JPH0553814A - Device for supporting preparation of event base retrieval system - Google Patents

Abstract

PURPOSE:To detect a featured and similar event together with a similar event having a high average point and to improve retrieval ability by making event to subvectors and detecting the featured and similar event by observing them in terms of the unit of the subvector. CONSTITUTION:A subvector similarity calculation part 8 calculates similarity between a question and the event by calculating the total sum of attribute similarity included in the subvector at every subvector divided by a vector division part 6. For example, data are collected by sampling the daily body temperature of a patient at every other one hour and when daily body temperature information is defined as one event, it is expressed by 24 attribute vectors since the information is sampled at every other one hour. When one day is segmented into the subvectors like defining hours from 21:00 to 5:00 as 'midnight', from 6:00 to 11:00 as 'forenoon', from 12:00 to 17:00 as 'afternoon' and from 18:00 to 20:00 as 'evening', in the case of retrieving the event similar to a certain question, the event having a body temperature change not be similar so much concerning the 'forenoon', 'afternoon' and 'evening' but to be quite similar concerning the 'midnight' can be retrieved.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a case-based search system creation support device used in the field of artificial intelligence.

[0002]

[Prior Art] Case-based Reasoning
g, hereinafter abbreviated as CBR) is a new inference paradigm that has recently received attention in the field of artificial intelligence (references:
"Case-Based Reasoning Expected to be Practical as a New Problem Solving Method" Shigenobu Kobayashi, Kotaro Nakamura, Nikkei AI Separate Volume 199
1 spring issue).

The structure of the CBR is shown in FIG. The case base stores a large number of cases composed of a set of problems solved in the past and a solution method. When a new problem to be solved is given, the case that solves the problem most similar to the problem is retrieved from the case base (search phase). C
In BR, searching for similar cases is an important process. When searching the case base, exactly match the problem to be solved (exac
without partial matching (partial matchi
ng). Next, in order to absorb the difference between the problem expression in the searched similar case and the problem expression to be solved, the system modifies the part and adaptively corrects a part of the similar case (correction phase). Is presented as a solution to the problem to be solved. The solution created in this way is not always correct, so try applying it. If the result is correct, it is stored in the case base as a positive case showing "new case and its solution", and when it fails, the reason for failure is analyzed and "new problem and wrong solution and its error “Criteria for not repeating” is stored in the case base (evaluation / repair phase). The success or failure of the application result may be determined by the CBR system itself or by the user. Thus, in CBR, when a new problem is given, whether the solution succeeds or fails, the result is always added to the case base of the system. If it fails, index it so that you do not make the same mistake twice.

When a system is made based on the concept of CBR, there are many difficult problems in each local phase of CBR, and no local phase can be easily created. For example: How would you describe the accumulated cases (both problem and solution)? -In what structure is the case stored in the case base? • How do you find examples that are "close" to a given problem in the search phase? Partial matchi
ng) is defined? -In the correction phase, how are the similar cases taken out corrected according to the difference between the two?・ In the repair phase, when a solution created by transforming a similar case was applied, but it was incorrect, how would the failure be reflected in the system?

Although many studies have been made for each problem, general-purpose and decisively good ones have not yet appeared. One way to deal with such problems is to defeat each one by a straightforward method, but all of the above problems are very difficult and general solution methods are available. Is difficult to present immediately. As another coping method in such a case, as an engineering idea, there is a method of effectively avoiding the problem of the CBR by partially limiting the potential capability of the CBR. Based on this policy, a tool for CBR has been proposed (reference: uninterface for
Case-based knowledge acquisition, C.I. Leedsbeck, An Interface for Case-Based Knowledge Acqu
isition, Proc. of Workshop on Case-Based Reasoning
, DARPA, 1988. ). A schematic diagram of this system is shown in FIG. In FIG. 16, 101 is a case attribute vector, 102 is a case base, 103 is a question storage attribute vector, 104 is a partial matching definition unit, 105 is a weight definition unit, 107 is an attribute similarity calculation unit, and 109 is a total similarity calculation unit. ,. Although FIG. 16 does not have the same configuration as this embodiment in the above reference, it is the same in principle. Many cases expressed in the form of a vector of attributes are stored in the case base 102. When the problem to be solved is stored in the problem storage attribute vector 103, the cases in the case base 102 are extracted one by one, The case attribute vector 101 is loaded, and the degree of similarity is calculated. The definition of the partial matching between the corresponding element attributes is defined in the partial matching definition unit 104, and the weight of the element attribute is stored in the weight definition unit 105. When the attribute similarity calculation unit 107 corresponding to each attribute calculates the attribute similarity, the total similarity calculation unit calculates the sum.

[0006] There are the following problems to be solved in such a conventional technique.

The first problem is that, in the prior art, when a case and a problem to be solved are expressed as a vector of several attributes that have been defined in advance, all the attributes in the entire attribute vector are fairly similar. However, it does not look like an important similar case from the viewpoint that the average score is high,
A part of the attribute vector is very similar to the problem to be solved, which means that it is not possible to search for cases that deserve attention enough. A part of this attribute vector may be one element attribute,
It may also be a set of several element attributes.

The second problem is as follows. When expressing a case, it is a concrete value because the value that is actually experienced or existing is used. Therefore, with the conventional method, the case definer who has the experience of “going to Disneyland every day and always having fun” was “going to Disneyland in January was fun” and “going to Disneyland in February was fun” "I had a lot of fun going to Disneyland in March" ... "I had a lot of fun going to Disneyland in December", so I had to give 12 cases. If you allow the expression "perfectly matches anything", in the above example, you can define only one, "at the time the attribute value is" anytime "and it was fun to go to Disneyland". Hopefully this isn't possible with conventional technology. However, if such an expression method is allowed, whether it is a real existence case, a kind of generalized case with some speculation, or a so-called common sense or speculation that does not support facts at all. Even if the second problem is solved by providing some new expression means, there is a possibility that another problem will be caused because the case-based credibility will be suspected if it is not clearly divided.

