JPH10301942A - Data mining device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To obtain a data mining device capable of simultaneously satisfying mutually opposed purposes, i.e., the prevention of cutting-off of necessary candidates and the suppression of generation of unnecessary candidates, in order to detect regularity based on user's interactive utilization. SOLUTION: Prior to the input of data to a data mining engine 101 for generating a correlation engine, a preprocessing part 102 is arranged so that a user can refer to the attribute of the data and select the level or attribute of the data to be processed by interactively operating the engine 101 to generate regularity having required accuracy as a mining result. Prior to the generation and display of an output from the engine 101, a post processing part 103 is arranged to enable the user to interactively select the obtained regularity, stores its history and reuse the history at succeeding processing and after, so that the selection can be gradually and automatically advanced and the device can be allowed to correspond also to much unnecessary regularity generation.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a data mining apparatus for generating a more useful association rule from a large amount of data in a database by using information related to the data.

[0002]

2. Description of the Related Art As one method of data mining for finding regularity from a large amount of data, reference 1
(Agrawal, R., “Fast Algorithm for Mining Associati
on Rules ") and the corresponding system that frequently occurs in the same record as described in" System and method for mining sequential patterns in a large-scale database "in JP-A-8-263346. Association rule methods for finding groups have been proposed.

In the association rule method, basically, the item names appearing in the data are directly subjected to mining, but this may be too detailed as a rule useful for the user. There are several possible solutions to this problem, but reference 2 (Srikant, R.,
Agrawal, R., “Mining Generalized Association Rule
s ", pp. 407-419, VLDB '95) and the corresponding" database mining system "of JP-A-8-287106), a method of generalizing rules to a level automatically considered to be effective. Has been proposed.
In this method, the system automatically removes the rules that are deemed unnecessary.

[0004]

On the other hand, from the standpoint that only the user can determine the effective level for the user, the level is determined interactively, and the rule is determined interactively. There has not been proposed a method for selecting the information. Find rules based on data items that are useful to data mining users, and delete unnecessary rules before showing them to users. Rarely occurring events, while important in some cases, tend to be buried in large amounts of data. In the conventional association rule method, candidates based on frequency and reliability are given as criteria for discarding candidates, but in rare cases, large values are discarded as early as the frequency discard criteria, while small values are selected. This leads to the generation of a large amount of mining results, and it is highly likely that valuable results will be buried in the process of selecting a large amount of results this time by the user.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-described problems of the prior art, and prevents the pruning of necessary candidates in order to discover valuable regularity based on interactive use of a user. It is an object of the present invention to obtain a data mining apparatus that can satisfy the conflicting goals of suppressing generation of unnecessary candidates at a time.

[0006]

According to the present invention, there is provided a data mining apparatus comprising: a first database storing a plurality of related data items for each record; an attribute name hierarchized for each data item; And a second database storing data including attribute values and data items of each record stored in the first database and the second database.
The hierarchical structure is read for each data item stored in the database of the second database, and the data item of each record stored in the first database is read in response to the data item selection operation input. A pre-processing unit that outputs a record obtained by converting a data item by referring to a hierarchical data structure for each data item stored in a pre-processing unit. A data mining engine that picks up data items that frequently appear in the data items, generates an association rule that satisfies the thresholds of the appearance frequency and reliability between those data items, and outputs the result as a mining result; and a mining output from the data mining engine. Each data stored in the second database is based on the result of the selection operation input of the selection. A post-processing unit that outputs a mining result that is deleted or rearranged based on a priority designation with reference to a data structure hierarchized for each data item, and a third database that stores the mining result output from the post-processing unit It is provided with.

The preprocessing unit reads a data item of each record stored in the first database, and outputs a frequency additional information for a corresponding data item each time the data item is read. A data item processing unit for calculating and storing an appearance frequency of a corresponding data item each time information is input, comparing the appearance frequency with a set reference value, and outputting an identification display instruction when the frequency is higher than the reference value; A display unit for displaying the data items of the record read through the read data interpretation unit and identifying and displaying the corresponding data items based on the identification display instruction.

[0008] The pre-processing unit is based on an input unit for receiving a selection operation input used for mining, and item conversion information input for a data item of a record input via the read data interpretation unit. A data conversion unit that outputs a record obtained by converting the data item to the data mining engine, wherein the data item processing unit performs the second data conversion on the data item of the record input via the read data interpretation unit. A hierarchical data structure of a corresponding data item is input from the database and displayed on the display unit, and mining is performed from the data structure of the hierarchical data item based on a selection operation input input via the input unit. Is selected, and an instruction to identify and display the selected data item is output to the display unit and the selected data item is selected. The item conversion information over data item is characterized in that the output to the data conversion unit.

