JPH10247197A - Data mining device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To decrease the number of steps of a program that is produced by a user and to attain the effective retrieval of an existing relational data base by converting a data mining command into a program sentence that includes an SQL sentence to retrieve the data base. SOLUTION: When a data mining command is inputted, a command history storage means 17 is called out to store the data mining command in a command history 18. When a program is inputted, a data mining detection means 12 detects the data mining command out of the program and gives an instruction, i.e., a 'mark' to translate the command part of the program into a module that includes an SQL sentence. Then an SQL sentence generation means 13 produces an SQL sentence that is necessary for retrieval of a relational data base based on the obtained information. Thus, it's not required for a user to produce a program by himself by means of the SQL sentence.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a data mining technique for extracting rules between attributes in a database composed of various attributes.

[0002]

2. Description of the Related Art First, basket analysis will be described as one example of data mining. This is to extract rules indicating the tendency of simultaneous purchase from a set of receipts, such as "When buying bread, buy milk at the same time". It is described by a logical formula composed of the corresponding product in the form of "milk". However, it is conceivable that some receipts contain bread but not milk, and not all receipts satisfy this rule. Therefore,
Only rules having a high percentage of receipts that are satisfied among all rule candidates are extracted. As a guideline for extracting or rejecting this rule, the ratio of the number of receipts that satisfy the rule to the total number of receipts (support)
And "Ratio of the number of receipts that include all the products on the left and right sides of the rule to the number of receipts that include the products on the left side of the rule" (confidence). The former is the generality of the rule, and the latter is the rule's generality. Indicates accuracy. In the basket analysis, thresholds for these two indices are determined in advance, and the presence or absence of extraction as a rule is determined based on whether or not the index of a candidate rule exceeds the threshold.

[0003] Such a rule can be generated from a table having various attributes. Rules such as “weather = fine, gender = male → product = milk” can be extracted from a table having receipt ID, weather, gender, and product as attributes. This extraction method is the same as the basket analysis described above. In this example, rule extraction and rejection are performed in this example by the ratio of the number of receipts “weather = fine, gender = male, product = milk” to the total number of receipts (support), and “weather = Threshold, the ratio of the number of receipts “weather = fine, gender = male, product = milk” to the number of receipts “gender = male = male” (confidence). Is done.

In order to create a program for performing the above data mining process, the program must be created in a general-purpose language from the generation of rule hypotheses to data search. In addition, even when an existing relational database is used, An application program developer needs to write an SQL statement to search a database to be subjected to data mining.

[0005]

Since the conventional data mining apparatus has been realized as described above, the number of steps for creating a program relating to a part handling data in the data mining algorithm is increased, and the number of created steps is increased. The program also has a problem that it lacks versatility.

Further, when a data mining device is configured using an existing relational database, it is necessary to write an SQL sentence for a data search portion, and thus there is a problem that knowledge of the SQL language is required. Furthermore, searching from a normalized relational table has a problem that it is difficult to maintain performance because records of a plurality of tables need to be joined based on an attribute serving as a key.

SUMMARY OF THE INVENTION The present invention has been made to solve the above problems, and has as its object to reduce the number of steps of a program created by a user and to obtain a data mining device with high search efficiency from an existing relational database. And

[0008]

According to a first aspect of the present invention, there is provided a data mining apparatus comprising data mining command conversion means and a database to be subjected to data mining processing, wherein the data mining command conversion means comprises a data mining command or a data mining command. User input / output means for program input and result output including the command, command detection means for detecting a data mining command embedded in the program, data mining command and table structure of the database SQL statement generating means for generating an SQL statement by using
By providing a table structure acquisition unit for acquiring a database table structure, a data mining command is converted into a program statement including an SQL statement for searching a database.

According to a second aspect of the present invention, in the data mining apparatus according to the first aspect, the data mining apparatus includes a table reconstructing means for generating a search table used at the time of database search, thereby improving the efficiency of the database search for data mining. It is intended to improve.

According to a third invention, in the data mining apparatus according to the second invention, the table restructuring means comprises command history storage means for storing and retrieving a history of data mining commands issued so far, The search table is reconfigured based on the information.