The third problem is that, in the conventional technology, there is no means for clearly indicating such a situation when an exceptional or singular point case is included in a higher-level concept including a case. For example, let's say that Yokohama is a good place to travel. However, in Yokohama City, only "Tsurumi" is a factory zone and is not suitable for travel. At this time, it is assumed that the case was given that "I traveled to Tsurumi in November, but it was not fun". here,
Given that the question "Is Yamashita Park suitable for travel?" And there is only one example of the travel report in Yokohama City from the above perspective, this only case becomes a similar case.
You will end up deciding that it is not suitable for travel. This is because the person who stored the case "I traveled to Tsurumi in November but it was not fun" can be said to be "not fun because the place is" Tsurumi ", and the matching partner who inquires is also" Tsurumi ". However, even if it is in the vicinity of Tsurumi (eg Yamashita Park), I would like to make the attributes of both do not match. "

The fourth problem relates to the definition of partial matching for each element attribute and the change of the weight value for each element attribute in order to improve the performance of the CBR system. Generally, the performance of similarity search in CBR is determined by the definition of partial matching for each element attribute and whether or not the weight is given for each element attribute. In a general CBR application system development process, a certain partial matching definition and attribute weight are determined, various problems are given, and similar examples output to them are looked at, and the partial matching definition and attribute weight are used as parameters. It is about changing. However,
When these parameters are actually changed, the performance as a CBR system does not consistently improve, and it may happen that the previous parameters have better performance. Therefore, a problem left in the prior art is that it is convenient to save the tried partial matching definitions and attribute weights and retrieve them as needed, but such a mechanism is not incorporated.

The fifth problem is that it is effective to have a plurality of partial matching definitions and attribute weight definitions and to switch the partial matching definitions and attribute weight definitions to be used according to the type of case. Then that kind of mechanism is not incorporated. For example, in the conventional method, when there is a chart for each person and performance is improved when the partial matching definition or weight attribute definition of a specific attribute is switched depending on the gender, the switching cannot be performed depending on the gender.

The sixth problem is related to the fourth problem. C
When the definition of partial matching and the value of the weight for each element attribute are changed for the similar case search of the BR system, the similar case for a certain problem was correctly searched before the change, but the order is not correct after the change. May be searched by. If such a situation occurs, the change must be corrected immediately, but the traditional method is to accurately monitor this situation and
No means was provided to notify the R system builder of the occurrence of such a situation.

[0013]

SUMMARY OF THE INVENTION The object of the present invention is to add a mechanism for solving the above six problems to a conventional CBR system creation tool so that the problems of the CBR can be satisfactorily avoided and the CBR can be easily implemented.
It is to provide a construction device of a creation tool that can create an application problem in the framework of.

[0014]

[Means for Solving the Problems] The first aspect of the present invention is to solve the first problem by providing a case attribute vector which collectively represents an input to a system and a corresponding output from the system, and a case. A case base holding a plurality of attribute vectors, a problem storage attribute vector having the same structure as the case attribute vector storing the problem to be solved, and matching between the case attribute vector and the corresponding element attribute value of the problem storage attribute vector A partial matching definition unit that defines a degree, a weight definition unit that defines weights for each element attribute of the case attribute vector and the problem storage attribute vector, and a case attribute vector and the problem storage attribute vector in one or more sub A vector division unit for dividing into vectors, and a match between the values of the corresponding element attributes of the case attribute vector and the problem storage attribute vector An attribute similarity calculation unit that calculates a degree with reference to the partial matching definition unit and outputs a value multiplied by the corresponding weight defined in the weight definition unit for each element attribute, and the vector division unit. For each subvector divided by, a subvector similarity calculation unit that outputs the sum of the outputs of the attribute similarity calculation units corresponding to the element attributes belonging to the subvector, and the subvector similarity calculation unit A total similarity calculation unit that outputs a sum of outputs, and stores a problem to be solved in the problem storage attribute vector, extracts one case from the case base, and stores the case in the case attribute vector, A control unit for instructing to calculate the attribute similarity, the sub-vector similarity, and the total similarity for each element attribute; Compare the attribute similarity, sub-vector similarity, and total similarity for each elementary attribute between cases, select the case with a large total similarity, sub-vector similarity, or attribute similarity, and select the problem to be solved. A case-based search system creation support device including a case selection / presentation unit that is presented as a similar case.

The second aspect of the present invention is to solve the second problem. In the first aspect, the value of the element attribute in the case attribute vector or the question storage attribute vector "completely matches any value". When the attribute similarity calculation unit receives the expression, the function that interprets the degree of matching between attribute values as “completely matched” is output as the attribute similarity calculation. It is a case-based search system creation support device added to the department.

The third aspect of the present invention is to solve the third problem. In the first aspect of the invention, the value of the element attribute in the case attribute vector or the question storage attribute vector is "when there is no complete match, The expression "meaning that all the similarities between the case attribute vector and the question storage attribute vector are set to 0" is allowed, and when the attribute similarity calculation unit receives the expression, the degree of matching between the attribute values is changed. , If the attribute values are equal, "completely matched"
A case in which a function of interpreting the output of the attribute similarity calculation unit and the output of the sub-vector similarity calculation unit including the element attribute and the total similarity calculation unit as the similarity 0 is added It is a base search system creation support device.

