JPH10312288A - Combined analysis system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a data analysis method capable of performing combined analysis for a table form data and data different in form such as a transaction type. SOLUTION: In this system, example data 101 constituted of a single value attribute and a plural values attribute and attribute information 102 are inputted. They are converted from the table form data or the data to be analysed of the transaction form. One or more attributes are set as target attribute examples and explanation attribute value example group to generate rules from the combination of the target attribute example and the example attribute value example group. The rules are evaluated and selected, are indicated to a user or are transmitted to an external system.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] 1. Field of the Invention [0002] The present invention relates to a data analysis technique for examining relationships between data attributes.

[0002]

2. Description of the Related Art There are many data analysis tools. There are many spreadsheet tools that handle so-called spreadsheet data. These implement mathematical calculations between data items. Data mining is also known as a technique for searching for and finding combinations of data items having large correlations and dependencies. For the data mining technology for tabular data, see “Knowledge Discovery Technology from Databases” (Maeda, Systems / Control / Information, Vol. 39, No.
4, pp. 185, 1995). This is to automatically extract a combination of items that have a strong dependency in the data. For the time series analysis, especially the combination analysis technology for transaction data, see "Visualization of Association Rules" (Fukuda, Morishita, IEICE Technical Report, Vol.95,
No. 81, 1995). This automatically extracts a combination of events having a high probability of occurring simultaneously in a transaction.

[0003]

SUMMARY OF THE INVENTION For decision making,
Data analysis using various data is required. For example, in the analysis of customer purchase trends in the sales business, information on static (which does not change on a daily basis) characteristics such as gender and age of customers, and customer purchase records (changes in time series with each purchase) Analysis) using information. Since these data take various forms depending on their properties, they are usually analyzed by different methods. But,
In order to discover potential correlations and dependencies between data, it is necessary to have a method of simultaneously analyzing these different types of data. An object of the present invention is to provide a data analysis method for performing analysis by combining tabular data and data in different formats such as a transaction format.

[0004]

In order to analyze different types of data in combination, there is provided means for generating a set of attribute and attribute value for both single-value attribute data and multi-value attribute data. Further, it has means for setting a purpose attribute and a description attribute. Further, it has a means for combining the purpose attribute and the description attribute to generate a rule. It has means for evaluating each of these rules and means for selecting a rule according to the evaluation result. It has means for presenting the selected rules to the user and transmitting them to an external system. Further, it has means for converting the data to be analyzed into the single-value attribute data and the multi-value attribute data.

[0005]

FIG. 1 shows an example of an embodiment of the present invention. Here, case data 101 and attribute information 102
Is the input. The case data 101 is a set of records. Each record is one of the cases to be analyzed.
It corresponds to one. FIG. 2 shows an example of the case data 101. This is customer data at a retail store, where one case is one customer. Thus, each record corresponds to one customer. Case data 1
01 is composed of one or more fields, and each field corresponds to an attribute related to a case. For customer data, the customer number, gender, age,
This corresponds to an item that describes information about a customer, such as a purchased item. In each field of each record, the attribute value of the corresponding attribute of the corresponding customer is stored. Attribute information 102 is attribute information such as which field corresponds to which attribute and what value each attribute can take. FIG. 3 shows an example of the attribute information. This is a set of information about each attribute. The attribute information includes information such as “attribute name”, “corresponding field”, “single value / multiple value distinction”, and “definition area”. "Attribute name" is a name given to each attribute, and the attributes are distinguished from each other. In the following description, the attribute is indicated by the attribute name. For example, an attribute having an attribute name “sex” is called “sex” attribute. The “corresponding field” is a field number corresponding to each attribute in the case data. "Single value / multiple value distinction" is a symbol indicating whether each attribute is a single value attribute or a multiple value attribute. The single-value attribute and the multi-value attribute will be described later. “Defined area” is a set of attribute values that each attribute can take. For example, the domain of the "customer number" attribute is a numeric string from 00000 to 99999, and the domain of the "sex" attribute is {male, female}.

