JPH11250084A - Data mining device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To set diverse preprocessing procedures, to easily repeat trial and error on various conditions in a mining process and to simplify various kinds of setting by designating a procedure file to be used in a processing executing means and the order of its application. SOLUTION: In a preprocessing setting part 24, a user prepares the procedure file 312 and selects the file 312 to be used. Next, a correlation rule is extracted from an item database 304 by mining execution 22. The parameter of mining and the item designation of the condition part and the result part of the correlation rule are set by a mining setting part 25 by the user in advance. Finally, the extracted correlation rule is picture-displayed by the display of a result 23. At this time, the correlation rule including only some item is sampled or data some correlation rule can be applied to is referred to advance analysis. When the result is not satisfactory, the settings 24 and 25 are executed again.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] The present invention relates to a data mining apparatus for extracting an association rule from a relational database.

[0002]

2. Description of the Related Art As one method of data mining for finding regularity from a large amount of data, there is extraction of an association rule. An association rule is a rule in the form of “A → B”. “A record containing A in the database is
In many cases. " Hereinafter, A is referred to as a condition part, and B is referred to as a conclusion part.

In order to process a database with a mining system, preprocessing such as discretization for converting a numerical attribute into a symbolic value is required. A method of configuring a data mining apparatus including such preprocessing is particularly described. Kaihei 8-77
No. 010, "Data analysis method and apparatus" (hereinafter referred to as "prior art").

[0004]

When performing data mining, it is rare that all useful rules can be extracted by one execution, and trial and error must be repeated many times by changing preprocessing and mining settings. Often must.

The prior art has shown an integrated data mining device that includes preprocessing. However, the processing performed in the pre-processing is only discretization of numerical attributes and processing of missing values, except for the constraint setting of the generated rules. Processing functions such as attribute value grouping and record selection satisfying specific conditions are not available. Did not. For this reason, fine-grained preprocessing is impossible, and therefore, there is no function for efficiently performing trial and error of preprocessing application using abundant preprocessing procedures.

A first object of the present invention is to solve the above-mentioned problems, to enable abundant setting of preprocessing procedures, and to easily repeat trial and error by changing conditions little by little in a mining process. Another object of the present invention is to provide an apparatus which can easily perform various settings.

[0007] In addition, in the display of association rules obtained by mining, it is often the case that association rules that include only a certain item are extracted and viewed. Often, the numbers are found to be too large to be examined by hand, and the process must be repeated. In the related art, there is no function of displaying information on an association rule including an item for each item in advance.

[0008] A second object of the present invention is to solve the above-mentioned problem and provide an apparatus which allows a user to easily determine which item of association rule is to be extracted and viewed in the association rule display. To provide.

[0009]

According to a first aspect of the present invention, there is provided a pre-processing execution means for inputting a relational database and outputting an item database composed of a plurality of records formed by collecting a plurality of records in accordance with a user's specification. And mining executing means for inputting an item database, extracting a correlation rule between the items of the item database, and outputting the result as a result rule file, and result displaying means for displaying the contents of the result rule file. In a data mining apparatus, a procedure file for recording a procedure for converting from a relational database to an item database is input, and a procedure for discretizing numerical attributes, a procedure for grouping attribute values, and a procedure for Procedures for replacing with values, procedures for deleting attributes Come, procedures for the selection of specific conditions are met records, procedures for itemized to grant the attribute information in the attribute values, procedures for grouping of items,
Procedure file editing means for editing such as addition, deletion or change of any of the procedures, and one or more procedure files to be used in the preprocessing execution means and a procedure file application setting for specifying the order of application Means,
And a pre-processing setting means having

According to a second aspect of the present invention, the data mining apparatus of the present invention has a database analyzing means for inputting a relational database and outputting a data dictionary in which attribute names and attribute value information in the relational database are described for each attribute. , A data dictionary, and one or more ordered procedure files already edited specified by the procedure file application setting unit, and outputting an intermediate data dictionary as a result of applying the procedure file to the data dictionary The data dictionary conversion means, the data dictionary or the intermediate data dictionary, and the attribute hierarchical structure information which describes the relation between the attributes of the relational database as a hierarchical structure are input, and the attribute not included in the intermediate data dictionary is extracted from the attribute hierarchical structure. A display attribute hierarchy generator that deletes and outputs a hierarchical structure in the form of adding attribute values below each attribute DOO further comprising a pre-processing setting means is arranged to further include a processing time attribute hierarchy display means before the display attribute hierarchy generating means for displaying an output hierarchy structure.

According to the third invention, the pre-processing attribute hierarchy display means of the data mining apparatus of the present invention allows the user to distinguish between an attribute group, an attribute, and an attribute value when displaying a hierarchical structure. As described above.

According to the fourth invention, the procedure file editing means of the data mining apparatus of the present invention is configured so as to enable editing of a procedure relating to a node selected by the user by the pre-processing attribute hierarchy display means. You.

According to the fifth invention, the procedure file editing means of the data mining apparatus of the present invention is characterized in that the user selects a certain node in the pre-processing attribute hierarchy display means and performs a procedure or an attribute relating to discretization of numerical attributes. A similar procedure is automatically added to all the attribute nodes under the node based on the additional procedure performed for the deletion of.

According to the sixth aspect, the preprocessing hierarchical attribute display means of the data mining apparatus of the present invention, when displaying the hierarchical structure, highlights a node corresponding to each procedure in the procedure file being edited. It is configured to

According to the seventh invention, the preprocessing setting means of the data mining apparatus of the present invention comprises: a first procedure file including a procedure relating to discretization of a numerical attribute; a grouping of attribute values; A second procedure file that includes procedures related to replacement with null values, deletion of attributes, and selection of records that satisfy specific conditions, and a second procedure file that depends on the first procedure file; itemization for adding attribute information to attribute values; and item grouping. A third procedure file is created which includes procedures related to the first procedure file and the second procedure file, and is applied to the relational database in the order of the first procedure file, the second procedure file, and the third procedure file. It is configured to guarantee that

According to the eighth invention, the pre-processing setting means of the data mining apparatus of the present invention comprises the pre-editing first, second, and third procedure files of the procedure files having correct dependencies among the first, second, and third procedure files. It is configured to include a procedure file application setting unit capable of selecting only a set.