A fourth aspect of the present invention is to solve the third problem. In the first aspect, the value of the element attribute in the case attribute vector or the question storage attribute vector is "when not completely matched. , The attribute similarity calculator between the case attribute vector and the question storage attribute vector and the value of the subvector similarity including the target element attribute are set to 0, and the attribute similarity calculator calculates the expression. When the attribute values are received, the degree of matching between the attribute values is output as "completely matched" if the attribute values are equal to each other. Otherwise, the attribute similarity calculation unit and the sub-vector sub-vector including the element attributes are output. A case-based search system creation support apparatus having a function of interpreting the output of a vector similarity calculation unit so as to output the similarity as 0.

A fifth aspect of the present invention is to solve the third problem. In the first aspect, the value of the element attribute in the case attribute vector or the question storage attribute vector is "when there is no complete match. The expression "meaning that the attribute similarity between the case attribute vector and the question storage attribute vector is 0" is allowed, and when the attribute similarity calculator receives the expression, the matching between attribute values is performed. A case-based search with a function of interpreting the degree of output as “completely matched” if the attribute values are equal to each other, and outputting the output of the attribute similarity calculation unit as a degree of similarity of 0 otherwise This is a system creation support device.

The sixth aspect of the present invention is to solve the fourth and fifth problems, and in the first aspect, it is possible to have a plurality of the partial matching definition parts for one element attribute. Of the multiple partial matching definition parts,
It is a case-based search system creation support device to which a function of performing calculation by the attribute similarity calculation unit using an externally designated one is added.

A seventh aspect of the present invention is to solve the fourth and fifth problems, and in the first aspect, it is possible to have a plurality of the weight defining parts for one element attribute. The case-based search system creation support apparatus to which the attribute similarity calculation section performs a calculation by using one of the sets of weight definition sections specified from the outside.

The eighth aspect of the present invention is to solve the second problem. In the first aspect of the present invention, all the attribute values actually exist for each case stored in the case base. A case type is used for a thing, a case definer generalizes the values that actually existed for some element attribute values and stores them in the case base, and a generalized case type is used for all the element attributes. The common sense type is allowed to be given to a value that does not actually exist, and when the case selection presenting section presents a case similar to the given problem to be solved, the common sense type is added at the same time. It is a case-based search system creation support device provided with a function of presenting a type.

A ninth aspect of the present invention is to solve the fourth and fifth problems, and in the sixth aspect, when a plurality of the partial matching definition parts are defined, the attribute similarity calculation part is defined. , The partial matching definition parts
The sub-vector similarity calculation unit individually extracts the attribute similarity calculation results, and the sub-vector similarity calculation unit individually calculates the sub-vector similarity using the attribute similarity calculation results of the element attributes using the respective partial matching definition units. The total similarity calculation unit calculates the total similarity individually using the sub-vector similarity calculation results obtained by using the partial matching definition units. A case in which a function of individually outputting the attribute similarity calculation result of the calculated element attribute, the subvector similarity calculation result, and the total similarity calculation result is given to each of the partial matching definition parts It is a base search system creation support device.

A tenth aspect of the present invention is to solve the fourth and fifth problems, and in the seventh aspect, when a plurality of weight defining sections are defined, the attribute similarity calculating section is , Each of the weight defining units is individually extracted and attribute similarity calculation is performed individually, and the sub-vector similarity calculating unit individually calculates the attribute similarity of the element attribute using each of the weight defining units. To calculate the sub-vector similarity, the total similarity calculation unit individually calculates the total similarity using the sub-vector similarity calculation results using the weight definition unit, the case selection presentation The unit has a function of individually outputting the attribute similarity calculation result of the calculated element attribute, the subvector similarity calculation result, and the total similarity calculation result for each of the plurality of weight definitions. It is given the case-based search system creating support device.

The eleventh invention is to solve the sixth problem. In the first invention, a problem to be solved is given, and a result similar to the problem presented by the case selection and presentation unit is presented as a problem. When all of them are stored as a set and the user changes the partial matching definition part, the stored problems are taken out one by one, the similarity is calculated, a new similar case sequence is newly obtained, and the calculated result is obtained. A case-based search system creation support device provided with a mechanism for notifying the user when the similar case sequence is different from the similar case sequence paired with the stored problem.

A twelfth aspect of the present invention is to solve the sixth problem, and in the first aspect, a problem to be solved is given, and a result similar to the problem presented by the case selection presenting unit is presented as a problem. When all the sets are stored and the user changes the weight definition part, the stored problems are taken out one by one, the similarity degree is calculated, a new similar case sequence is obtained, and the calculated result is obtained. When the similar case sequence is different from the similar case sequence paired with the stored problem,
It is a case-based search system creation support device provided with a mechanism for notifying a user.

[0026]

FUNCTION AND PRINCIPLE The present invention is expressed by the means described in claims 1 to 12. Claim 1 represents the entire basic processing, and claims 2 and below are those in which important functions are individually added to claim 1. Therefore, (action and principle)
In the section (1), when claim 1 is mainly described, and when the contents of claim 2 and the subsequent claims are explained, it is made clear in the text which claim is mentioned.

To explain the operation of the present invention, a very simple example will be used. In the example system, "How much fun would you have if you traveled to a place that departed at one time and came back at another time?" And store it. When a user searches for a case base, he or she inputs the departure time, the return time, and the destination as problems, and searches for cases that are close in time and place and presents them to the user.

[0028] Claim 1 is an apparatus for solving the problem 1. Consider constructing this example system with the case-based search system creation support device described in claim 1. The case attribute vector includes “departure time”, “return time”, “place”, and “fun” as attributes. Therefore, one case is defined by an attribute vector consisting of four attributes. The case base stores actual travel experience cases represented by this attribute vector. For example, when "departing in February, traveling in the Mediterranean and returning to Japan in March, there are 35 points of enjoyment" is represented by (February, March, Mediterranean, 35).

The problem expression is stored in the problem storage attribute vector, and the problem storage attribute vector has the same attribute vector structure.