Here, the single-value attribute and the multi-value attribute will be described. A single-valued attribute is an attribute that cannot take two or more values at the same time for one case. FIG. 4 is an example of a single-valued attribute. In the case of the "gender" attribute, the gender for one case, that is, one customer is only one of "male" and "female" and cannot take two or more values. However, it is possible that the gender is unknown. In such a case, the gender attribute of the customer does not take any value, and the field is blank. On the other hand, a multi-value attribute is an attribute that can take a plurality of values simultaneously for one case. FIG. 5 is an example of a multi-value attribute. "suit",
“Shoes”, “coat” and the like are attribute values of the “purchase item this month” attribute. In the case of the “purchase this month” attribute, one customer may purchase a plurality of products during the period. Also,
The number of products purchased by one customer during the period is undefined. Therefore, a plurality of attribute values are stored in the field corresponding to the “purchase item this month” attribute, and the number is indeterminate. If no attribute value is taken, the field is blank. FIG. 6 is another example of a multi-valued attribute. Here, the stored attribute value is one or blank for any one field of any one record, but there are a plurality of fields corresponding to the “purchase item this month” attribute. The “purchase this month” attribute will take multiple attribute values at the same time. In this case, FIG.
In the "field" section of the attribute information shown in (1), a plurality of field numbers corresponding to the "purchase item this month" attribute are described.

FIG. 7 shows the input / output relationship of the explanation attribute value example group generating means 103. One of the attributes in the attribute information 102
Then, a set of one of the attribute values belonging to the domain of the attribute is referred to as an attribute value example, and is represented by an attribute value name and an attribute value. For example, (sex, male) is an example of an attribute value in which an attribute “sex” and its attribute value “male” are paired. The input of the explanation attribute value example group generation unit 103 is a set of these attribute value examples 701. A set of attribute value examples 701 to be input to the explanation attribute value example group generation unit 103 is particularly called a combination candidate set 702. If no particular condition is set for the combination candidate set 702, the attribute information 10
2 are all elements of the combination candidate set 702. Explanation attribute value example group generation means 1
In step 03, the input attribute value examples are combined to generate and output an attribute value example group 703 having one or more attribute value examples as elements. A plurality of attribute value examples having the same single numerical value attribute cannot be included in the same attribute value example group 703. For example,
Since the “gender” attribute is a single-valued attribute, two attribute value examples (sex, male) and (sex, female) cannot be included in the same attribute value example group. On the other hand, for the multi-value attribute, a plurality of attribute value examples having the same multi-value attribute can be included in the same attribute value example group 703. For example, the “Purchase this month” attribute is a multi-value attribute, so
And (an item purchased this month, court) can be included in the same attribute value example group. When there is no condition for the combination of the attribute value examples other than the condition of the same attribute, if the number of attribute value examples 701 belonging to the combination candidate set 702 is large, the number of possible combinations becomes enormous. Therefore, the condition 104 is set in the way of creating a combination, such as setting the upper limit of the number of attribute value examples which one attribute value example group 703 has as an element to a relatively small number, for example, three. Explanation attribute value example group generation means 103
Generates all possible attribute value example groups 703 within the range of the combination condition 104. Explanation attribute value example group generation means 10
The attribute value example group generated in 3 is referred to as an explanation attribute value example group.

The target attribute value example setting means 105 receives a set of attribute value examples similar to those of the explanation attribute value example group generating means 103 and selects one or more of them as a target attribute value example. The output of the purpose attribute value example setting means 105 is a set of the selected purpose attribute value examples. Which of the input attribute value examples is selected as the target attribute value example is predetermined or designated by the user. Unless otherwise specified, all attribute value examples are selected as target attribute value examples.