Further, the data item processing section is characterized by simultaneously displaying the frequency information of the appearance frequency calculated and stored for the data item displayed on the display section.

[0010] The data item processing section may select a plurality of data items input through the input section with respect to the data structure of the hierarchized data items displayed on the display section. A plurality of data items to be used for mining are selected from the data structure of the hierarchized data items, and item conversion of the plurality of data items selected to output records of the plurality of data items having a hierarchical relationship to the data mining engine. The information is output to the data conversion unit.

The data item processing unit reads the attribute value of each data item stored in the second database with respect to the data structure of the hierarchized data item displayed on the display unit, and reads the attribute value. The attribute value is also displayed in the corresponding data item displayed in the section, and the data item or attribute value to which the attribute value is added based on the attribute value selection operation input input through the input section is changed. Item conversion information is output to the data conversion unit in order to output a record to which a data item has been added.

[0012] The data item processing unit may further include an attribute for each attribute name corresponding to each data item stored in the second database with respect to the data structure of the hierarchical data items displayed on the display unit. A data item or attribute value obtained by reading the hierarchical relationship of values, displaying the hierarchical attribute values on the display unit, and adding the attribute values based on the attribute value selection operation input input via the input unit. Item conversion information is output to the data conversion unit in order to output a record to which a data item in which is changed is added.

The data item processing section stores all operation inputs input via the input section as operation histories in the second database, and stores the operation histories input via the input section. An operation history is read out from the second database based on a display operation input, and a result of applying the operation history instead of a user operation is displayed on the display unit.

Further, the data item processing section stores the operation history to which the identifier input through the input section is added in the second database, and is selected by the identifier input through the input section. The operation history
And the result of applying the operation history instead of the user operation is displayed on the display unit.

[0015] The post-processing section includes a mining result holding section for temporarily storing a mining result output from the data mining engine, a display section for displaying the mining result, and a post-processing process for the displayed mining result. An input unit for receiving an operation input for processing, a mining result processing unit for processing and outputting the mining result held in the mining result holding unit based on the operation input input via the input unit, An operation history holding unit that temporarily holds all operation inputs input via the input unit as an operation history, and all operation logs temporarily held by the operation history holding unit outside the post-processing unit. And a fourth database for storing

Further, the mining result processing unit deletes a rule specified to be deleted from the mining result displayed on the display unit based on an operation input for specifying rule deletion input through the input unit. It is characterized by the following.

The mining result processing section refers to a hierarchical data structure of data items stored in the second database based on an operation input for specifying a hierarchical distance input through the input section. Then, similar rules composed of synonyms corresponding to the hierarchical distance are also collectively deleted.

Further, the mining result processing unit sets a rule corresponding to the priority designation among the rule group displayed on the display unit based on a priority designation operation input input via the input unit, to a head position. It is characterized by being moved and identified.

The mining result processing unit refers to a hierarchical data structure of data items stored in the second database based on an operation input for specifying a hierarchical distance input via the input unit. Further, the similarity rule constituted by a synonym corresponding to the hierarchical distance is also displayed on the display unit based on the priority designation.

Further, the mining result processing unit refers to the hierarchical data structure of the data items stored in the second database, and, for a rule having a large number of items in the condition part in the rule group, , A rule with a large number of items in the condition part was deleted or given priority processing for a rule with a small number of items in the condition part where the conclusion part matches or is similar to the rule and all the items in the condition part match or are similar. At the time, similar processing is executed.

The mining result processing section stores the operation history added with the identifier input through the input section in the fourth database through the operation history holding section, and stores the operation history through the input section. The operation history selected by the input identifier is read out from the fourth database, and a mining result reflecting the operation history is displayed on the display unit.

Furthermore, a fifth database storing a selection rule to which an identifier created outside is added is further provided, and the identifier stored in the fifth database is stored in the post-processing unit. An externally created selection rule holding unit that reads and temporarily holds the attached selection rule is further provided, and the mining result processing unit stores a mining result reflecting the selection rule held in the externally created selection rule holding unit. The information is displayed on the display unit.

[0023]

BEST MODE FOR CARRYING OUT THE INVENTION

Embodiment 1 FIG. The data mining device according to the first embodiment of the present invention has a configuration as shown in FIG. Generally, data mining can be said to generate an association rule as a mining result to be stored in the database 105 from a database 104 storing a plurality of related data items for each record.