According to a fourth aspect, in the data mining apparatus according to the second aspect, the table reconstructing means includes a rule storing means for managing and storing a rule set determined to be valid through the data mining processing up to now. , The search table is reconfigured based on the rules in the set of rules.

According to a fifth aspect, in the data mining apparatus according to the fourth aspect, the rule storage means is configured based on a set of know-how rules held by a specialist.

According to a sixth aspect, in the data mining apparatus according to the second aspect, the table reconstructing means includes an SQL history storing means for storing and retrieving a history of the SQL statement generated through the data mining process. , The search table is reconfigured based on the history information of the SQL statement.

According to a seventh aspect, in the data mining apparatus according to the second aspect, the data mining command conversion means and a database to be subjected to data mining are provided on first and second computers connected via a network. Each is provided.

[0015]

BEST MODE FOR CARRYING OUT THE INVENTION

Embodiment 1 FIG. Hereinafter, a first embodiment of the present invention will be described with reference to FIGS.

FIG. 1 is a diagram showing an example of a system configuration using a data mining device according to the present embodiment.
One of the two computers connected by the network has a target database 51 to be subjected to data mining processing.
And a search-only database 52 used in the data mining process, and these databases are managed by a relational database management system 50. The other computer is equipped with a data mining command conversion device 10 in which a program is created. It is assumed that the program is executed on the computer 2 on which the relational database management system 50 is installed. doing. The sequentially generated transaction data is added to the target database 51, and the search database 52 is updated collectively when the data mining command converter 10 is started.

In the present embodiment, it is assumed that a program in which a command is converted into an executable form is executed on the computer 1 on which the command conversion device 10 is mounted. It depends on which computer it runs on. For example, when the target database, executable program, program creation and command conversion are performed by the same computer, a system configuration as shown in FIG. 2A is adopted, and the target database and the search database are managed by the same database management system. You. When the program creation, command conversion, and execution format programs operate on the same computer, and the target database is mounted on another computer, the system configuration is as shown in FIG. 2B. . FIG. 2 (b)
In the configuration (1), a relational database management system is also provided on a computer on which command conversion is performed, and a search database is arranged there.

Next, based on the configuration diagram of the data mining command changing device shown in FIG. 3, a procedure for creating a data mining execution program will be described by taking as an example a target database composed of a relational table shown in FIG. In the present embodiment, the command for performing data mining is “MING0”, which is defined in the following format. MING (attribute 1, attribute 2, ..., attribute n, su
(port, confidence) Here, the number of attributes to be specified is two or more, and the last specified attribute n is an attribute of the conclusion clause, and the previous attribute is to extract a rule forming a conditional clause. supp
“ort” and “confidence” are thresholds of support and certainty of the rule to be extracted. For example, for the command MING (weather, merchandise, 0.01, 0.1), weather = rain → merchandise = hot coffee, s = 0.03
c = 0.2 (s is support, c is confidence) weather = cloudy → product = milk, s = 0.05, c = 0.
15 Weather = sunny → Product = ice cream, s = 0.1,
A rule such as c = 0.5 is extracted. This command can be input directly from the command prompt, executed by the data mining command converter, and the result can be displayed. Alternatively, the command can be embedded in a program written in a general-purpose programming language, and that part can be converted into the data mining command converter. It is also available in the form of converting by a device and storing the result in a variable.

Hereinafter, each means constituting the command conversion device will be described with reference to FIG. First, the user input / output means 11 will be described. This means has a role of analyzing an input given by the user to the data mining command conversion device, activating other means, and returning a final result to the user. The instructions given by the user include a direct input of a data mining command, a program in which the data mining command is embedded and its conversion instruction, or an end instruction of the data mining command conversion device. When a data mining command is directly input, the execution result is displayed. When a program statement including the data mining command and its conversion instruction are passed, the program result obtained by converting the command part into a module including the SQL statement is displayed. Output.

Next, the data mining command detecting means 12 will be described. This means detects a data mining command in the input program and replaces that part with S
It has a function of giving an instruction for translating into a module including a QL sentence, that is, a “mark”.