A set consisting of one example group of explanatory attribute values and one example of target attribute values is called a rule. The rule generation means 106 receives as input the set of the explanation attribute value example groups 703 output from the explanation attribute value example group generation means 103 and the set of the purpose attribute value examples 801 output from the purpose attribute value example setting means 105, A set of rules generated by combining these is output. FIG. 8 shows an example of input and output of the rule generation means 106. The rule generation means 106 generates a rule 802 that satisfies the rule condition among all possible combinations of the explanation attribute value example group and the purpose attribute value example, and outputs a set of the generated rules 802. FIG.
Shows the conditions for the rules. Regarding single-valued attributes, a description attribute value example group having the same attribute and a target attribute value example cannot be combined. For example, (gender, male)
For the purpose attribute value example, (gender, male)
A description attribute example group having attribute value examples of (sex, female) is not combined (901, 902). Further, with respect to the multi-value attribute, the same attribute value example, that is, the description attribute value example group having the same attribute and the same attribute value and the target attribute value example cannot be combined. For example, for a target attribute value example of (purchase item this month, suit), a description attribute value group having an attribute value example of (purchase item this month, suit) is not combined (903). The explanation attribute value group having the attribute value example of “coat” can be combined because the attribute values are different (904).

FIG. 10 shows the input / output of the rule evaluation means 107. The rule evaluation means 107 outputs the case data 101
And the rule 8 generated by the rule generation means 106
02, and the attribute information 102 are input, each piece of evaluation information 1002 is calculated for all of the input rules 802, and the evaluation information 1002 is added to the rule 802.
Here, what added evaluation information to a rule is called "rule evaluation". And rule evaluation 1001
Output the set of Now, it is assumed that one record R and one attribute value example a: (attr, val) are given. In record R, the set of attribute values stored in the field corresponding to attribute attr is written as VAL (R, attr)
far. VAL (R, att if attribute attr is a single-valued attribute
The number of elements in r) is 1 or 0. Also, the attribute attr
Is a multi-valued attribute, the number of elements in VAL (R, attr) is
It is 1 or more or 0. Here, when VAL (R, attr) = φ, R is said to be "missing" for attribute attr only then. Also, val ∈ VA
If L (R, attr), then only if
And record R are said to "match" each other (FIG. 11). Also, attribute value example group A: {a_i} is given, and all of the attribute value example a_i belonging to A are records R
Attribute value example group if and only if
Let us say that A and record R match each other (FIG. 12). Also, given rule r, record R
Satisfies both the target attribute value example and the description attribute value example group that constitute the rule r, then only when the rule r
And record R match each other (FIG. 13). FIG. 14 shows a procedure for obtaining evaluation information on a rule. Let r be the target rule, t be the purpose attribute value example that constitutes rule r, A be the description attribute value group, and {a_1, a_2, ... a be the attribute value example that constitutes A.
_n}. Where n is the number of example attribute values that make up A. For all records in the case data 101, the number of records that match the explanatory attribute value example group A is counted in processing 1401, and is set as S_A. Processing 14
In 02, the number of records that match the target attribute value example t is counted, and is set as S_t. In the processing 1403, the number of records matching the rule r is counted, and this is set as S_r. In processing 1404, the attribute value examples t and a_i (where 1
Count the number of records that are not missing in any of the attributes of ≤ i ≤ n), and let it be N. When determining the match or the loss, it is necessary to know which field corresponds to which attribute. This can be known by referring to the attribute information 102. In processing 1405, the evaluation value U of the rule r is calculated according to the following equation.

[0011]

U = G · log [P (S_r | S_A) / P (S_A)] where P (S_A) = S_A / N, P (S_r) = S_r / N, P (S_r |
S_A) = P (S_r) / P (S_A). G is S_A, S_
This is a function of t, S_r, and N. Specifically, the following equation is used.

[0012]

[Equation 2] G = S_A / N

[0013]

[Equation 3] G = S_r / S_t

Also,

In Expression 1, a power of G may be used instead of G. In processing 1406, these U, S_A, S_t, S_
r and N are added to the rule r as evaluation information 1002 to generate a rule evaluation 1001 of the rule r. The rule evaluation means 107 generates a rule evaluation 1001 for all the inputted rules 802, and
Output the set of

FIG. 15 shows the input and output of the rule selecting means 108. In the rule selection means 108, the rule evaluation means 10
7, a set of rule evaluations 1001 output from
Based on this as a selection candidate, based on the rule selection condition 109,
A part or all of the rule evaluations that are selection candidates are selected based on the evaluation information of each rule, and a set of selected rule evaluations 1501 is output. FIG. 16 shows an example of a rule selection method. Here, the number of rule evaluations to select, L
Are determined in advance by the rule selection condition 109, and L rule evaluations are selected from the input rule evaluations 1001 in order of the evaluation value.