Basically, the data mining engine 1
01 picks up data items that frequently appear in the same record with respect to all records stored in the database 104 and generates an association rule that satisfies the threshold of the appearance frequency and reliability between those data items. In the first embodiment, as a method of improving the quality of a rule, a method of appropriately using related information is employed. Specifically, the method includes a data structure and an attribute name hierarchized for each data item. Item dictionary / attribute dictionary database 106 storing data including attribute and attribute values, database 10 storing user operation history of rule selection
7. Use the database 108 storing the externally created selection rules.

In order to use these databases effectively, the pre-processing unit 1
02 and the post-processing unit 103 are added. That is,
The pre-processing unit 102 uses the dictionary of the database 106, and the post-processing unit 103 uses the rule selection rule of the databases 107 and 108. The user can use these pre-processing units 10
2 and in the processing stage by the post-processing unit 103, in order to obtain the most convenient result for the user, an operation input is interactively given in these pre-processing and post-processing stages. Finally, a simple and highly convenient mining result stored in the database 105 is obtained.

That is, the preprocessing unit 102 reads the data structure of each record stored in the database 104 and the data structure hierarchized for each data item stored in the database 106, and selects the data item. In response to an operation input, the data item of each record stored in the database
Reference is made to a hierarchical data structure for each data item stored in 06, and a record obtained by converting the data item is output. In addition, the post-processing unit 103 transmits the database 106 to the mining result output from the data mining engine 101 based on a selection operation input of selection.
The mining result, which is deleted or rearranged based on the priority designation, is output with reference to the data structure hierarchized for each data item stored in. In FIG. 1, reference numeral 109 denotes a user who gives an operation input in an interactive manner.

Here, as the pre-processing unit 102, FIG.
Is provided. As shown, the pre-processing unit 10
Reference numeral 2 denotes a read data interpreting unit 102a which reads data items of each record stored in the database 104 and outputs additional frequency information to the corresponding data item each time the data item is read, and a corresponding data item each time the additional frequency information is input The data item processing unit 102b calculates and stores the appearance frequency of the data item, compares the appearance frequency with a set reference value, and when the frequency is higher than the reference value, identifies the corresponding data item by color or symbol, and reads the read data interpretation unit 102a. And a display unit 102c for displaying the data items of the record read via the display unit 102, and displaying the corresponding data items based on the identification display instruction. The user can receive the information displayed on the display unit 102c. It is provided as information on items used for mining in the current situation, and is identified and displayed. Data item is made to be able to readily identifiable to be a mined.

The pre-processing unit 102 is provided with an input unit 10 for receiving a selection operation input used for mining from a user.
2d, a data conversion unit that outputs a record obtained by converting a data item based on item conversion information input to a data item of a record input via the read data interpretation unit 102a as data to be sent to the data mining engine 101 102e, and the data item processing unit 102b includes the read data interpretation unit 10b.
2a, a hierarchical data structure of the corresponding data items is input from the database 106 to the data items of the record, displayed on the display unit 102c, and selected by the input unit 102d. A data item to be used for mining is selected from the data structure of the hierarchized data items based on the operation input, an identification display instruction of the selected data item is output to the display unit 102c, and the item conversion of the selected data item is performed. Information is output to the data conversion unit 102e.

For example, from the database 104, "man, 3
8 years old, hepatitis, gastroenteritis, metal allergy, living along national highway,
The record of "weight gain, large amount of drinking, drug A effective" is read out and displayed on the display unit 102c, and among the data items, for example, the hierarchical structure (tree structure) of metal allergy is read out from the database 106 and displayed. In the case where the data item is displayed in the section 102c, if the user selects “allergy” higher than that from the hierarchical structure of “metal allergy” as a data item to be mined as a selection operation input from the user, the selected data item The identification display is output to the display unit 102c, and the data conversion unit 102e
The data is converted into a record of "male, 38 years old, hepatitis, gastroenteritis, allergy, living along the national highway, being overweight, having a large amount of drinking, and drug A effective". The data item used for mining can be specified, and the data can be converted.

The database 106 contains, for example, a binary relation or "3,""metal allergy, isa, allergy,""medicine A effective, isa, antibiotics effective,""hepatitis, incubation period, 3 years." A dictionary that internally generates hierarchical relationships from item relationships is stored.
Here, in the ternary relation, the first indicates a data item, the second indicates an attribute name, and the third indicates an attribute value.
isa indicates that the preceding item is lower than the following item.

The data item processing unit 102b
At this time, the frequency information of the appearance frequency calculated and stored for the data item displayed on the display unit 102c is simultaneously displayed, so that it is possible for the user to determine which frequency information is used for mining. I have.