Next, the SQL sentence generating means 13 will be described. This means is a data mining command detecting means 12
, A relational database table to be accessed from the data mining command with the “mark” and its argument, and the contents of the select statement are set, and a module including the SQL statement is generated.

Next, the table structure obtaining means 14 has a function of obtaining the table structure from the target database 51 or the search database 52.

Next, the command execution means 15 will be described. This means has a function of, when a command is directly passed to the data mining command conversion device, compiling a module obtained by converting the command, converting the module into an executable form, accessing a database, and executing this.

Next, the table reconstructing means 16 will be described. This means determines the structure of a table to be created in the search database 52 using a data mining command issued in the past, a rule set, an SQL statement history, or the like, creates the table in the search database, and stores the data in the search database. It plays a role to bring from the target database.

Next, the command history storage means 17 will be described. This means has a function of storing in the command history 18 the data mining commands that have been requested and issued to the system in the past in the order in which they were issued. Further, a function of searching the stored command history 18 is also provided.

Next, the rule storage means 19 will be described. This means has a function of storing, in the rule set 20, rules that are regarded as valid as a result of past data mining processing and know-how rules of experts. It also has a function to search for stored rules.

Next, the SQL statement history storage means 21 will be described. This means uses the data mining command created by the data mining command converter in the past.
It has a function of storing the L sentence in the SQL sentence history 22. It also has a function to search stored SQL statements.

The operation of the data mining command converter according to this embodiment will be described with reference to the flowchart shown in FIG. In the present embodiment, it is assumed that data mining is performed from a target database including the three tables in FIG.

First, the pre-processing of step 100 will be described. The execution of this step is performed by the table structure acquisition means 14 of the data mining command conversion device. Here, the table structures of the target database 51 to be subjected to data mining and the search database 52 are acquired, and then the search database 52 is updated. This means that data added to the target database 51 since the previous update of the search database 52 is added to the search database 52.

Next, the apparatus enters a state of waiting for an input in step 200. This step is executed by the user input / output means 1
In this case, when the user inputs a termination command, the apparatus stops. On the other hand, when a data mining command or a program including the same is input, step 30 is executed.
Proceeding to 0, a command check is performed. When a data mining command is input, the command history storage unit 17 is called, and the input data mining command is stored in the command history 18.

Next, the command check in step 300 will be described. When a program is input, the data mining command detecting means 12 detects a data mining command in the input program,
An instruction for translating the part into a module including the SQL statement, that is, a “mark” is given. Then, check the arguments of each command. Here, rule = MINING (weather, merchandise, s = 0.1, c
= 0.4) is issued in the program input by the user. This command has "weather" as the attribute of the conditional clause, "product" as the attribute of the conclusion clause, and supports 0.1.
As described above, this means that all rules having a certainty factor of 0.4 or more are extracted and stored in the variable rule.

Next, the attribute check in step 400 will be described. Here, the data mining command detection unit 12 checks whether each attribute specified by the data mining command exists in any table of the target database 51. Then, check whether all attributes are set on one table,
If set, the search target of the SQL statement to be generated is set as the target database 51, and the search table search in the next step 500 is skipped.

Next, the search processing of the search table in step 500 will be described. Here, the table structure acquisition unit 14 checks whether or not a search necessary for executing the given data mining command can be performed in the search table. Specifically, it is to check whether or not a table including all the attributes specified in the data mining command exists in the search database 52. If it exists, the table name is output. If it does not exist, a message to that effect is output. In this example, it is assumed that the table of the search database 52 has not been created at this stage.

Next, the generation of the SQL statement in step 600 will be described. Here, the SQL sentence generating means 13 generates an SQL sentence necessary for relational database search based on the information obtained up to the previous step. When the table does not exist in the search database 52 by the command in the above example, the present apparatus generates a module including the following SQL statement. Since a search table has not been generated in the search database 52, the SQL statement is issued to the target database 51. SQ
Parts other than the L statement, such as for statements and variable substitution, are replaced with descriptions in a general-purpose programming language used by the input program. connect (target database) for x all in (weather) fory all in (product) ｛select ^* from receipt, product, weather where receipt. ID = product. ID, receipt. Day = weather. Sun, weather. Time = receipt. Time Weather = x → countl select ^* from receipt, commodity, weather where receipt. ID = product. ID, receipt. Day = weather. Sun, weather. Time (start) <receipt. Time <weather. Time (end), weather = x, product = y → count2 if count2 / sum> 0.1 and count2 / count1> 0.5 adopt_rule (x → y, rule) 作成 This SQL statement creation is a block diagram of FIG. It is executed based on the procedure shown in FIG. The details will be described below.