FIG. 17 shows another example of a rule selection method. Here, when two rules r1 and r2 have the same target attribute value example, and the explanatory attribute value example group A1 of r1 is a subset of the explanatory attribute value example group A2 of r2, the rule r2 is
Let's say that it is "included" by rule r1 (Fig. 1
8). In the rule selection method shown in FIG. 17, with respect to any two rules r1 and r2, if r1 includes r2 and the evaluation value of r2 does not exceed the evaluation value of r1, the rule evaluation of r2 is excluded from the selection candidates. . Then, from among the rule evaluations remaining as selection candidates, L rule evaluations are selected in descending order of the evaluation value. Note that the value of L is determined by the rule selection condition 109
Is predetermined.

FIG. 19 shows another example of a rule selection method. First, the input rule evaluation 1001 is divided into groups according to the purpose attribute value examples. Then, a rule evaluation is selected in each group. As a selection method in each group, a selection method in a simple evaluation value order as shown in FIG. 16 or a selection method in consideration of inclusion of a rule as shown in FIG. 17 is used.

FIG. 20 shows the input and output of the output means 110. The output unit inputs the set of rule evaluations 1501 selected by the rule selection unit 108 and presents the set to the user or shapes it into a form suitable for transmission to an external system. Or transmitted to an external system via a storage medium 2002 or a communication path 2003.

FIG. 21 shows a display example of the screen 2001.
What is displayed here is the entered rule evaluation 15
01 is a table listing the evaluation values in order, and each row represents one rule evaluation. In addition, each column represents a purpose attribute value example, a description attribute value example group, rule accuracy, and rule coverage. Taking the first line as an example, an example of the purpose attribute value is (purchased last month, coat), and a group of explanatory attribute values is (sex, male), (age group, 30s), (purchased last month, suit ), The rule accuracy is 60% and the rule coverage is 40%. Here, the accuracy and the coverage will be described. The accuracy is obtained by the following equation.

[0020]

## EQU4 ## where S_r is the number of records that match the rule, and S_A is the number of records that match the group of explanatory attribute values. Also,
The coverage is obtained by the following equation.

[0021]

## EQU5 ## Coverage = S_r / S_t where S_r is the number of records matching the rule, and S_t is the number of records matching the target attribute value example. Accuracy is
It shows how accurately the example attribute value group describes the purpose attribute value example. For example, a rule that has a purpose attribute value example “purchase court last month” and a description attribute value group “male, 30s, suit purchase last month” that has 60% accuracy means that “rules for men, 30s” 60% of customers who buy suits last month buy coats last month. In other words, it can be interpreted that the explanation that "a customer who is" male, 30s, buying a suit last month "has purchased a coat last month" is 60% correct. The coverage indicates how much of the rule explains the entire purpose attribute value example. For example,
A rule with an objective attribute value example “purchase court last month” and a description attribute value group “male, 30s, suit purchase last month” has a coverage rate of 40%. This means that they account for 40% of the "purchased" customers. The user evaluates the usefulness of the explanation that "the customer" male, 30s, purchasing a suit last month "has purchased a coat last month" while checking the accuracy and coverage. Use for customer trend analysis. Further, the coverage may be obtained by the following equation.

[0022]

## EQU6 ## Coverage = S_A / N where N is the number of records that are not missing for any attribute in the rule. When outputting the rule evaluation to an external system, the selected rule evaluation is stored in an appropriate data structure, and transmitted to the external system via the storage medium 2002 or the communication path 2003.