The data item processing unit 102b
The data structure of the hierarchized data items displayed on the display unit 102c is changed from the data structure of the hierarchized data items based on a selection operation input of a plurality of data items input through the input unit 120d. A plurality of data items used for mining are selected, and records of a plurality of data items having a hierarchical relationship are stored in the data mining engine 1
Item conversion information of a plurality of data items selected to be output to No. 01 can be output to the data conversion unit 102e.

For example, the data mining engine 101
The record to be output to "male, 38 years old, hepatitis, gastroenteritis, allergy, living along the national highway, a bit fat, high drinking,
In addition to "Effective for drug A", data item "Allergy"
In addition, if "metal allergy" is also selected as a mining target, the record to be output to the data mining engine 101 is "male, 38 years old, hepatitis, gastroenteritis, metal allergy, living along the national highway, fat, heavy drinking, drug A Effectiveness"
Can also be output, and a plurality of data items used for mining can be selected interactively. That is,
When item B is lower than item A and item C is lower than item A, for example, both upper item A and lower item C can be selected in a partial hierarchy of the entire hierarchy.

The data item processing unit 102b
For the data structure of the hierarchical data items displayed on the display unit 102c, the attribute values of each data item stored in the database 106 are read and the display unit 10 is read.
The attribute value is also displayed in the corresponding data item displayed in 2c, and the data item or attribute value to which the attribute value is added is changed based on the attribute value selection operation input input through the input unit 102d. Item conversion information can be output to the data conversion unit 102e in order to output a record to which the added data item is added, and it is possible to specify which attribute value is specified as a mining target.

The data item processing unit 102b
The hierarchical structure of attribute values for each attribute name corresponding to each data item stored in the database 106 is read from the data structure of the hierarchical data items displayed on the display unit 102c, and the hierarchical structure is displayed on the display unit 102c. The displayed attribute value is also displayed, and a record to which a data item to which an attribute value is added or a data item to which an attribute value is changed is added based on an attribute value selection operation input input via the input unit 102d is displayed. The data conversion unit 102e for outputting
In this case, it is possible to output which level of the attribute value is to be designated as the mining target from the hierarchical relationship of the attribute value for each attribute name.

The data item processing unit 102b
All the operation inputs input via the input unit 102d are temporarily stored and stored in the database 106 as an operation history.
In the database 10 based on the operation input of the operation history input through the input unit 102d.
6, the operation history is read out, and the result of applying the operation history in place of the user operation can be displayed on the display unit 102c, and the saved history can be reused at the next execution. Making it possible.

The data item processing unit 102b
The operation history to which the identifier input via the input unit 102d is added is stored in the database 106, and the operation history selected by the identifier input via the input unit 102d is read from the database 106 and the display unit The result of applying the operation history instead of the user operation can be displayed in 102c, and it is possible to select and reuse an appropriate one from various histories by an identifier at the next execution. ing.

The post-processing section 103 has the configuration shown in FIG. As illustrated, the post-processing unit 103
Is a mining result holding unit 10 that temporarily holds mining results output from the data mining engine 101.
3a, a mining result display unit 103b for displaying the mining result, an input unit 103c for receiving an operation input for post-processing of the displayed mining result,
A mining result processing unit 103d that processes and outputs the mining result held in the mining result holding unit 103a based on an operation input input through the input unit 103c, and a mining result processing unit 103d input through the input unit 103c. A user operation history holding unit 103e that temporarily stores all operation inputs by the user as operation history and stores the held operation history in the database 107, and an external unit that reads out and temporarily stores the sorting rules stored in the database 108. The apparatus includes a creation / selection rule holding unit 103f, and selectively selects a convenient rule from correlation rules that are inevitably generated in large quantities so that a rule with a high value is treated with higher priority. Further, by storing the operation history in the database 107 or reading out the sorting rules stored in the database 108, the saved history or the sorting rules can be reused at the next execution to perform complicated operations. Can be used for post-processing without repeating.

In the post-processing unit 103, as a mining result from the data mining engine 101, for example, “resident on the national road, allergy, →, medicine A valid, 0.0
24, 0.63 "is input. here,
The left side of the arrow indicates the condition part of the rule, and the right side indicates the conclusion part. The number of data items in the condition part is indefinite.
Data item "valid", the relative frequency (appearance frequency) in which all data items include "living along the national road", "allergy" and "medicine A valid" in all records, and all data items in the condition part ( The relative frequency (reliability) of the appearance of the data item of “medicine A valid” continues under the condition that “the two data items of“ living along the national road ”and“ allergy ”) appear at the same time.