First, step 610 in FIG.
Is executed. The database of the connection destination is determined by whether or not a valid search table exists in the search of the search table in step 500. If a valid search table exists, connect
t (database for search), conn if not present
ect (target database) is set.

Next, the for statement setting step of step 620 is executed. Here, x, y,... Are sequentially assigned to each attribute specified in the data mining command.
And variables are assigned, and each variable is processed by a for statement.

Next, a table determination step of step 630 is executed. Here, the table referred to in the SQL statement is determined. This will be the minimum number of tables containing the attributes specified in the data mining command. con
If the connection destination in the nect statement is the search database 52, the number of tables is always one. Specifically, step 40
At the stage of attribute check of 0, all tables including each attribute are listed as candidates. Then, for each attribute, a combination having the smallest number of tables among all combinations of each candidate is determined as a table to be searched. For example, when table 1 and table 2 including attribute 1 and table 2 including attribute 2 are listed as candidates,
(Attribute 1, Attribute 2) → (Table 1, Table 2),
Although there are two combinations of (Table 2 and Table 2), the second combination is only one table (Table 2), and this is used as a search table.

Next, an SQL statement setting step of 640 is performed. Here, two SQL select statements are arranged in a portion enclosed by a for statement in the module. One is to set each variable in the where clause for the attribute of the conditional clause, and the other is to set each variable in the where clause for all attributes of the conditional clause and the conclusion clause. As for from, the tables determined in the table determination step of step 630 are arranged. where
The clause also adds conditions on attributes that need to be matched in the join key that joins the tables.

Finally, the verification part setting step of step 650 is performed. Here, two of the number of the corresponding records returned as a result of two select statements are set, and an assignment statement is arranged. Next, a code for comparing the support value and the certainty value threshold value based on these two numerical values is added, and a code adopted as a rule when the threshold value is satisfied is also added.

Next, step 800 in FIG. 4 will be described. Target database 5 from this created SQL statement
When the access to No. 1 is performed, the number of times of access to the target data, which is secured in advance, is incremented. This numerical value is retained even after the system is terminated. If the number of data accesses exceeds a certain threshold value, this value is reset to “0”, and the process proceeds to step 700 for table reconstruction. On the other hand, if it does not exceed the threshold, the converted program is output, and the process returns to the state of waiting for input.

Next, the reconstruction of the relational table in step 700 will be described. The table after this reconfiguration is stored in the search database 52. When the creation and the structure of the search table are determined, data is copied from the target database 51 to the search table.

The following five methods are prepared as table construction methods for reconstructing a search table.

(A) First Method The first method is to determine the table configuration based on the past data mining command issue history. The history of the data mining commands is managed by the command history storage unit 17, and is stored in the command history 18 when the command is input.

The table configuration according to this method is executed according to the procedure shown in the block diagram of FIG. Hereinafter, the details of this method will be described using examples. It is assumed that the following is stored as a past data mining command issuance history. (1) MINING (weather, temperature, product name, s = 0.0
5, c = 0.5) (2) MING (temperature, product name, s = 0.05, c
= 0.6) (3) MINING (product name, age, s = 0.1, c =
0.5) (4) MINING (age, product name, s = 0.05, c
= 0.6) (5) MINING (product name, age, s = 0.05, c
= 0.3) (6) MINING (weather, product name, s = 0.1, c =
0.3)

First, the first step 711 in FIG.
A level grouping process is performed. In this step, for each data mining command in the data mining command issuance history, those having the same conclusion clause are divided into groups. In the above example, (1), (2),
(4) and (6) have the same product as the conclusion node, and (3) and (5) have the same age as the conclusion node. The group formed here is called a level 1 group.