For each individual attribute, the definition of the matching degree when the values of the attribute are compared is described in the partial matching definition section. For example, in FIG. 14, "departure month"
Alternatively, the degree of matching between “return months” and the degree of partial matching of “place” are defined in FIG. Similarly, each attribute is given a weight that represents how important the attribute is as a whole, and they are stored in the weight defining section. Now, the weight of "departure time" is 0.4, the weight of "return time" is 0.1, and the weight of "place" is 0.5.

The attribute similarity calculation unit stores each case in the case base in a case attribute vector, and calculates the degree of matching between the values of the element attributes corresponding to the case attribute vector and the question storage attribute vector. .. The calculation formula is "Departure month"
In the case of, the partial matching definition part shown in FIG. 14 is referred to. If one is “December” and the other is “February”, the two belong to the category of “winter”, so the match degree is 0.8. The final match degree is a value obtained by multiplying this by a weight of 0.4 and normalizing it to a value between 0 and 1, that is, (0.8 * 0.4) / (0.4 + 0.1 +
0.5) = 0.32. Also, “December” and “6
In the case of “month”, both parties do not match unless they trace back to the highest level category of “season”, so the match degree is 0.
The attribute similarity calculator calculates such a match degree for each element attribute.

The vector division unit defines the attribute vector into sub-vectors divided into meaningful units. In this example, “departure month” and “return month” are treated as one subvector, and “place” and “fun” are defined as one subvector with one attribute. The sub-vector similarity calculation unit sums up the attribute similarity of the attributes included in each sub-vector divided by the vector division unit to calculate the sub-vector similarity between the problem and the case in that sub-vector To do.

The sub-vectorization is remarkably effective when the information which changes with time is used as one case. For example, if the patient's daily body temperature is sampled every hour and data is collected, and one patient's daily body temperature information is taken as one case, the data is collected every hour, so 2
It is represented by four attribute vectors. This attribute vector is defined as "midnight" from 21:00 to 5:00, "am" from 6:00 to 11:00, "pm" from 12:00 to 17:00, and "evening" from 18:00 to 20:00. If you divide it into vectors, you can search for similar cases given a certain problem. For example, "Am", "Afternoon", "Evening" are not very similar, but about "midnight" changes in body temperature. It is possible to search for "cases that are very similar".
The feature of claim 1 is that cases can be treated as subvectors in this way, and characteristic similar cases can be detected in terms of subvectors. The total similarity calculation unit is the sum of the outputs of the individual sub-vector similarity calculation units.

The second aspect of the invention solves the second problem. According to the second aspect, the description "perfectly matches any value" is allowed as the expression described in the element attribute, and the interpretation execution function is added to the attribute similarity calculation unit of the first aspect.

Claim 3 solves the third problem. Claim 3 permits the description "when all of the similarities between the case attribute vector and the question storage attribute vector are 0" when the expressions are not completely matched in the element attribute, the interpretation is executed. The function of claim 1
It is added to the attribute similarity calculation unit, the sub-vector similarity calculation unit, and the total similarity calculation unit. For example, if there is a description of must-completely-exact-match ("Tsurumi") as the value of the element attribute in the question storage attribute vector or case attribute vector, if the value of the corresponding attribute is "Tsurumi", then this The degree of matching between attributes is 1, but other values are 0, and the similarity of subvectors that include this attribute is also 0.
And the total similarity between this case attribute vector and this question storage attribute vector is also zero.

Claim 4 solves the third problem. According to claim 4, as a description described in the element attribute, a description "when there is no complete match, the subvector similarity and the attribute similarity between the case attribute vector and the question storage attribute vector are set to 0" is allowed. The interpretation execution function is added to the attribute similarity calculation unit and the sub-vector similarity calculation unit of claim 1. For example, if there is a description "must-almost-completely-exact-match"("Tsurumi") as the value of the element attribute in the question storage attribute vector or the case attribute vector, if the corresponding attribute value is "Tsurumi" , The matching degree between the attributes is 1, but other values are 0, and the sub-vector similarity including this attribute is also 0.

Claim 5 solves the third problem. According to a fifth aspect of the present invention, as an expression described in the element attribute, "when there is no complete match, the attribute similarity between the case attribute vector and the question storage attribute vector is set to 0".
Is allowed and the function of interpretation execution is added to the attribute similarity calculation unit of claim 1. For example, if there is a description of must-exact-match-this-attribute (“Tsurumi”) as the value of the element attribute in the question storage attribute vector or the case attribute vector, if the corresponding attribute value is “Tsurumi” , The degree of matching between the element attributes is 1, but it is 0 for other values.

Claim 6 solves the fourth and fifth problems. Claim 6 is an extension of the partial matching definition part so that a plurality of matching definition information can be stored for one element attribute, and when a user instructs which partial matching definition information to use from the outside. It is used when calculating attribute similarity.

Claim 7 solves the fourth and fifth problems. Claim 7 is an extension of the weight definition section so that a plurality of weight information can be stored for one element attribute, and when the user instructs which weight definition information is to be used from the outside, that attribute is Used when calculating similarity.

Claim 2 is an apparatus for solving the second problem. According to claim 8, one storage area is added to the case attribute vector as the case expression stored in the case base, and a value of “case” or “generalization case” or “common sense” can be stored therein for each case. It is an extension to.

The ninth aspect is an apparatus for solving the fourth and fifth problems. A ninth aspect is a development of the sixth aspect, wherein the partial matching definition part is expanded so that a plurality of partial matching definition information can be stored for one element attribute, and a certain problem is given, and a similar case search process is performed. When performing, the attribute similarity calculation is performed by using a plurality of partial matching definition information individually, and the subvector similarity calculation is performed by using the attribute similarity calculation results calculated independently of each other.
This is an extension to perform the total similarity calculation using the sub vector similarity calculation results calculated independently. Therefore, the calculation result of the similarity between the problem attribute vector and one case is generated by the number of held partial matching definition information.