According to the ninth aspect, the data mining apparatus of the present invention comprises: a mining attribute hierarchy display unit for displaying the hierarchical structure output by the display attribute hierarchy generation unit; , May appear only in the condition part of the association rule, or may appear only in the conclusion part of the association rule, or may appear in the condition part of the association rule, or must not appear in the association rule And a condition part conclusion part setting means for designating any one of the four types of modes. The condition part conclusion part setting means selects a node of the mining attribute hierarchy display means and designates the mode. Thus, all nodes under the node are automatically designated as the same mode.

According to the tenth aspect, the attribute hierarchy display means at the time of mining of the data mining apparatus of the present invention makes it possible for the user to distinguish the display form of each node by specifying the mode by the condition part conclusion part setting means. And so on.

According to the eleventh aspect, the attribute hierarchy display means at the time of mining of the data mining apparatus according to the present invention includes, for each node, the node and the subordinate of the node by the designation of the mode by the condition part conclusion part setting means. When the modes of the nodes are all the same, the node is configured to be highlighted.

According to the twelfth aspect, the result display screen of the data mining apparatus of the present invention is configured to display all items and the number of association rules including each item.

[0021]

Embodiments of the present invention will be described below with reference to the drawings. In the present embodiment, a case will be described in which association rules are extracted from a health examination database.

FIG. 1 shows an example of a health examination database.
This database is composed of a plurality of records of each person. Each person's record contains general information such as name ID, gender, age, etc., body measurement data such as height, medical questionnaire response data, precise diagnosis data, etc. Is included. For example, ID00
Looking at the record of the person No. 01, the gender is “male”, the age is “22”, the height is “159”, the grip strength (right) is “44”,
Data such as "always" is stored in the column of stiffness of the questionnaire response data. By processing a plurality of data in such a database with a data mining device, a user can extract the relationship between lifestyle and health status and use it for disease prevention. Although there are various data mining methods, the data mining apparatus extracts an association rule. For example, an association rule such as “high obesity → high blood pressure” is extracted from the health examination database. This association rule means that people with high obesity often have high blood pressure.

FIG. 2 shows the flow of the data mining process. In the data mining process, first, an item database 304 is created from the relation database 301 in which the records to be analyzed are stored by the preprocessing execution 21. The item database 304 is different from the relational database 301 in that a plurality of records of indefinite length each consisting of a set of items are collected. In the preprocessing, processes such as discretization of continuous values, narrowing down of data, and itemization for adding attribute information to attribute values are performed, and details thereof will be described later. The contents of the pre-processing are set in advance by the user using the pre-processing setting 24. In the preprocessing setting 24, the user creates a procedure file and selects a procedure file to be used. Next, an association rule is extracted from the item database 304 by the mining execution 22. The user sets the mining parameters and the item designation of the condition part and the conclusion part of the association rule in advance by the mining setting 25. Finally, the extracted association rules are displayed on the screen by the result display 23. At this time, the analysis is advanced by extracting and seeing an association rule including only a certain item or referring to data to which a certain association rule applies. If the result is not satisfactory, the pre-processing setting 24 and the mining setting 25 are redone and the execution is repeated.

FIG. 3 is a detailed block diagram of the entire data mining apparatus shown in FIG. FIG. 4 is a detailed block diagram of the preprocessing setting unit 311 of FIG.
FIG. 5 is a diagram showing a detailed block diagram of the mining setting means 313 of FIG.

The execution of the pre-processing will be described with reference to FIGS. 2 and 3. Each stored procedure file of the procedure file 312 is composed of one or a plurality of procedures. Each procedure is composed of a character string describing the conversion method for the database. There are three types of procedure files: a first procedure file, a second procedure file, and a third procedure file.

The first procedure file includes a procedure relating to the discretization of the numerical attribute. The discretization of the numerical attribute means that the range of the numerical attribute is divided into several regions by boundary values, and each divided region is given a different name to obtain a new attribute value. If the number of records containing a certain item is small relative to the whole, a correlation rule that includes the item is not extracted due to the nature of the algorithm for extracting the association rule, so a discretization procedure is required.

The second procedure file includes procedures for narrowing down data, that is, grouping attribute values, replacing attribute values with null values, deleting attributes, and selecting records satisfying specific conditions. The grouping of attribute values is to collect several attribute values into one attribute value. Replacing an attribute value with a null value means deleting an attribute value. When the ratio of the number of records including a certain item to the whole is large, an enormous number of association rules including the item are generated due to the nature of the algorithm for extracting the association rule.
To prevent this, the attribute value is deleted.

The third procedure file includes a procedure relating to itemization for adding attribute information to attribute values and grouping of items. If the itemization is not applied, the attribute information is not reflected.
A correlation rule that should be “high blood pressure” is output in a format such as “high → high”. Therefore, in order to output a meaningful association rule, itemization is essential. By applying the itemization procedure to the database, an item database composed of a plurality of records of indefinite length is generated.

In FIG. 3, when the user 318 issues an instruction to execute pre-processing, the pre-processing executing means 302 selects the first, second, and second pre-selected by the procedure file application setting means 44 shown in FIG. The three-procedure file is sequentially applied to the relational database 301 in this order to the item database 30.
4 is generated.