In the above description, all the attribute information included in the attribute information 102 has been input to the explanation attribute value example group setting means 103. However, some attributes do not make sense to use as example description attribute values. For example, the attribute value of the “customer number” attribute is a serial number assigned to the customer, and it will not be considered that the value is related to the trend or sales of the customer. It is better not to use such an attribute when generating the example attribute value group. FIG. 22 shows a means for designating an attribute used for a group of explanatory attribute values. In the description attribute setting means 2201, the attribute information 102
The attribute to be used by the explanation attribute value example group generation unit 103 and the attribute value are set and output among the attributes possessed by. This setting is predetermined or performed by the user as needed. FIG. 23 shows an example of the operation screen when the user sets the description attribute. The operation screen 2301 displays a list 2302 of attributes included in the attribute information 102 and a list 2303 of attribute values. The user selects and marks one or more attributes to be used as descriptive attributes from the attribute list 2302. In addition, for each of the selected attributes, one or more attribute values to be used as the explanation attribute value examples are selected. The explanation attribute setting unit 2201 outputs a set of the selected attribute and attribute value. For example, if “age” is selected as one of the description attributes and “20s” and “30s” are specified as the attribute values, “20s”, “20s” 2 in the "30s"
Only one attribute is output. Therefore, in the explanation attribute value example group generation means 103, (age group, 20s),
An attribute value example such as (age group, thirties) is used, but an attribute value example such as (age group, forties) is not used, and thus a rule having such an attribute value example is not generated. Also, a list 230 of attribute values on the operation screen 2301
In 3, the initial setting is “all attribute values”, and if no attribute value selection operation is performed, all attribute values of the selected attribute are output.

The above description has been made on the assumption that the case data 101 and the attribute information 102 have already been given. However, in the case of an actual application, the data to be analyzed may not be in such a format. Many. Therefore, a data conversion unit that converts data to be analyzed into case data 101 and attribute information 102 will be described. FIG. 24 is an example of a system configuration to which data conversion means is added. The data conversion means 2402 inputs the analysis target data 2401,
From this, case data 101 and attribute information 102 are generated. Here, taking the case of customer trend analysis in a retail store as an example, what kind of data conversion is performed will be described. The customer trend analysis uses customer attribute information such as gender and age of each customer and purchase information indicating when and what each customer purchased, and performs an analysis by combining the two. The analysis target is a customer. Therefore, both customer attribute information and purchase information are converted into case data in which one record corresponds to one customer.

First, an example of converting customer attribute information into case data will be described. FIG. 25 shows one of the records of the analysis target data. The customer attribute information is a set of attribute information such as gender and age for each customer, and is the same in format as the single-valued attribute of the case data 101 described above. However, conversion may be required depending on the nature of the data and the purpose of the analysis. For example, in order to analyze customer trends by region in Japan, address information of each customer is required. Customer attribute information 250 of FIG.
1 has an address field, which can be used. However, using the address information of the customer attribute information 2501 as an attribute value as it is is too fine. To study regional trends throughout Japan, information at the regional level, such as Tohoku, Kanto, and Chubu, or at the prefectural level would be appropriate. Therefore, the address information of the customer attribute information 2501 is converted into attributes at the regional level and at the prefectural level. In addition, when it is desired to examine customer trends by age, the age information of the customer attribute information 2501 will be used.
In some cases, you may want to investigate by age group. In the example of FIG. 25, the age information of the customer attribute information 2501 is not used as it is as an attribute value, but is converted into an age group attribute such as 20s and 30s. In such a case, the domain of each attribute is clearly known from the conversion method. For example, the domain of the regional attribute is {Hokkaido, Tohoku, Kanto, Chubu, Kinki, ‥‥}
It will be like. The domain of the prefecture attribute is all prefecture names, and the domain of the age group attribute is {10 years old, 10s, 20s, ‥‥}. As a result of the above conversion, a “value range” attribute, a “prefecture” attribute, an “age group” attribute, and the like are generated from the customer attribute information 2501, and attribute information 2503 for these attributes is generated. Also, case data 2502 having fields and attribute values corresponding to these attributes is generated.