[0042] Claim 10 is an apparatus for solving the fourth and fifth problems. Claim 10 is a development of Claim 7,
The weight definition section is expanded so that a plurality of weight definition information can be stored for one element attribute, and when a certain problem is given and a similar case search process is performed, the plurality of weight definition information are individually used, Attribute similarity calculation is performed, and the attribute similarity calculation result calculated independently is used to calculate the subvector similarity, and the subvector similarity calculation result calculated independently is used to calculate the total similarity. It is an extension to do. Therefore, there are as many similarity calculation results between the problem attribute vector and one case as the weight definition information held.

The eleventh aspect is an apparatus for solving the sixth problem. Claim 11 is a case example in which the case-based definer changes the partial matching definition in order to improve the search performance, and automatically recalculates a problem that has worked so far, and the change does not work correctly. It is presented to the base definer.

A twelfth aspect of the invention is an apparatus for solving the sixth problem. Claim 12 is a case-based case example in which the case-base definer automatically recalculates a problem that has worked up to now as a result of changing the weight definition in order to improve search performance, and the change does not work correctly. It is presented to the definer.

[0045]

DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 is a diagram showing the first invention. In FIG. 1, 1 is a case attribute vector, 2 is a case base, 3 is a problem storage attribute vector, 4 is a partial matching definition part, 5
Is a weight definition unit, 6 is a vector division unit, 7 is an attribute similarity calculation unit, 8 is a sub-vector similarity calculation unit, 9 is a total similarity calculation unit, 10 is a control unit, and 11 is a case selection presentation unit. In FIG. 1, since the number of attribute vectors is four, the vector length of the case attribute vector 1, the record length of one record of the case base 2, the vector length of the problem storage attribute vector 3, the number of partial matching definition parts 4 , The number of weight definition units 5 is four, but this number does not limit the present invention. Further, in FIG. 1, as the subvectors of the attribute vector, two attributes from the left, one attribute, and one attribute are defined as one subvector by the vector division unit 6 as a total of three subvectors. This does not limit the claims. Along with this, the number of sub-vector similarity calculation units 8 is three, but this also does not limit the present invention.

When the problem to be solved is given and stored in the problem storage attribute vector 3, the control unit 10 takes out the cases stored in the case base 2 one by one and stores them in the case attribute vector 3. go. The attribute similarity calculation unit 7 calculates the degree of partial matching of the values of the element attributes corresponding to the case attribute vector 1 and the question storage attribute vector 3 with reference to the definition of the partial matching definition unit 4. Since the inside of the partial matching definition unit 4 is defined in advance as shown in FIGS. 14 and 15, the degree of partial matching of two attribute values is calculated by referring to it. Here, the value is assumed to be a value between 0 and 1. The weight definition unit 5 stores the weight for each attribute. Figure 1
In the above, the weight values are set to 0.4, 0.1, 0.3 and 0.2 so that the total sum of the weights becomes 1. However, the present invention is not limited to this. The attribute similarity calculation unit 7 multiplies the value of the degree of partial matching calculated with reference to the partial matching unit 4 by the value of the weight definition unit 5 and outputs the result. The sub-vector similarity calculation unit 8 sums up and outputs the input outputs of one or more attribute similarity calculation units 7.
Further, the total similarity calculation unit 9 sums the outputs of all the sub-vector similarity calculation units 8 and outputs the sum. The outputs of the sub-vector similarity calculation unit 8 and the total similarity calculation unit 9 are passed to the case selection presentation unit 11. When the above calculation is completed for one case in the case base, the control unit 10 extracts the next case from the case base 2 and stores it in the case attribute vector 1 to perform the same calculation. After all the cases in the case base 2 and the similarities in the problem storage attribute vector 3 are calculated, the case selection presenting unit 11 holds the similarity and the total similarity for each subvector of each case. Has been done. The case selection presenting unit 11 presents the values in descending order of total similarity. In addition, a case in which one or more sub-vector similarity values take a significantly large value will be similarly presented. An external user gives the user how many "remarkably large values" to set. In the case of the present embodiment, the value of the sub-vector similarity is 0 or more and 0.5 or less. Therefore, for example, if 0.45 is given as the lowest value as a “significantly large value”, the sub-vector similarity is more than that value We will present examples taken as a degree.

FIG. 2 shows the second aspect of the present invention. Figure 2
Shown by means of expanded internal processing of the attribute similarity calculation unit 7 of FIG. This allows the interpretation and execution of new instructions.

FIG. 3 shows the third aspect of the present invention. Figure 3
Shown by means of expanded internal processing of the attribute similarity calculation unit 7 of FIG. This allows the interpretation and execution of new instructions.

FIG. 4 shows the fourth aspect of the present invention. Figure 4
Shown by means of expanded internal processing of the attribute similarity calculation unit 7 of FIG. This allows the interpretation and execution of new instructions.

FIG. 5 shows the fifth aspect of the present invention. Figure 5
Shown by means of expanded internal processing of the attribute similarity calculation unit 7 of FIG. This allows the interpretation and execution of new instructions.