FIG. 6 is a diagram showing a flow of a pre-processing execution process in this embodiment. In FIG. 6, at step 61, data is read from the relational database 301, at step 62, the read data is discretized, at step 63, data is narrowed down, at step 64, data items are formed, and at step 65, data items are formed. Generate the database 304. At this time, it is possible to set so that not only the item database 304 to which all the procedure files have been applied but also the intermediate database 303 to which each procedure file has been applied is stored. in this case,
In step 66, an intermediate database 303 is generated from the data discretized in step 62, and
In step 7, the data narrowed down in step 63 is generated as the intermediate database 303. These intermediate databases 303 have the same structure as the relational database 301.

According to such pre-processing execution, when re-executing pre-processing with slightly changing conditions, it is only necessary to edit the existing procedure file and apply it again, so that trial and error can be easily performed. This has the effect. In addition, by providing a data mining device having a framework for sequentially applying the above three types of procedure files,
There is an effect that necessary and sufficient preprocessing can be performed.

Further, by saving the intermediate database 303 after application of each procedure file, the input to the pre-processing execution means 302 is replaced with the intermediate database 303 instead of the relational database 301, so that the already executed processing can be performed again. Can be omitted.

[0033] Examples of procedures file Next, an example of a first procedure file in FIG. In FIG. 7, each line corresponds to one procedure. For example, the procedure 71 on the first line of the first procedure file 70 is
"Attribute <age> is bounded by <30><40><50> and attribute value <30><30-40><40-50><50
The above is discretized into>. " By the procedure processing of the first line, the attribute <age> is 30, 40, 5
With the boundary point of 0 as a branch point, the attribute value is <30 or less><3
0-40><40-50><50 or more>. The same applies to other procedure lines. As in the procedure 72 and the procedure 73, a discretization procedure that takes a different boundary value depending on the value of another attribute can also be supported. The procedure 72 on the second line may be, for example, “set the attribute <height> under the condition <sex = male>
0>, it is discretized into attribute values <low><medium><high>". By the procedure processing in the second line, the attribute <height> is discretized into the attribute values <low>, <medium> and <high> under the condition of <sex = male> with the boundary points of 160 and 180 as the branch points. . The procedure 73 on the third line is, for example, “condition <
Under gender = female> attribute <height> with boundary value <150><1
70>, the attribute values are discretized into <low>, <medium>, and <high>. " By the procedure processing of the third line, <
Under the condition of gender = female, the attribute <height> is discretized into attribute values <low>, <medium> and <high> with the boundary point between 150 and 170 as a branch point. The procedure 74 on the fourth line may be, for example, “set the attribute <grip strength (right)> to the boundary value <30><40><50><6
0> and attribute value <30 or less><30-40><40-50
><50-60><60 or more>". By the procedure processing of the fourth line, the attribute <
The grip force (right)> is defined as the branch point at the boundary point of 30, 40, 50, 60, and is <30 or less><30-40><40-50>
<50-60> Discretized to <60 or more>.

FIG. 8 shows an intermediate database obtained as a result of applying the first procedure file of FIG. 7 to the relational database of FIG. 8, in the procedure 71 on the first line in FIG. 7, the attribute <age> is <30 or less><30-
40><40-50><50 or more>
In the procedures 72 and 73 on the third and third rows, the attribute <height> is discretized into <low>, <medium> and <high>. In the procedure 74 on the fourth row, the attribute <grip strength (right)> is <30 or less. ><30-40><40-50><50-60><60
It can be seen that only the discretized results are displayed in the above.

FIG. 9 is a diagram showing an example of the second procedure file. In FIG. 9, a procedure 91 on the first line is an example of a procedure for deleting an attribute. The procedure 91 is defined as “deleting the attribute <ID>”. The procedure 92 on the second line is an example of a procedure for selecting a record satisfying a specific condition. The procedure 92 is defined as “select the attribute value <low><high> of the attribute <height>”. The procedure 93 on the third line is an example of a procedure for deleting an attribute. The procedure 93 is defined as “deleting the attribute <gripping strength (right)>”. The procedure 94 on the fourth line is an example of a procedure for grouping attribute values. In the procedure 94, the attribute value of “attribute <shoulder” <always>
<Sometimes> are grouped into attributes <yes>. " The procedure 95 on the fifth line is an example of a procedure for replacing attribute values with null values. The procedure 95 is defined as “deleting the attribute value <No> of the attribute <shoulder>”.

FIG. 10 shows an intermediate database obtained as a result of applying the second procedure file of FIG. 9 to the intermediate database of FIG. In FIG. 10, in the procedure 91 on the first line in FIG. 9, all the fields of the attribute <ID> are deleted, and in the procedure 92 on the second line, the attribute <height>
Only the attribute values <low> and <high> are selected, the attribute <grip strength (right)> is deleted in the procedure 93 on the third line, and the attribute value of the attribute <shoulder> is deleted in the procedure 94 on the fourth line. <Always><sometimes> are grouped into attribute <yes>
In the row procedure 95, the attribute value <attribute <
It can be seen that only the result in which “No” was deleted is displayed.

FIG. 11 is a diagram showing an example of the third procedure file. In FIG. 11, the procedure 111 on the first line is
It is defined as "itemize attribute value <male> of attribute <gender> to <sex: male>". The procedure 112 on the second line is
It is defined that "attribute value <female> of attribute <gender> is itemized as <sex: female>". The procedure 113 on the third line is
It is defined as "attribute attribute value <male> of attribute <sex> and attribute value <30 or less> of attribute <age> to <youth>." The procedure 114 on the fourth line is defined as “itemize attribute value <30 or less> of attribute <age> to <age: 30 or less>.