Next, the processing of the second level grouping in step 712 is performed. In this step, each level 1
The level 1 group is further divided into a level 2 group with reference to the attribute of the conditional clause of the data mining command in the group. Specifically, the data mining commands of the group of the level 1 group are sequentially taken out one by one, allocated to the existing level 2 group, and all the data mining commands are allocated to the level 2 group. If the data mining command to be allocated is included in any of the attribute sets of the existing level 2 group, include it in that set.
If not, create a new level 2 group and put the data mining command in it. In the above example, if the level 1 groups of (1), (2), (4), and (6) are created, first, at the stage of allocating (1), there is no level 2 group in this level 1 group. , (1) are created.
Next, regarding (2), the conditional clause (temperature) is:
Since it is included in the conditional clause (weather, temperature) of (1), it is put into the level 2 group including (1). Regarding (4), since the conditional clause (age) is not included in the attribute set of any data mining command of the level 2 group including (1) and (2), a new level 2 group is created and insert. As for (6), the conditional clause (weather) is included in the conditional clause (weather, temperature), so (1),
Place in level 2 group including (2). (1), (2), (6) as a whole, (3), (5)
The above three level 2 groups are created.

Next, a table structure determination process in step 713 is performed. In this step, the table structure is determined based on the level 2 group created in step 712. First, for each level 2 group, a table including a conditional clause of a data mining command having the attribute of the most conditional clause, its conclusion clause, and an ID is set as a candidate. Then, if there is no search database 52 or a target database 51 having the same configuration (consisting of the same attribute) in the search database 52 created up to that time, the candidate table is generated. decide.
In the above example, if the search database does not have a table (that is, the search database is not created), as shown in FIG. 10, (ID, weather, temperature, product) (ID, weather , Temperature, and merchandise) and (ID, age, merchandise). In addition, although there are three level 2 groups formed in the stage of (2), (3),
(5) made as a candidate of the table can from those consisting of (4) from (ID, age, products) are either candidates for that the same structure that is discarded.

(B) Second Method The second method is to determine the table configuration based on the results of past data mining. This is a rule set that is automatically generated by data mining and that is selected by further determining whether or not the user is valid. This set of rules 20 is managed by the rule storage means 19,
Access is made through this.

This method is executed according to the procedure shown in the block diagram of FIG. The details of this method will be described with an example. It is assumed that the following set of rules is considered to be valid as a selection result of past data mining. (1) Weather = sunny, temperature = 20 ° C → product = iced coffee, s = 0.03, c = 0.5 (s is support, c is confidence) (2) Temperature = 30 ° C → product = milk , S = 0.05, c =
0.6 (3) Product = ice cream → age = teens, s = 0.
1, c = 0.5 (4) Age = 40's → Product = Oolong tea, s = 0.03, c
= 0.1 (5) Goods = French fries → Age = 20s, s = 0.
05, c = 0.3 (6) Weather = snow → product = hot coffee, s = 0.0
1, c = 0.4

First, step 72 in FIG.
One first level grouping process is performed. In this step, for each rule in the rule set, those having the same conclusion clause are divided into groups. In the above example, (1), (2), (4) and (6) are the conclusion clauses for products,
Since (3) and (5) have the same age as the conclusion clause,
Each is divided into the same group. The group formed here is called a level 1 group.

Next, the processing of the second level grouping in step 722 is performed. In this step, each level 1
The level 1 group is further divided into a level 2 group with reference to the attribute of the condition clause of the rule in the group. Specifically, the rules of the group of the level 1 group are sequentially taken out one by one, allocated to the existing level 2 group, and all the rules are allocated to the level 2 group. Regarding the allocation method, if the allocation rule is included in any attribute set of the existing level 2 group, it is included in that set. If not, create a new level 2 group and put the rules there. In the above example, if the level 1 groups of (1), (2), (4), and (6) are created, first, at the stage of allocating (1), there is no level 2 group in this level 1 group. , (1) are created. Next, regarding (2), since the conditional clause (temperature) is included in the conditional clause (weather, temperature) of (1), it is put into the level 2 group including (1). Regarding (4), the conditional clause (age) is not included in the attribute group of any rule of the level 2 group including (1) and (2), so a new level 2 group is created and Put in. Regarding (6), since the conditional clause (weather) is included in the conditional clause (weather, temperature) of (1), it is put into the level 2 group including (1) and (2). Overall (1),
The above-mentioned three level 2 groups are generated, which are composed of (2) and (6), composed of (3) and (5), and composed of (4).