FIG. 6 is a diagram for explaining the sixth aspect of the present invention.
In FIG. 6, 1 is a case attribute vector, 11 is a partial matching definition storing unit, 3 is a problem storing attribute vector, 4 is a partial matching defining unit, 5 is a weight defining unit, 12 is a partial matching definition designating unit, and 7 is an attribute similarity. It is the degree calculator. Sixth
The invention of is a partially expanded version of the first invention.
The expanded component is the partial matching definition storage unit 11
And the partial matching definition designation unit 12, and the other components are exactly the same as those in FIG. In FIG. 6, for simplification of description, one element attribute in the case attribute vector 1 and the question storage attribute vector 3 is focused and shown.
Further, in FIG. 6, not all the constituent elements of FIG. 1 are shown, but only the constituent elements directly related to the expanded partial constituent matching definition storage unit 11 and the partial matching definition specifying unit 12 are expressed. Partial matching definition storage unit 11
The partial matching definition expression 12 stores a plurality of partial matching definition expressions. When the user specifies in the partial matching definition specifying unit 12, the specified definition is selected and the partial matching definition unit 4 is selected. To load. The subsequent processing is exactly the same as that of the first invention.

FIG. 7 is a diagram for explaining the seventh invention.
In FIG. 7, 1 is a case attribute vector, 21 is a weight definition storage unit, 3 is a problem storage attribute vector, 4 is a partial matching definition unit, 5 is a weight definition unit, 22 is a weight definition designating unit, 7
Is an attribute similarity calculation unit. The seventh invention is a partial extension of the first invention. The extended components are the weight definition storage unit 21 and the weight definition designation unit 22,
The other components are the same as those in FIG.
In FIG. 7, for simplification of description, one element attribute in the case attribute vector and the question storage attribute vector 3 is focused and shown. Further, in FIG. 7, not all of the constituent elements of FIG. 1 are shown, but only constituent elements directly related to the weighted definition storage unit 21 and the weighted definition designation unit 22, which are expanded constituent elements, are expressed. A plurality of weight definition expressions are stored in the weight definition storage unit 21, and when the user designates which one of them is to be used by the weight definition designation unit 22, the designated definition is selected and the weight definition is selected. Load in part 5. The subsequent processing is exactly the same as that of the first invention.

FIG. 8 is a diagram for explaining the eighth aspect of the present invention.
In FIG. 8, 1 is a case attribute vector, 2 is a case base, 3 is a problem storage attribute vector, 4 is a partial matching definition unit, 5 is a weight definition unit, 6 is a vector division unit, 7 is an attribute similarity calculation unit, 8 Is a sub-vector similarity calculation unit, 9 is a total similarity calculation unit, 10 is a control unit, and 11 is a case selection presentation unit. This is the same as that of the same number in FIG. Three
Reference numeral 1 is a case type storage unit. Case type storage unit 31
Is used to store the type of a case, corresponding to each record in case base 2. In the case type storage unit 31, three types of "case", "generalization case" and "common sense" are prepared. However, this claim
It is not limited to one type. These types are those that the person who defines the case and stores it in Case Base 2 stores it together when it is stored. The value of the case type storage unit 31 is not used during the similar case inspection. When the case selection presenting unit 11 presents a similar case to a given problem, the value of the case type storage unit 31 corresponding to the similar case is referred to and presented together.

Although not described here, one possibility is that, when given a problem, rather than searching all cases in Case Base 2, only cases of a particular type are specified by the user. It is also possible to add a function of searching for.

FIG. 9 is a diagram for explaining the ninth invention.
In FIG. 9, 1 is a case attribute vector, 11 is a partial matching definition storage unit, 3 is a problem storing attribute vector, 4 is a partial matching definition unit, 5 is a weight defining unit, 42 is a partial matching definition designation control unit, and 7 is an attribute. It is a similarity calculator.
The ninth invention is a partial expansion of the sixth invention. The components expanded from the first invention are a partial matching definition storage unit 11 and a partial matching definition designation control unit 4.
2 and other components are exactly the same as those in FIG. The partial matching definition storage unit 11 is the same as the component having the same name and the same number in FIG. In FIG. 9, for simplification of explanation, the case attribute vector 1 and the question storage attribute vector 3
I paid attention to one element attribute in. Also, FIG.
In FIG. 1, not all of the components shown in FIG. 1 are written, but only the components that are directly related to the expanded partial matching definition storage unit 11 and the partial matching definition designation control unit 42 are expressed. The partial matching definition storage unit 11 stores a plurality of partial matching definition expressions. When the user sets the problem to be solved in the problem storage attribute vector 3, a similar case is searched for as in the first invention. At this time, the partial matching definition designation control unit 42 is used.
Takes one partial matching definition from the partial matching definition storage unit 11 and stores it in the partial matching definition unit 4. When the similarity calculation of all the cases in the case base for the problem stored in the problem storage attribute vector 3 is completed and the result is presented, the partial matching definition designation control unit 42 then causes the partial matching definition storage unit 11 Another partial matching definition is taken out from and is set in the partial matching definition unit 4, and the similarity of each case in the case base is calculated again for the same problem as before. This process is performed for all the partial matching definitions in the partial matching definition storage unit 11.

FIG. 10 is a diagram for explaining the tenth invention. In FIG. 10, 1 is a case attribute vector, 21 is a weight definition storage unit, 3 is a question storage attribute vector, 4 is a partial matching definition unit, 5 is a weight definition unit, 52 is a weight definition designation control unit, and 7 is an attribute similarity degree. It is a calculation unit. The tenth invention is a partial extension of the seventh invention. The components expanded from the first invention are the weight definition storage unit 21 and the weight definition designation control unit 52, and the other components are the same as those in FIG. The weight definition storage unit 21 is the same as the component having the same name and the same number in FIG. In FIG.
For simplification of description, one element attribute in the case attribute vector 1 and the question storage attribute vector 3 is focused and shown. Further, in FIG. 10, all the constituent elements of FIG. 1 are not written, but the weight definition storage unit 21 which is an expanded constituent element.
Only the components directly related to the weight definition designation control unit 52 are represented. The weight definition storage unit 21 stores a plurality of weight definition expressions. When the user sets the problem to be solved in the problem storage attribute vector 3, the similar case is searched in the same manner as in the first invention. At this time, the weight definition designation control unit 52 causes the weight definition storage unit 11 to execute the search. One weight definition is extracted and stored in the weight definition unit 5. When the similarity calculation of all the cases in the case base for the problem stored in the problem storage attribute vector 3 is completed and the result is presented, the weight definition designation control unit 52 then separates from the weight definition storage unit 21. One weight definition is taken out and set in the weight definition unit 5, and the similarity of each case in the case base is calculated again with respect to the same problem as before. This process is performed for all weight definitions in the weight definition storage unit 21.