The method of converting the address information into the regional attribute and the prefecture attribute, and the method of converting the age information into the age group attribute, that is, the correspondence between the customer attribute information before the conversion and the attribute value after the conversion is described above. Often it is experimentally clear. For example, the address information “Yokohama-ku, Yokohama-shi, Kanagawa”
It is apparent a priori that the prefecture attribute is converted to the attribute value “Kanagawa”. However, the conversion method may not be apparent a priori. In the above example of the age group attribute, a conversion method that is apparent a priori was used, such as “20 years old or older and less than 30 years old → 20s”, but when dividing into age groups, 20 years old, 30 years old, It may not be clear whether a break should be placed at the age of 40. In that case, the conversion method is determined using the characteristics of the original data, such as the age distribution of the customer in the customer attribute information. This example is shown in FIG. FIG. 26 is a histogram 2601 showing the customer age distribution in the original customer attribute information.
According to this, the age distribution of the customers is not uniform, and therefore, it is likely that grouping can be performed based on the age distribution.
Various methods of grouping are also conceivable.
In the example of the above, around the peak around the age of 30, around the peak,
Three groups are formed below the peak and above the peak, and another group over the age of 40 is grouped into one group, for a total of four groups.
These four groups are given names for distinguishing each other, and they are attribute values. Thus, the conversion method 2602 for the age group attribute is determined, and based on this, the data conversion from the customer attribute information 2603 to the case data 2604 is performed, and the attribute information 2605 is generated.
Attributes related to customer attribute information are usually single-valued attributes.

Next, an example of converting purchase information into case data will be described. FIG. 27 shows an example of purchase data. One record corresponds to one purchase of a product, and stores the purchase date and time, the customer number of the purchased customer, the product name, and the price. The purchase data is a series of these records, and is usually arranged in the order of purchase, that is, in the order of purchase date and time. This data can be said to be a series of transactions of purchase. In the case of the customer attribute data in FIG. 25, one record corresponds to one customer, and therefore, information about one customer exists only in one record. Can be. However, in the case of purchase data, one record corresponds to one product purchase, not one customer. Of course, since one customer may purchase a plurality of products, purchase information about one customer exists in a plurality of records. Therefore, in order to convert a record into case data corresponding to a customer, it is necessary to convert the data format. Further, there are various data conversion methods depending on what is focused on and used for analysis in the purchase information.

FIG. 28 shows an example of conversion of purchase data. This is data conversion focusing only on "what you purchased this month". First, an attribute “purchased this month” is generated, and a field corresponding to the attribute “purchased this month” is provided in the case data. Since one customer can purchase a plurality of products, this attribute is a multi-value attribute, and the domain is the product names of all the products. Next, the purchase data 2801 for the target period (in this case, this month) is searched, and the product name of the product purchased by each customer is entered in the field corresponding to the “purchased this month” attribute of the case data. Store. It is possible that one customer purchases the same product more than once during the period, but the same product name is not redundantly stored in one field. Here, attention is paid only to "whether purchased", and the number of times (number) is not a problem.

FIG. 29 shows another example of conversion of purchase data. In the example of FIG. 28, attention is paid to whether or not each product is purchased. However, depending on the purpose of the analysis, there is a case where it is desired to treat not only individual products but also a plurality of products including a time relationship. For example, the analysis is related to a purchase pattern such as "purchased a suit and coat in one month" and "purchased a coat after purchasing a suit". When dealing only with individual products, it is sufficient to arrange the product names in the field. However, in order to handle the purchase pattern, not the individual product but an attribute value representing the purchase pattern itself is used. For example, purchase patterns related to periods such as “purchased a suit and coat in one month” and “purchased a suit and coat in one year” include “suit & coat in one month”, “suit & coat in one month”, and
A symbol such as “coat in 12 months” is used as the attribute value.
Also, for the purchase pattern related to the order, such as "I purchased a coat after purchasing a suit,"
A symbol such as "en coat" is used as the attribute value. Since a single customer can apply to more than one of these purchase patterns, the attributes associated with these purchase patterns are usually multi-valued attributes. in this way,
If a symbol is defined according to the purchase pattern to be expressed, various purchase patterns can be treated as a multi-value attribute. What purchase patterns need to be expressed depends on the purpose of the analysis. The conversion procedure is as follows:
First, a purchase pattern to be focused on is determined (2902). Thereby, a set of possible purchase patterns, that is, a domain of a “purchase pattern” attribute is determined (2903). Next, the purchase data 2901 is searched for an applicable purchase pattern for each customer (2904), and the symbol corresponding to the found purchase pattern is stored in the case data 2905.