FIG. 12 shows an intermediate database obtained by applying the third procedure file of FIG. 11 to the intermediate database of FIG. In FIG. 12, in the procedure 111 on the first line in FIG. 11, the attribute value of the attribute <sex>
Man> is itemized as <sex: man>, and procedure 11 in the second line is
2, the attribute value <female> of the attribute <sex> is itemized as <sex: female>, and in the procedure 113 on the third line, the attribute value <male> of the attribute <sex> and the attribute value of the attribute <age><3
<0> is itemized as <Youth>, and procedure 11 in the fourth line is
4, it can be seen that only the result in which the attribute value <30 or less> of the attribute <age> is itemized as <age: 30 or less> is displayed.

Selection of Procedure File The process of selecting a used procedure file in the preprocessing setting 24 of FIG. 2 will be described. Basically, the procedure file depends on the previous procedure file. That is, the second
The procedure file depends on the first procedure file, and the third procedure file depends on the second procedure file. For example, when the attribute value <age> is discretized into the attribute value <20 or less and the attribute value <20 or more> in the first procedure file, the attribute value <3
This procedure will not be applied normally. Attribute value <20 or less> in the first procedure
This is because there is no attribute value of <30 or more> because the attribute value is discretized to <20 or more>.

FIG. 13 shows an example of a dependency relationship between procedure files.
Shown in In FIG. 13, the third procedure files AAA and AAB depend on the second procedure file AA, the third procedure files ABA and ABB depend on the second procedure file AB, and the third procedure file BAA depends on the second procedure file BA. Dependent, the second procedure files AA and AB are dependent on the first procedure file A, and the second procedure file BA is dependent on the first procedure file B. Such a file dependency is stored in the file management unit 31 of FIG.
7 is managed.

FIG. 14 is a diagram showing a selection screen for selecting a usage procedure file. In FIG. 14A, the file A is selected in the first procedure file list box 141, the files AA and AB having a dependency relationship with the file A are displayed in the second procedure file list box 142, and the file AA is further selected. Then, in the third procedure file list box 143, files AAA and AAB that have a dependency relationship with the file AA are displayed.

FIG. 14 (b) shows a case where file AB is selected in the second procedure file list box of FIG. 14 (a).
Shown in In FIG. 14B, the file A is selected in the first procedure file list box 141, the files AA and AB having a dependency relationship with the file A are displayed in the second procedure file list box 142, and the file AB is further selected. Then, in the third procedure file list box 143, files ABA and ABB that have a dependency relationship with the file AB are displayed.

FIG. 14 shows a case where file B is selected in the first procedure file list box of FIG.
It is shown in (c). In FIG. 14C, a file B is selected in the first procedure file list box 141, and a file BA having a dependency with the file B is displayed in the second procedure file list box 142, and a file BA is further selected. In the third procedure file list box 143, a file BAA having a dependency relationship with the file BA is displayed. By adopting the above-described method, it is possible to guarantee the application of a set of procedure files having correct dependencies.

Mining Execution and Mining Setting When the preprocessing execution 21 of FIG. 2 is completed and the item database 304 to be mined is generated, the mining execution 22 is performed next. Prior to the execution of mining, the user performs mining setting 25 in advance. FIG. 5 is a diagram showing details of the mining setting. In FIG. 5, a parameter setting means 53 sets mining parameters, and a condition part conclusion part setting means 52 sets items of a condition part and a conclusion part of an association rule. The mining attribute hierarchy display means 51 displays the mining attribute hierarchy for the user 318 in response to an instruction from the display attribute hierarchy generation means 314. The mining parameters depend on the algorithm used.

When the user 318 issues an instruction to execute mining, the mining executing means 305 shown in FIG. 3 executes the association rule extraction processing with the item database 304 as an input, and outputs the result as a result rule file 306. The association rule extraction processing can be executed by a known conventional technique. For example, an association rule such as “age: 50 or more → shoulder: yes” is extracted from the item database 304 of FIG.

Result Display FIG. 15 shows a result display screen displayed by the result display means 316 of FIG. Although the display in the association rule list box 151 is omitted here, it is assumed that an enormous number of association rules are displayed. After selecting an item in the item list box 152, radio buttons 153, 1
By selecting any one of the condition part, the conclusion part, the condition part, and the conclusion part at 54 and 155 and clicking the narrow down button 156, the association rule including the item is displayed in the association rule list box 151. Here, in the item list box 152, the number of association rules including the item in the condition part is stored in the condition part column 157 corresponding to each item, and the number of association rules including the item in the conclusion part is stored in the conclusion part column 158. The number of association rules is displayed. The number of correlation rules corresponding to each item is calculated when the content of the result rule file 306 in FIG. 3 is read when the result display screen is started. FIG.
5 indicates that, for example, the number of association rules including the item <sex: male> in the condition part is 258, and the number of association rules including the item <gender: male> in the conclusion part is 130. Since the number of association rules that include each item is displayed,
There is an effect that the user 318 can select whether or not to actually narrow down by referring to it.

The pre-processing setting Next, the pretreatment set 24 of FIG. 2, a description will be given of a process to create a procedure file. FIG. 16 shows a first procedure file editing screen. Here, a case where the first procedure file is edited for the relational database 301 of FIG. 1 is shown. The attribute hierarchy display portion 161 includes:
Attributes in the relational database 301 are displayed according to an attribute hierarchy set in advance by the user.

FIG. 17 is a diagram showing the meaning of each icon. Here, the attribute group is a superordinate concept of the attribute. As described above, in the present embodiment, the attribute value group,
Since the display mode of each node corresponding to the attribute and the attribute value can be changed, there is an effect that the user can easily understand.