Next, the processing for determining the table structure in step 723 is performed. In this step, the structure of the table to be created is determined based on the level 2 groups generated up to the previous step. First, for each level 2 group, a table including a conditional clause of a rule having the attribute of the most conditional clause, its conclusion clause, and an ID is set as a candidate. And the search database 5 created so far
2 or in the target database 51, if there is no one having the same configuration (consisting of the same attribute),
The table generation of the candidate is determined. In the above example, if the search database has no tables (ie, no search database has been created),
As shown in FIG. 10, two tables, (ID, weather, temperature, and product) (meaning a table including ID, weather, temperature, and product as attributes), and (ID, age, and product) are determined. In addition, although there are three level 2 groups formed in the stage of (2), the table candidates made up of those composed of (3) and (5) and those composed of (4) are (I
D, age, is one of the candidates for that is the same configuration that the products) is discarded.

(C) Third Method In the third method, the table configuration is determined based on a set of know-how rules possessed by an expert. This set of rules is also managed by the rule storage unit 19. The method of determining the table structure from the rule set 20 can be realized in the same manner as (B) the second method.

(D) Fourth Method A fourth method is to determine the table configuration from an SQL sentence created from past data mining commands. The history of the past SQL sentence is managed by the SQL sentence history storage unit 21. By the way, the SQL statement newly generated from the data mining command is
At the stage of output of 00, it is added to the SQL sentence history 22 through the SQL sentence history storage means 21.

This method is executed according to the procedure shown in the block diagram of FIG. Hereinafter, the details of this method will be described using examples. It is assumed that the following SQL statements are created in the past and stored in the SQL statement history 22. (1) select ^* from receipt, weather, product where receipt. ID = product. ID, receipt.
Day = weather. Sun, weather. Time = receipt. Time Weather = x, Temperature = y, Product name = z (2) select ^* from receipt, weather, product where receipt. ID = product. ID, receipt.
Day = weather. Sun, weather. Time = receipt. Time Product name = x, Temperature = y (3) select ^* from receipt, product where receipt. ID = product. ID Product name = x, Age = y (4) select ^* from receipt, product where receipt. ID = product. ID Product name = x, Age = y (5) select ^* from receipt, product where receipt. ID = product. ID Product name = x, age = y (6) select ^* from receipt, weather, product where receipt. ID = product. ID, receipt.
Day = weather. Sun, weather. Time = receipt. Time Product name = x, Weather = y (7) select ^* from receipt where time = x, age = y

First, step 73 in FIG.
In step 1, the non-target removal processing is performed. In this step, s
For each select statement in the select statement (SQL statement) set, the number of designated tables below the from clause is one. In the example, since (7) corresponds to this, it is excluded.

Next, the process of creating an attribute sequence in step 732 is performed. In this step, w is selected for each of the select statements remaining in the non-target removal processing in step 731.
Except for items of the format “table.1 attribute 1 = table 2.attribute 2” in the here clause, a column of attributes that are item-designated elsewhere is created. In the example, for each of (1) to (6) remaining in the non-target removal processing in step 731, (1) → weather, temperature, product name (2) → temperature, product name (3) → product name, There is an attribute column of age (4) → age, product name (5) → product name, age (6) → weather, product name.

