 15, 1991) (58) Field surveyed (Int. Cl. ⁶ , DB name) G06F 17/30

Claims (5)

(57) [Claims] 1. A case base for storing at least one case information including data representing a problem and data representing a solution to the problem in advance, and a problem similar to the problem input via input means provided. Case search means for searching from the case base based on a procedure stored in the storage means provided in advance in the case search means, in the case search procedure stored in the storage means, Field-dependent procedure storage means for storing a search procedure for each field depending on the field of the case information stored in the base is provided, wherein the storage means includes a general-purpose search procedure independent of the field of the case information stored in the case base. Characterized in that only the search procedure for each field stored in the field-dependent procedure storage means is changed when the case base is changed. Stem. 2. The case search system according to claim 1, wherein the search procedure for each field stored in the field-dependent procedure storage means is a problem of each case information stored in the case base by the case search means. A similarity calculation procedure for calculating a similarity to the input problem. 3. The case search system according to claim 1, wherein said input problem and said case search means corresponding to said input problem are stored in said field-dependent procedure storage means. A case search system comprising: a case use unit for performing inference based on case information searched using the above, and obtaining a solution to the input problem. 4. A case retrieval method for a case retrieval system according to claim 1, wherein the input problem and the problem of the case information stored in the case base are classified into a set of attributes. Or, it includes at least one of a set of attribute pairs, which is a pair of an attribute and a value of the attribute, and the field-specific search procedure for the input problem stored in the field-dependent procedure storage means includes: The case search means comprises a procedure for calculating a similarity or a similarity for each part including at least one of the similarities for each attribute pair.
Calculating the similarity for each part of the input problem and the case information problem using at least one of the above-described similarity calculating procedure or the similarity calculating procedure for each attribute pair. A case search method for a case search system, characterized in that a case search is performed. 5. The case search method of a case search system according to claim 4, wherein said field-dependent procedure storage means stores a similarity between said input problem calculated by said case search means and said case information problem for each part. A total similarity calculation procedure for calculating the overall similarity of the question of the case information with respect to the input problem, and the case search means uses the total similarity calculation procedure to calculate the input similarity. A case search method for a case search system, wherein a similarity between the problem and the problem of the case information as a whole is obtained, and a case from the case base for the input problem is selected based on the obtained similarity. .