In FIG. 16, a check box 162
Is assigned to each attribute node, and specifies whether or not the attribute is a target of creation of the discretization procedure. If it is checked, it means that a discretization procedure is to be created. If it is not checked, it means that it is not a discretization procedure creation object. Non-numeric attributes are automatically considered out of the discretization procedure creation target, and the check box is disabled. By default, all numerical attributes are targeted for creation of the discretization procedure, but can be changed by the user.

Next, a procedure editing method will be described. The target procedure in the first procedure file editing is a discretization procedure. As shown in FIG. 4, the procedure file generating means 41 creates an empty first procedure file and stores it in the procedure file 312. In the attribute hierarchy display portion 161 of FIG. 16, the user 318 first selects a node using input means such as a mouse. After that, when the discretization button 163 is clicked, a dialog for setting the discretization method appears. The discretization method includes the method of dividing the attribute value equally using the upper and lower limits of the attribute value, the method of setting the boundary value so that the frequency is counted and the frequency equal, and the method in which the user sets the boundary value list. The method of giving is supported. Further, for example, a discretization procedure having a conditional part such as procedures 71 and 72 in FIG. 7 can also be set.
If an attribute node is selected and a discretization procedure is created, a discretization procedure for the attribute is created.If an attribute group node is selected and the discretization procedure is performed, a discretization procedure is created below the attribute group. Discretization procedures are created for all the attributes for which certain discretization procedures are to be created. When a new discretization procedure is added to an attribute for which a discretization procedure already exists, the old discretization procedure is replaced with a new discretization procedure.

As described above, in the present embodiment,
Since the procedure is generated by selecting a node, there is an effect that the procedure can be easily edited. further,
Since a procedure can be created collectively for a plurality of attributes by specifying an attribute group, there is an effect that the labor for creating the procedure can be greatly reduced especially when the number of attributes is large.

When an attribute having a discretization procedure is selected in the attribute hierarchy display portion 161 of FIG. 16, a discretization procedure relating to the attribute is displayed in a procedure list box 164. After selecting the discretization procedure in the procedure list box 164, the user can click the delete button 165 to delete the procedure, and click the change button 166 to change the procedure. The attribute for which the discretization procedure has been created is highlighted in the attribute hierarchy display portion 161. Therefore, there is an effect that the editing state is easy to understand.

For example, the state after selecting the node corresponding to the attribute <height> in FIG. 16 and creating the procedure is as shown in FIG. In FIG. 18, since the node corresponding to the attribute <height> is selected, the node 182 is highlighted and, for example, “attribute <height> is added to the boundary value <160><180>, the attribute value is discretized to <low><medium><high>."

Next, a method of displaying the attribute hierarchy display portion 161 in FIG. 16 will be described. Only when the data dictionary 308 does not exist, the database analysis unit 307 of FIG.
08 is created. The data dictionary 308 is a file in which attribute names of the relational database 301 and attribute value information for the attributes are described for each attribute. As attribute value information,
A list of attribute values appearing in the relational database 301, an attribute value type, and the like are required. On the other hand, the user 318 creates in advance attribute hierarchical structure information 315 that expresses the relation between all the attributes in the relation database 301 as a tree structure. When creating the attribute hierarchical structure information 315, attribute information described in the data dictionary 308 can be used.

FIG. 19 is a diagram showing an example of the attribute hierarchy structure information 315. In FIG. 19, all data are “interview”,
"Physical examination" is classified into "physical examination", etc. Further, "interview" is classified into "stiff shoulders", ..., etc. "Physical examination" is "height", "grip strength (right)", .. Are generated.

The display attribute hierarchy generating means 314 in FIG.
The attribute hierarchical structure information 315 is output as it is. Then, the pre-processing attribute hierarchy display means 43 of FIG. 4 displays the tree structure output by the display attribute hierarchy generation means 314 of FIG. The attribute hierarchy display portion 161 in FIG. 16 is a display example. In FIG. 16, the result of the user's operation is interpreted by the procedure file editing means 42 in FIG. 4, and the procedure file 312 is actually edited.

FIG. 20 shows a second procedure file editing screen. Since the contents of the second procedure file depend on the contents of the first procedure file, it is necessary to select the first procedure file in advance. Here, the case where the first procedure file of FIG. 7 is selected is shown. In the attribute hierarchy display portion 201 of FIG. 20, the attributes in the relational database 301 are displayed according to the attribute hierarchy set by the user 318 in advance. The attribute value for each attribute is displayed below each attribute.

As for the editing of the procedure, one or a plurality of nodes are selected in the attribute hierarchy display part 201 and the delete button 202 and the select button
3. This is performed by clicking any one of the attribute value grouping buttons 204. However, what can be clicked when the attribute group node is selected is the delete button 202.
Only. When an attribute group node is selected and the delete button 202 is clicked, an attribute deletion procedure is generated for all the attributes below the node. In addition, the deletion or change of the existing procedure can be performed in the same manner as in the above-described editing of the discretization procedure.

Here, a method of displaying the attribute hierarchy display portion 201 in FIG. 20 will be described. The data dictionary conversion means 309 of FIG. 3 refers to the first procedure file,
An intermediate data dictionary 310 is created by adding an attribute value list for each numerical attribute to be discretized to the data dictionary 308.
If the intermediate data dictionary 310 to which the first procedure file has been applied already exists, the above processing is not performed.

The display attribute hierarchy generating means 314 of FIG.
The intermediate data dictionary 310 and the attribute hierarchical structure information 315 are input, and a tree structure is created in which an attribute value relating to the attribute is added below each attribute in the attribute hierarchical structure. And FIG.
The pre-processing attribute hierarchy display means 43 displays the tree structure output by the display attribute hierarchy generation means 314 in FIG. FIG.
The attribute hierarchy display portion 201 of 0 is a display example of a tree structure output by the display attribute hierarchy generation unit 314.