Next, the grouping process of step 733 is performed. In this step, grouping is performed with reference to the attribute sequence of each select statement in the select statement set. More specifically, select statements are sequentially taken out one by one, allocated to existing groups, and all select statements are allocated to groups. Regarding the allocation method, if the attribute sequence of the select statement to be allocated is included in any attribute set of the existing group, it is included in the set. If not, create a new group and put the rules there. In the above example, since a group is not formed at the stage of allocating (1), a group including (1) is created. Next, regarding (2), since the attribute column (temperature, product name) is included in the attribute column (weather, temperature, product name) of (1), it is put into the level 2 group including (1). Regarding (3), since the attribute sequence (product name, age) is not included in any of the attribute sequences including (1) and (2), a new group is created and put into it. The same applies to other select statements. In this way, a group consisting of (1), (2) and (6) and a group consisting of (3), (4) and (5) are created.

Finally, the grouping process of step 734 is performed. In this step, the structure of a table to be created is determined based on the groups generated up to the previous step. First, for each group, a table including an attribute column of a select statement having the most attribute in the attribute column and an ID is set as a candidate. Then, if there is no search database having the same configuration (having the same attribute) in the search database created so far, the table generation of the candidate is determined. In the above example, if the search database has no tables, that is, if the search database has not been created, (1), (2),
From (6) to (ID, weather, temperature, product name) (meaning a table consisting of ID, weather, temperature, product name as attributes), from (3), (4), (5) to (ID, age, product) 2) are determined to be created as shown in FIG.

(E) Fifth Method The fifth method is a method in which the user is shown the data mining command history and the rule result history so far, and the user decides the table configuration by referring to the history. When the function that realizes this method is activated, the user can search the command history 18 and the rule set 20, and the system provides the user with a function of classifying and displaying these for each creating user and each conclusion clause attribute. provide.

When the table reconstruction is completed, the operation of the present apparatus returns to the search table search processing of step 500 in FIG.
Subsequently, the SQL sentence generation in step 600 is performed. If a necessary table exists in the tables in the search database, the connect statement in the first part of the generated program is a connect (search database). At this stage, the search table 1 including (ID, weather, temperature, and product name) is created on the search database by the table reconstruction in step 700. Therefore, since this table is found by searching the search table in step 500, the SQL in step 600
The SQL statement in the program created by the generation becomes this table of the search database (tentatively referred to as search table 1), and the following modules are generated. connect (database for search) for x all in (weather) for all in (product name) ^* select ^* from search table 1 where weather = x → count1 select ^* from search table 1 where weather = x, product name = y → count2 if count2 / sum> 0.1 and count2 / count1> 0.3 adopt_rule (x → y, rule) 、 This module has fewer tables to refer to than the module that searches the target database created first. Therefore, more efficient database search is realized.

Finally, the output of step 900 will be described. Here, the user input / output means 11 replaces the data mining command marked with "" in the input program with the module including the generated SQL statement and outputs the data in the form of a file or the like. The SQL sentence in the program output here is added to the SQL sentence history 22 managed by the SQL sentence history storage unit 21. The above is a description of a case where a program is input. However, when a command is directly input from the user by means such as a prompt, the module is compiled by the command executing means, converted into an executable program, and executed.

Embodiment 2 Next, a second embodiment of the present invention will be described. In the first embodiment, all functions of rule extraction are realized by one data mining command. However, in the development of a data mining algorithm, an attempt is made to improve the performance by reducing the number of database searches. Therefore, it is necessary to provide a search function for a part of the database instead of performing a search for the entire database. In the present embodiment, this function can be similarly designated by a data mining command. For example, if a rule having a weather as an attribute of a conditional clause and a product name as an attribute of a conclusion clause is extracted and an algorithm is to find a rule in which the weather clause is “weather = fine”, rule = MINING (weather = Fine, product, s = 0.
(1, c = 0.3)
This part is expanded into the following program. connect (target database) x = fine for all in (product) ｛select ^* from receipt, product, weather where receipt. ID = product. ID, receipt. Day = weather. Sun, weather. Time = receipt. Time Weather = x → count1 select ^* from Receipt, goods, weather where receipt. ID = product. ID, receipt. Day = weather. Sun, weather. Time (start) <receipt. Time <weather. Time (end), weather = x, product = y → count2 if count2 / sum> 0.1 and count2 / count1> 0.5 adopt_rule (x → y) 生成 This program is generated in step 600 in the first embodiment. Fo, the second step in the SQL statement generation of
In the setting of the r statement, for the attribute for which the value is set, f
This is the same as the first embodiment, except that an operation of arranging a statement for assigning a value specified in a data mining command to a variable before a for loop without forming an or statement is performed.