FIG. 11 is a diagram for explaining the eleventh invention. In FIG. 11, 61 is the entire embodiment of FIG.
Is a partial matching definition change detection unit, 63 is a retest control unit, 64 is a past question answer storage unit, and 65 is a search result comparison unit. An eleventh invention is one in which a function is added to the first invention. In FIG. 11, 61 represents the entire embodiment of FIG. When the user gives a question as an input to the whole example 61 of FIG. 1 and the whole example 61 of FIG. 1 outputs a similar case to it, the past problem answer storage unit 64 outputs all input question and output pairs. Save it. When the user or the case-based administrator changes the partial matching definition unit 4 in FIG. 1, the partial matching definition change detection unit 62 recognizes it and notifies the retest control unit 63 of it. The retest control unit 63 takes out a set of question input and output one by one from the past question answer storage unit 64, transfers the question input to the embodiment 61 of FIG. 1, and searches for a similar case. Example 6 of FIG.
1 outputs a similar case of the given question input, but the search result comparison unit 65 outputs the output corresponding to the question input during the retest taken from the past question answer storage unit 64 and the embodiment of FIG. Compare the output from 61 to see if they are equal. When both are equal, nothing is notified, but
When the two are not equal, the user is notified that "because the partial matching definition has been changed, an example in which similar cases have been successfully searched in the past now operates differently".

FIG. 12 is a diagram for explaining the twelfth invention. In FIG. 12, 71 is the entire embodiment of FIG. 1, 72 is a weight definition change detection unit, 73 is a retest control unit, 74 is a past question answer storage unit, and 75 is a search result comparison unit. 12th
The invention of is a function added to the first invention. In FIG. 12, 71 represents the entire embodiment of FIG. When the user gives a question as an input to the whole embodiment 71 of FIG. 1 and the whole embodiment 71 of FIG. 1 outputs a similar case to it, the past problem answer storage unit 74 outputs all the input problem and output pairs. Save it. When the user or the case-based administrator changes the weight definition unit 5 in FIG. 1, the weight definition change detection unit 72 recognizes it and notifies the retest control unit 73 of it. The retest control unit 73 retrieves a set of question input and output from the past question answer storage unit 74 one by one, transfers the question input to the embodiment 71 of FIG. 1, and retrieves a similar case. The embodiment 71 of FIG. 1 outputs a similar case of the given question input, but the search result comparison unit 75 outputs the output corresponding to the question input during the retest taken from the past question answer storage unit 74 and the figure. The output from Example 71 of No. 1 is compared to see if they are equal. When both are equal, nothing is notified, but when both are not equal, the user said that "the weight definition was changed, so that the case where similar cases were successfully searched in the past behaves differently". To notify.

[0059]

According to the present invention, in the similar case search operation of the case-based system construction support system, not only the similar case in which the average point is good, but also the characteristic point is detected as a characteristic similar case, although the average point is not good. Since the operations and commands useful for detecting the case to be prepared are prepared, the ability of the similar case search is improved.

Further, since the basic mechanism necessary for the construction support environment in the case base construction can be provided, there is an effect of improving the development environment of the case base system construction support system.

[Brief description of drawings]

FIG. 1 is a diagram of an embodiment of the first invention.

FIG. 2 is a diagram of an embodiment of the second invention.

FIG. 3 is a diagram of an embodiment of the third invention.

FIG. 4 is a diagram of an embodiment of the fourth invention.

FIG. 5 is a diagram of an embodiment of the fifth invention.

FIG. 6 is a diagram of an embodiment of the sixth invention.

FIG. 7 is a diagram of an embodiment of the seventh invention.

FIG. 8 is a diagram of an embodiment of the eighth invention.

FIG. 9 is a diagram of an embodiment of the ninth invention.

FIG. 10 is a diagram of an embodiment of the tenth invention.

FIG. 11 is a diagram of an eleventh embodiment of the present invention.

FIG. 12 is a diagram of an embodiment of the twelfth invention.

FIG. 13 is a diagram showing a configuration of general case-based reasoning.

FIG. 14 shows a definition example of a partial matching definition part, in which the definition of partial matching of “month” and “month” is described.

FIG. 15 shows an example of definition of a partial matching definition part, in which the definition of partial matching of “place” and “place” is described.

FIG. 16 is a diagram showing a conventional device.

Claims (12)