FIG. 21 shows a third procedure file editing screen. Since the contents of the third procedure file depend on the contents of the first procedure file and the second procedure file, it is necessary to select the first procedure file and the second procedure file in advance. Here, FIG. 21 shows a case where the first procedure file of FIG. 7 and the second procedure file of FIG. 9 are selected. In the attribute hierarchy display part 211 of FIG. 21, the attributes and attribute values related to the database after the first procedure file 70 and the second procedure file 90 are applied to the relational database 301 are the attribute hierarchy set by the user in advance. It is displayed according to.

Since the itemization procedure is required for all attribute values, it is assumed that a default itemization procedure is generated for all attribute values when the third procedure file is generated. This is the procedure file generating means 41 of FIG.
Is performed by referring to the intermediate data dictionary 310.
In this case, in the attribute hierarchy display portion 211 of FIG.
Highlight the node corresponding to the procedure changed from the default itemization procedure.

As in the editing of the discretization procedure, one or a plurality of nodes are selected in the attribute hierarchy display portion 211 in FIG. 21 and either the itemization button 212 or the itemization grouping button 213 is edited. This is done by clicking. In addition, the deletion or change of the existing procedure can be performed in the same manner as in the above-described editing of the discretization procedure.

Here, a method of displaying the attribute hierarchy display portion 211 of FIG. 21 will be described. The data dictionary conversion means 309 of FIG. 3 refers to the first procedure file,
An attribute value list of each numerical attribute to be discretized is added to the data dictionary 308, and by referring to the second procedure file, processing such as deletion of attributes and attribute values, conversion of attribute values, and the like is performed to create the intermediate data dictionary 310. create. When the intermediate data dictionary 310 after the application of the first procedure file 70 already exists, the intermediate data dictionary 310 may be created by applying the second procedure file 90 to this. Also, the intermediate data dictionary 3 already applied with the first and second procedure files
If 10 exists, the above processing is not performed.

The display attribute hierarchy generating means 314 of FIG.
The intermediate data dictionary 310 and the attribute hierarchical structure information 315 are input, and an attribute not included in the intermediate data dictionary 310 is deleted from the attribute hierarchical structure, and a tree structure in which an attribute value is added below each attribute is output. Then, the pre-processing attribute hierarchy display means 43 shown in FIG.
The tree structure output by 14 is displayed. The attribute hierarchy display portion 211 in FIG. 21 is a display example of a tree structure output by the display attribute hierarchy generation unit 314.

Mining Setting Next, a description will be given of a part related to item specification of a condition part and a conclusion part of the mining setting. FIG. 22 shows a condition part conclusion part item designation screen. Since the specification of the condition part and the conclusion part depends on the contents of the procedure file, it is necessary to select a procedure file to be used in advance. Here, FIG. 7 is used as the first procedure file, and FIG. 9 is used as the second procedure file.
And the case where FIG. 11 is selected as the third procedure file. Attribute hierarchy display part 221 in FIG.
Shows the attributes and attribute values of the database after applying the first, second, and third procedure files to the relational database 301 according to the attribute hierarchy set in advance by the user.

Here, the meaning of each icon regarding the item designation of the condition part and the conclusion part is as shown in FIG. In FIG. 23, for example, (1) it may appear only in the condition part of the association rule, (2) it may appear only in the conclusion part of the association rule, or (3) it may appear in the condition part of the association rule. Four modes of “good” and (4) must not appear in the association rule are prepared. In FIG. 22, for example, it is specified that the item <height: low> and the item <height: high> must not appear in the association rule. Here, the setting of the attribute other than the item or the node of the attribute group does not affect the mining execution.

An item may appear only in the condition part of the association rule, may appear only in the conclusion part, may appear in either the condition part or the conclusion part, or may appear in either the condition part or the conclusion part. There are four modes that must not appear,
Since the information is displayed in a different form, there is an effect that the setting state is easily understood by the user.

The designation of a lightly displayed icon has the same meaning, but the designation of the node and the designation of a node below the node are not all the same. Conversely, an icon that is not lightly displayed, that is, an icon that is highlighted, indicates that the designation of the node and the nodes below it are all the same. Therefore, when the icons related to the item designation of the condition part and the conclusion part are not light, the effect is that the setting state can be recognized without expanding and displaying the items below the item.

Now, in the attribute hierarchy display part 221 of FIG. 22, when a certain node is designated and the condition part check box 222 and the conclusion part check box 223 are set and the apply button 224 is clicked, the node and the node All lower nodes are designated as the same mode. For example, when a node corresponding to the attribute group <body measurement> is selected, only the condition part check box 222 is checked, and the apply button 224 is clicked, the <body measurement> node is displayed as in the attribute hierarchy display portion 241 in FIG. And all nodes below it are specified as "may appear only in the condition part of the association rule". In addition, since the designations are all the same, the icon display in the <body measurement> node is not a faint display but a highlighted display. Since such a designation method is employed, there is an effect that the load of the user setting can be greatly reduced.

Here, a method of displaying the attribute hierarchy display portion 221 in FIG. 22 will be described. The data dictionary conversion means 309 in FIG. 3 adds the attribute value list of each numerical attribute to be discretized to the data dictionary by referring to the first procedure file 70, and In addition, the intermediate data dictionary 310 is created by performing processing such as deletion of attribute values, conversion of attribute values, and the like, and conversion of attribute values to items and processing of item grouping with reference to the third procedure file 110. . If the intermediate data dictionary 310 after the application of the first procedure file already exists, the intermediate data dictionary 310 may be created by applying the second and third procedure files thereto. Similarly,
Intermediate data dictionary 31 after application of first and second procedure files
If 0 already exists, the intermediate data dictionary 310 may be created by applying the third procedure file to this.
Further, if the intermediate data dictionary 310 to which the first, second, and third procedure files have been applied already exists, the above processing is not performed.