[0064]

As described above, according to the present invention, there is an effect that a user does not need to create a program for performing data mining processing from an existing relational database using an SQL statement by himself.

Further, since a table for database search is newly created by providing a table reconstructing means, the efficiency of searching the relational database for the output program can be increased, and the performance can be improved. In addition, even if there is a program for operating the target database other than the program issuing the data mining command, the program does not need to be affected by the table reconfiguration.

Further, the database search table is stored based on the data mining command history issued so far, or the rules and know-how rules determined to be valid through the data mining process, and the history information of the previously generated SQL statement. Since the reconfiguration is performed, there is an effect that an efficient search that reflects the database operation status can be realized.

Further, since the command conversion device and the database to be processed are realized separately on each computer via a network, there is an effect that a data mining device suitable for the operating environment can be realized.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a system configuration of a data mining device according to a first embodiment of the present invention.

FIG. 2 is a diagram illustrating another example of a system configuration of the data mining device.

FIG. 3 is a system configuration diagram of a data mining command conversion device according to the first embodiment of the present invention.

FIG. 4 is a flowchart illustrating a processing procedure of data mining command conversion according to the first embodiment.

FIG. 5 is a diagram showing a table configuration of a target database used for explaining the first embodiment.

FIG. 6 is a flowchart illustrating a procedure of generating an SQL sentence according to the first embodiment.

FIG. 7 is a flowchart illustrating a procedure for reconstructing a table based on history information of a data mining command according to the first embodiment.

FIG. 8 is a flowchart illustrating a table reconstruction method using a rule set according to the first embodiment.

FIG. 9 is a flowchart illustrating a table reconstruction method using an SQL statement according to the first embodiment.

FIG. 10 is a diagram illustrating an example of a table created in a search database as a result of table reconstruction in the first embodiment.

[Explanation of symbols]

Reference Signs List 10 data mining command converter, 11 user input / output means, 12 data mining command detecting means, 13 SQL sentence generating means, 14 table structure acquiring means, 15 command executing means, 16 table reconstructing means, 17 command history storing means, 18 Command history, 19 rule storage means, 20 rule sets, 21
SQL statement history storage means, 22 SQL statement history, 50 relational database management system, 51 target database, 52 search database.

Claims (7)

[Claims] 1. A data mining command conversion means,
In a data mining apparatus configured from a database to be subjected to data mining processing, the data mining command conversion means includes a data mining command or a program input including the command, and a user input / output means for outputting a result, and the program A command detection unit for detecting a data mining command embedded in the database, an SQL statement generation unit for generating an SQL statement based on the data mining command and the table structure of the database, and acquiring the database table structure A data mining apparatus, comprising: a table structure acquiring unit that converts a data mining command into a program sentence including an SQL sentence for searching a database. 2. The data mining apparatus according to claim 1, further comprising a table reconstructing means for generating a search table used for searching a database, thereby improving a database search efficiency of the data mining. 2. The data mining device according to 1. 3. The table reconstructing means includes command history storing means for storing and retrieving a history of data mining commands issued so far, and reconstructing a retrieval table based on the command history information. 3. The data mining command device according to claim 2, wherein: 4. The table reconstructing means includes a rule storing means for managing and storing a rule set determined to be valid through the data mining process up to now, and for retrieving based on rules in the rule set. 3. The data mining apparatus according to claim 2, wherein the table is reconfigured. 5. The data mining apparatus according to claim 4, wherein said rule storage means is configured based on a set of know-how rules held by an expert. 6. The table reconstructing means includes an SQL history storage means for storing and retrieving a history of SQL statements generated through data mining processing, and a retrieval table based on the history information of the SQL statements. 3. The data mining apparatus according to claim 2, wherein is reconstructed. 7. The data mining command conversion means and a database to be subjected to data mining are provided on first and second computers connected via a network, respectively. Data mining device as described.