[Claims] 1. A device for storing a past solved problem and its solution as an experience, and searching a past similar solution experience when a problem to be solved is given, input to a system and correspondence from the system. A case attribute vector that collectively represents the output, a case base that holds a plurality of case attribute vectors, a problem storage attribute vector that has the same configuration as the case attribute vector that stores the problem to be solved, the case attribute vector, and the problem A partial matching definition unit that defines the degree of matching between corresponding element attribute values of the stored attribute vector; a weight definition unit that defines weights for each element attribute of the case attribute vector and the problem storage attribute vector; 1 for the vector and the problem storage attribute vector
A vector division unit that divides the vector into one or more subvectors, a degree of matching between the values of the corresponding element attributes of the case attribute vector and the problem storage attribute vector is calculated by referring to the partial matching definition unit, and the weight definition is performed. An attribute similarity calculation unit that outputs a value multiplied by a corresponding weight defined for each element attribute, and for each subvector divided by the vector division unit, the element attribute belonging to the subvector A sub-vector similarity calculation unit that outputs the sum of the outputs of the corresponding attribute similarity calculation units, and a total similarity calculation unit that outputs the sum of the outputs of the sub-vector similarity calculation units, which should be solved When a problem is given, it is stored in the problem storage attribute vector, and one case is taken out from the case base and stored in the case attribute vector. A control unit for instructing to calculate the attribute similarity for each element attribute, the subvector similarity and the total similarity, and the attribute similarity for each element attribute and the subvector calculated for each case according to the instruction of the control unit. A case selection / presentation unit that compares the similarity and the total similarity between cases, selects a case with a large value of the total similarity, the subvector similarity, or the attribute similarity, and presents it as a similar case of the problem to be solved. A case-based search system creation support device consisting of. 2. An expression that means "completely matches any value" is allowed as a value of an element attribute in a case attribute vector or a question storage attribute vector, and when the attribute similarity calculation unit receives the expression. The case-based search system creation support apparatus according to claim 1, wherein a function of interpreting the degree of matching between attribute values so as to be output as "completely matched" is added to the attribute similarity calculation unit. 3. As a value of an element attribute in a case attribute vector or a question storage attribute vector, "when there is no complete match, all the similarities between the case attribute vector and the question storage attribute vector are set to 0". When the attribute similarity calculation unit receives the expression, the degree of matching between attribute values is output as “completely matched” if the attribute values are equal to each other,
The method according to claim 1, further comprising a function of interpreting outputs of the attribute similarity calculation unit, the sub-vector similarity calculation unit including the element attributes, and the total similarity calculation unit so as to output the similarity 0. Case-based search system creation support device. 4. As a value of an element attribute in the case attribute vector or the question storage attribute vector, “when not completely matched, the attribute similarity between the case attribute vector and the question storage attribute vector and the target element attribute When the attribute similarity calculation unit receives the above expression, the degree of matching between attribute values is the same, and if the attribute values are the same, Addition of a function of interpreting so as to output as "completely matched", and otherwise output the output of the attribute similarity calculation unit and the subvector similarity calculation unit of the subvector including the element attribute as the similarity 0. The case base search system creation support device according to claim 1. 5. The value of the element attribute in the case attribute vector or the question storage attribute vector is set to “When the values do not completely match, the attribute similarity between the case attribute vector and the question storage attribute vector is set to 0. When the attribute similarity calculation unit receives the expression, the degree of matching between attribute values is output as “completely matched” if the attribute values are equal to each other. If not, the case-based search system creation support apparatus according to claim 1, further comprising a function of interpreting the output of the attribute similarity calculation unit so as to output the similarity as 0. 6. A plurality of the partial matching definition parts can be provided for one element attribute, and the attribute similarity calculation part is used by using one of the plurality of partial matching definition parts specified from the outside. The case base search system creation support apparatus according to claim 1, wherein a function for performing calculation is added. 7. It is possible to have a plurality of the weight defining sections for one element attribute, and the attribute similarity calculating section is used by using one of the plurality of weight defining sections designated from the outside. The case-based search system creation support device according to claim 1, wherein a function for performing calculation is added. 8. For each case stored in the case base,
A case type is stored in the case base where the values of element attributes are all actually existing, and the case definer generalizes the values that actually exist for some element attribute values and stores them in the case base. It is allowed to add generalized case types to things, and common sense types to those that do not have all the element attribute values actually present, and the case selection presentation unit is similar to the problem to be solved given. When presenting the case
The case-based search system creation support device according to claim 1, further provided with a function of presenting the given type at the same time. 9. When defining a plurality of the partial matching definition units, the attribute similarity calculation unit takes out the plurality of partial matching definition units one by one and individually calculates the attribute similarity, The sub-vector similarity calculation unit individually calculates the sub-vector similarity using the attribute similarity calculation results of the element attributes using the partial matching definition units, and the total similarity calculation unit, The total similarity is calculated individually using the subvector similarity calculation results using each partial matching definition part,
The case selection presenting unit displays the calculated attribute similarity calculation result of the element attributes, the sub-vector similarity calculation result, and the total similarity calculation result for each of the plurality of partial matching definition units. The case-based search system creation support device according to claim 6, which is provided with a function of individually outputting. 10. When a plurality of the weight defining sections are defined, the attribute similarity calculating section sets the weight defining section to 1
The subvector similarity calculation unit individually calculates the attribute similarity, and the subvector similarity calculation unit individually calculates the subvector similarity using the attribute similarity calculation results of the element attributes using the respective weight definition units. The total similarity calculation unit individually calculates the total similarity using the subvector similarity calculation results using the respective weight definition units, and the case selection presentation unit is configured to calculate the plurality of weights. Definition part 1
The case base according to claim 7, wherein a function of individually outputting the attribute similarity calculation result of the calculated element attribute, the sub-vector similarity calculation result, and the total similarity calculation result is given to each of them. Search system creation support device. 11. A problem to be solved is given, and similar results of the problem are presented together with the result presented by the case selection / presentation unit, all of them are saved, and when the user changes the partial matching definition unit, The stored problems are taken out one by one, the similarity is calculated, a new similar case sequence is obtained, and the obtained similar case sequence is different from the similar case sequence paired with the stored problem. The case-based search system creation support apparatus according to claim 1, which is provided with a mechanism for notifying the user at times. 12. A problem to be solved is given, and all the similar cases of the problem presented by the case selection / presentation unit are stored together with the problem, and when the user changes the weight definition unit, the When the stored problems are taken out one by one, the similarity is calculated, a new similar case sequence is obtained, and the obtained similar case sequence is different from the similar case sequence paired with the stored problem. The case-based search system creation support device according to claim 1, wherein a mechanism for notifying a user is added to the.