If more product sets, complicated time relations, and order relations are to be handled, the number of possible combinations increases, and accordingly, the domain of the attribute values becomes very large. However, the number of possible combinations is large, but only a small part of them actually appear in the purchase data. Therefore, among the possible combinations, those that do not actually appear in the purchase data or those that have a very low frequency of occurrence are deleted from the attribute information and the domain is reduced, so that the subsequent combination generation, evaluation, Perform the selection process efficiently. This is shown in FIG. here,
It is assumed that attention is paid to "a set of products to be purchased within 12 months" (3001). In this case, a symbol “A & B in 12 months” (where A and B are product names) is taken as an attribute value, and its definition area is all possible combinations of two products (3002). Next, from the purchase data 3003,
For each customer, a purchase pattern that matches is searched (3004), and a symbol corresponding to the found purchase pattern is stored in the case data 3005. Next, the purchase pattern that did not appear in the process of searching for the purchase pattern or the purchase pattern with a low appearance frequency is deleted from the domain of the attribute value (3006), and the small attribute information 30 is deleted.
07 is generated. When the purchase patterns having a low appearance frequency are deleted from the definition area, they are not present in the attribute information 3007 but are present in the case data 3005. If necessary, these purchase patterns are also deleted from the case data 3005. In this way, by reducing the domain of the attribute, it is possible to prevent the description attribute example group generating unit 103 and the rule generating unit 106 from generating useless combinations.

As described above, by converting customer attribute information and purchase information into single-value attribute data and multi-value attribute data, case data combining customer attribute information and purchase information can be created. By using this case data as input and using the analysis system shown in FIG. 1, customer trend analysis combining customer attribute information and purchase information becomes possible. FIG. 31 shows an example of the result of the customer trend analysis.
For example, the first of the description columns in FIG. 31 is “In the early 20s,
Forty percent of men who purchased both suits and coats in a month have more than 200,000 purchases this month, which is 60.5% of those who have purchased more than 200,000 this month. " It is shown that.

[0032]

According to the present invention, data combination analysis can be performed on data in which an attribute taking a single value and an attribute taking a plurality of values are mixed at the same time. As a result, data in different formats can be analyzed together, and potential correlations and dependencies between data that cannot be obtained by separate analysis can be found. Further, by converting the information on the attribute of the case and the information on the series of events into single-valued attributes and multi-valued attributes, it is possible to analyze both in combination. This makes it possible to discover correlations and dependencies between static information (attribute information) and dynamic information (event series) relating to cases.

[Brief description of the drawings]

FIG. 1 is a diagram showing a configuration of one embodiment of the present invention.

FIG. 2 is a diagram showing an example of case data.

FIG. 3 is a diagram illustrating an example of attribute information.

FIG. 4 is a diagram illustrating an example of a single-valued attribute.

FIG. 5 is a diagram illustrating an example of a multi-value attribute.

FIG. 6 is a diagram showing another example of a multi-value attribute.

FIG. 7 is a diagram illustrating input / output of an explanation attribute value example group generation unit.

FIG. 8 is a diagram illustrating input / output of a rule generation unit.

FIG. 9 is a diagram illustrating an example of a combination of example attribute values in a rule generation unit.

FIG. 10 is a diagram illustrating input / output of a rule evaluation unit.

FIG. 11 is a diagram showing an example of a match between a record and an attribute value example.