The display attribute hierarchy generating means 314 of FIG.
An intermediate data dictionary and attribute hierarchical structure information 315 are input, an attribute not included in the intermediate data dictionary 310 is deleted from the attribute hierarchical structure, and a tree structure in which an item is added below each attribute is output. Then, the mining attribute hierarchy display means 51 of FIG. 5 is replaced with the display attribute hierarchy generation means 31 of FIG.
4 displays the tree structure output. The attribute hierarchy display portion 221 in FIG. 22 is a display example of a tree structure output by the display attribute hierarchy generation unit 314. In FIG. 22, the result of the operation by the user is interpreted by the condition part conclusion part setting means of FIG. 5, and the actual setting is performed.

[0073]

As described above, according to the data mining apparatus according to the first aspect of the present invention, as the preprocessing of mining,
A method that prepares a procedure file consisting of multiple procedures that describe conversion methods such as discretization, attribute deletion, and itemization for a relational database, and sequentially applies one or more procedure files to the relational database. Therefore, by changing the contents of the procedure file and applying it to the relational database, there is an effect that different preprocessing can be easily repeated.

According to the data mining apparatus of the second invention, a data dictionary in which attribute names and attribute value information in the relational database are described for each attribute is generated, and a procedure file is applied to the data dictionary to generate an intermediate data dictionary. Since it is possible to generate a data dictionary and display the attribute hierarchical structure information describing the relationship between the attributes of the relational database as a hierarchical structure based on the contents of the intermediate data dictionary, the hierarchical structure obtained can be displayed. There is an effect that the structure of data obtained by applying the procedure file to the relational database can be easily grasped.

According to the data mining apparatus of the third invention, in displaying the hierarchical structure, the nodes are represented by their types,
That is, since the display is made in a different form depending on whether the attribute group, the attribute, or the attribute value, it is easy for the user to understand.

According to the data mining device of the fourth aspect, it is possible to select a node in the displayed hierarchical structure and edit the procedure related to the node, so that the procedure file can be easily edited. There is an effect that can be.

According to the data mining apparatus of the fifth aspect, when a procedure is generated, the procedure creation processing is performed on the node on the attribute hierarchy display screen, so that the procedure for all the attributes below the node is performed. Are collectively created, which has the effect of reducing the load on the user.

According to the data mining apparatus of the sixth aspect, when editing the procedure file, the nodes corresponding to the respective procedures in the procedure file are highlighted in the display of the hierarchical structure. This has the effect of being easy to understand.

According to the data mining apparatus of the seventh aspect, the preprocessing procedure file is converted into the first procedure file relating to the discretization of numerical attributes, the grouping of attribute values, the replacement of attribute values with null values, and the replacement of attribute values. A second procedure file related to deletion and deletion of records satisfying specific conditions, and a third procedure file related to itemization and attribute grouping for assigning attribute information to attribute values are sequentially created and applied to the relational database. By providing such a framework, there is an effect that necessary and sufficient preprocessing can be performed.

According to the data mining apparatus of the eighth aspect, a procedure that allows only a set of procedure files having a correct dependency among the first, second, and third procedure files edited in advance to be selected. The provision of the file application setting means has an effect that necessary and sufficient preprocessing can be performed.

According to the data mining apparatus of the ninth aspect, when the condition part conclusion part of the mining is designated, the setting of the node on the attribute hierarchy display screen is transmitted to all nodes below the node. Therefore, there is an effect that the load on the user is reduced.

According to the data mining apparatus of the tenth aspect, when the condition part conclusion part of the mining is designated, the display form of the node on the attribute hierarchy display means at the time of mining depends on the mode of the condition part conclusion part specification of the node. Changes
There is an effect that the setting state is easily understood by the user.

According to the data mining apparatus of the eleventh aspect, when the mining condition part conclusion part is designated, each node on the attribute hierarchy display means at the time of mining is set to the mode of the node and the node below the node. Are all the same, the node is highlighted, so displaying the attribute hierarchy in an expanded form to an appropriate level has the effect of indicating the condition part conclusion part designation for all nodes.

According to the data mining apparatus of the twelfth aspect, the number of association rules including each item is displayed together with all items on the result display screen. There is an effect that it becomes a guide to the user when performing the process of displaying.

[Brief description of the drawings]

FIG. 1 is a diagram illustrating a health checkup database as an example of a relational database.

FIG. 2 is a diagram showing a flow of processing of data mining in the present embodiment.

FIG. 3 is a detailed block diagram of the entire data mining device in the present embodiment.

FIG. 4 is a detailed block diagram of a preprocessing setting unit shown in FIG. 3;

FIG. 5 is a detailed block diagram of a mining setting unit shown in FIG. 3;

FIG. 6 is a diagram showing a flow of a pre-processing execution process in the present embodiment.

FIG. 7 is a diagram illustrating an example of a first procedure file according to the present embodiment.

FIG. 8 is a diagram showing an intermediate database after the first procedure file of FIG. 8 is applied to the health examination database of FIG. 1;

FIG. 9 is a diagram illustrating an example of a second procedure file according to the present embodiment.

FIG. 10 is a diagram showing an intermediate database after the second procedure file of FIG. 9 is applied to the intermediate database of FIG. 8;

FIG. 11 is a diagram illustrating an example of a third procedure file according to the present embodiment.

FIG. 12 shows the intermediate database of FIG.
It is a figure showing an intermediate database after applying a procedure file.

FIG. 13 is a diagram illustrating an example of a dependency relationship between procedure files.

FIG. 14 is a diagram showing a usage procedure file selection screen according to the present embodiment.

FIG. 15 is a diagram showing a result display screen in the present embodiment.

FIG. 16 is a diagram showing a first procedure file editing screen according to the present embodiment.