FIG. 12 is a diagram illustrating an example of a match between a record and an attribute value example group.

FIG. 13 is a diagram showing an example of a match between a record and a rule.

FIG. 14 is a diagram illustrating a procedure for calculating an evaluation value of a rule.

FIG. 15 is a diagram illustrating input / output of a rule selection unit.

FIG. 16 is a diagram illustrating an example of a method of selecting rules in the order of evaluation values.

FIG. 17 is a diagram illustrating an example of excluding a rule included in another rule from candidates.

FIG. 18 is a diagram illustrating an example of a rule inclusion relation.

FIG. 19 is a diagram illustrating an example of a method of selecting a rule for each group according to an example of an objective attribute value.

FIG. 20 is a diagram illustrating input / output of an output unit.

FIG. 21 is a diagram illustrating an example of a screen displaying an analysis result.

FIG. 22 is a diagram illustrating input / output of an explanation attribute setting unit.

FIG. 23 is a diagram illustrating an example of a method of setting an explanation attribute.

FIG. 24 is a diagram illustrating input / output of a data conversion unit.

FIG. 25 is a diagram illustrating an example in which customer attribute information is converted into single-value attribute data.

FIG. 26 is a diagram illustrating an example of determining a conversion method based on characteristics of data.

FIG. 27 is a diagram illustrating an example of purchase data.

FIG. 28 is a diagram showing an example in which purchase data is converted into multi-value attribute data.

FIG. 29 is a diagram showing an example in which a purchase pattern in purchase data is converted into multi-value attribute data.

FIG. 30 is a diagram illustrating an example in which a purchase pattern that does not appear in purchase data is deleted from the attribute definition area.

FIG. 31 is a diagram illustrating an example of display of a result of customer analysis.

[Explanation of symbols]

101: case data, 102: attribute information, 103: explanation attribute value example group generation means, 104: combination condition, 105: purpose attribute value example setting means, 106: rule generation means, 107
... Rule evaluation means, 108 ... Rule selection means, 109 ...
Rule selection condition, 110 ... output means.

Claims (5)

[Claims] 1. A set of records, each of which corresponds to one of the cases to be analyzed, each of said records being composed of one or more fields, each of said fields being associated with a case. Data corresponding to one of the attributes and having one or less or a plurality of attribute values in each of the fields for each of the records is set as analysis target data, and a set of attributes related to the case is set as an attribute candidate set. Input, and one of the attributes belonging to the attribute candidate set
And a combination generating means for generating one or more attribute value example groups each including one or more attribute value examples, and one or more attribute value examples. A purpose setting means for setting one of the attributes and one of the possible attribute values of the attribute as an attribute value example, and setting one or more of the attribute value examples as a target attribute value example; The attribute value example group generated by the combination generating means is an explanation attribute value example group, a set of one explanation attribute value example group and one target attribute value example is a rule, and one or more rules are generated.
It has combination evaluation means for calculating an evaluation value for each of the rules by a predetermined algorithm, and has combination selection means for selecting the rule according to the evaluation value, and outputs the selected rule to the outside. Or a combination analysis system having output means for presenting to a user. 2. The combination analysis system according to claim 1, wherein an attribute relating to the case to be analyzed is generated based on data relating to the case to be analyzed, and a set of possible attribute values of the attribute is determined. A combination analysis system including an attribute generation unit that determines an attribute value of the attribute for each of the cases to be analyzed. 3. The combination analysis system according to claim 1, further comprising means for designating a set of attribute candidates by a user. 4. The objective setting means according to claim 1, wherein:
A combination analysis system having means for designating a target attribute value example by a user. 5. A set of records according to claim 1, wherein data storing attribute information relating to a case to be analyzed and data which is a series of events relating to the case are input. Each of the records corresponds to one of the cases to be analyzed, each of the records is composed of one or more fields, each of the fields corresponds to one of the attributes for the case, and for each of the records 2. The combination analysis system according to claim 1, further comprising means for converting the data into a format in which one or less or a plurality of attribute values are stored in each of the fields.