FIG. 17 is a diagram showing the meaning of an icon in a procedure file editing screen according to the present embodiment.

FIG. 18 is a diagram showing a screen after a new procedure is added to the attribute <height> in FIG.

FIG. 19 is a diagram illustrating an example of an attribute hierarchical structure.

FIG. 20 is a diagram showing a second procedure file editing screen according to the present embodiment.

FIG. 21 is a diagram showing a third procedure file editing screen according to the present embodiment.

FIG. 22 is a diagram showing a condition part conclusion part item designation screen in the present embodiment.

FIG. 23 is a diagram illustrating the meaning of icons in a condition part conclusion part item designation screen according to the present embodiment.

FIG. 24 shows an attribute group <body measurement> in FIG.
FIG. 11 is a diagram showing a screen after selecting and setting.

[Explanation of symbols]

301 relational database, 302 preprocessing execution means,
303 intermediate database, 304 item database, 305 mining executing means, 306 result rule file, 307 database analyzing means, 308 data dictionary, 309 data dictionary converting means, 310 intermediate data dictionary, 311 preprocessing setting means, 312 procedure file, 313 mining Setting means, 314 display attribute hierarchy generation means, 315 attribute hierarchy structure information, 316
Result display means, 317 File management means, 318
user

Claims (12)

[Claims] 1. A preprocessing execution means for inputting a relational database and outputting an item database consisting of a set of items constituted by a plurality of records according to a user's specification, and the item database having the item database as an input In a data mining apparatus comprising: a mining execution unit for extracting a correlation rule between the items and outputting the result as a result rule file; and a result display unit for displaying the contents of the result rule file:
A procedure file for recording a procedure for converting a relational database into the item database is input. The procedure file is related to a procedure relating to discretization of numerical attributes, a procedure relating to grouping of attribute values, and replacing an attribute value with a null value. Addition, deletion, or change of any of the following procedures: a procedure for deleting attributes, a procedure for selecting records that satisfy specific conditions, a procedure for itemization to add attribute information to attribute values, and a procedure for grouping items And a procedure file application setting means for specifying one or more procedure files used in the preprocessing execution means and an order of application thereof. A data mining device characterized by the following. 2. A database analysis means for inputting the relational database and outputting a data dictionary in which attribute names and attribute value information in the relational database are described for each attribute, the data dictionary and the procedure file application setting means Data dictionary conversion means for inputting one or a plurality of ordered procedure files which have already been edited and designated as, and outputting an intermediate data dictionary as a result of applying the procedure file to the data dictionary; and A dictionary or an intermediate data dictionary, and attribute hierarchical structure information describing a relationship between attributes of the relational database as a hierarchical structure, and deleting attributes not included in the intermediate data dictionary from the attribute hierarchical structure. Display attribute hierarchy generating means for outputting a hierarchical structure in a format in which an attribute value is added below The pre-processing setting means, data mining apparatus according to claim 1, further comprising a processing time attribute hierarchy display means prior to displaying the hierarchical structure the display attribute hierarchy generating means outputs. 3. The pre-processing attribute hierarchy display means displays attribute groups, attributes, and attribute values in different forms so that users can be distinguished when displaying a hierarchical structure. The data mining apparatus according to claim 2, wherein 4. The data mining apparatus according to claim 2, wherein said procedure file editing means enables editing of a procedure relating to a node selected by a user on said preprocessing attribute hierarchy display means. apparatus. 5. The procedure file editing means according to a procedure in which a user selects a certain node of the pre-processing attribute hierarchy display means and discriminates a numeric attribute or an additional procedure performed for deleting an attribute. 5. The data mining apparatus according to claim 4, wherein a similar procedure is automatically added to all attribute nodes under the node. 6. The data according to claim 4, wherein the pre-processing attribute hierarchy display means highlights a node corresponding to each procedure in the procedure file being edited when displaying the hierarchical structure. Mining equipment. 7. A first procedure file including a procedure relating to discretization of a numerical attribute, a grouping of attribute values, replacement of attribute values with null values, deletion of attributes, and specific conditions. A second procedure file which includes a procedure relating to selection of records satisfying the first procedure file and which depends on the first procedure file; a first procedure file which includes procedures relating to itemization for adding attribute information to attribute values and grouping of items; A third procedure file depending on a second procedure file is created, and it is guaranteed that the third procedure file is applied to the relational database in the order of the first procedure file, the second procedure file, and the third procedure file. Item 2. The data mining device according to Item 1. 8. A procedure file application setting means for selecting only a set of procedure files having a correct dependency among first, second, and third procedure files edited in advance. The data mining apparatus according to claim 2, comprising: 9. A mining attribute hierarchy display means for displaying a hierarchical structure output by the display attribute hierarchy generation means, and each item in the item database may appear only in a condition part of an association rule. Specifies one of four modes: may appear only in the conclusion part of the association rule, may appear in the conclusion part of the condition part of the association rule, or may not appear in the association rule Mining setting means having a condition part conclusion part setting means, wherein the condition part conclusion part setting means selects a node of the attribute hierarchy display means at the time of mining and designates a mode, so that all of the nodes under the node are designated. 3. The data mining device according to claim 2, wherein the node of the data mining is also automatically designated as the same mode. 10. The mining attribute hierarchy display means makes the display form of each node different so that users can be distinguished by specifying a mode by the condition part conclusion part setting means. The data mining device according to claim 9. 11. The mining attribute hierarchy display means may be configured such that the mode of the node and the nodes below the node are all the same for each node by the designation of the mode by the condition part conclusion part setting means. The data mining apparatus according to claim 10, wherein the node is highlighted. 12. The data mining apparatus according to claim 1, wherein the result display screen displays all items and the number of association rules including each item.