In the post-processing unit 103, the mining result processing unit 103d outputs the mining result displayed on the display unit 103b based on the operation input for specifying rule deletion input via the input unit 103c, ie, the above-described mining result processing unit 103d. As described above, a rule specified to be deleted is deleted from the rule group, and an unnecessary rule is deleted to obtain a simplified correlation rule.

The mining result processing section 103d
Is configured by a synonym corresponding to the hierarchical distance by referring to the hierarchical data structure of the data items stored in the database 106 based on the operation input for specifying the hierarchical distance input via the input unit 103c. Similar rules are also deleted at once, and all rules that include synonym data items corresponding to the specified hierarchical distance are also deleted at once, eliminating the complexity of specifying each data item one by one. It is possible to do.

The mining result processing section 103d
Moves the rule corresponding to the priority designation from the rule group displayed on the display unit 103b to the head position based on the operation input of the priority designation input via the input unit 103c, and identifies the rule by color or code. Is to be displayed,
Identifies priority rules.

The mining result processing section 103d
Is configured by a synonym corresponding to the hierarchical distance by referring to the hierarchical data structure of the data items stored in the database 106 based on the operation input for specifying the hierarchical distance input via the input unit 103c. The similarity rule to be displayed is also displayed on the display unit 103b based on the priority designation. For the similarity rule composed of synonyms, the priority can be designated by specifying a range by selecting a menu. ing.

The mining result processing section 103d
Refers to the hierarchical data structure of the data items stored in the database 106, and
For a rule with a large number of items in the condition part, for example, for a rule of "allergy, heavy drinking, living along a national road, →, drug A effective", the rule and the conclusion part are the same or similar, and all the conditions part Rules with a small number of items in the condition part where the items match or are similar, for example,
Allergy, →, Drug A is effective "
The same processing is executed when a rule having a large number of items in the condition section is deleted or subjected to priority processing, so that a rule having a matching or similar data item can be easily deleted or subjected to priority processing. I have to.

The mining result processing section 103d
Stores the operation history to which the identifier input via the input unit 103c is added in the database 107 via the user operation history storage unit 103e.
The operation history selected by the identifier input via the interface 3c is read out from the database 107 and the display unit 1 is read out.
03b displays a mining result reflecting the operation history, and in the next operation, the operation history held for each user or the operation history with an identifier can be appropriately made available to the user. I have to.

Further, the mining result processing section 103
d reads the selection rule with the identifier stored in the database 108 storing the selection rule with the identifier created externally and stores the selection rule with the identifier in the externally created selection rule holding unit 103f which temporarily stores the selection rule with the identifier. The mining result reflecting the held sorting rules is displayed on the display unit 103b, and the sorting rules created externally are stored as identifiers and can be appropriately used by the user in the same manner as the operation history. I have.

By providing both the pre-processing function and the post-processing function, it is possible to discover regularity that has been difficult to find by simply applying the association rule method. In other words, even for rarely occurring events, pruning can be prevented by reducing the frequency criterion, and those that are highly convenient in post-processing can be selected from association rules that are inevitably generated in large quantities. By choosing. Also, by grouping too detailed data items on the basis of similarity, it becomes a simpler association rule, while a correlation rule of higher interest is not pruned due to the frequency cut-off criterion and The effect of making it hard to be rejected can be expected.

[0048]

As described above, according to the data mining apparatus of the present invention, a first database storing a plurality of related data items for each record, and a hierarchical structure for each data item. A second database storing data including attribute names and attribute values, and a data item of each record stored in the first database and a data item stored in the second database. The hierarchical data structure is read, and the data item of each record stored in the first database is hierarchized for each data item stored in the second database in response to a data item selection operation input. Pre-processing unit that outputs a record in which data items are converted with reference to the data structure, and the same record for all records that are input through the pre-processing unit A data mining engine that picks up data items that frequently appear in the data items, generates an association rule that satisfies the thresholds of the appearance frequency and reliability between those data items, and outputs the result as a mining result; and a mining output from the data mining engine. Based on the selection operation input of the selection of the result, the mining result which is deleted or rearranged based on the priority designation is output by referring to the hierarchically structured data structure for each data item stored in the second database. Since a post-processing unit and a third database for storing mining results output from the post-processing unit are provided, valuable regularities are discovered based on the interactive use of the user, and necessary data items are found. Can meet the conflicting goals of preventing pruning of candidates and suppressing unnecessary generation of candidates at once, and use It is possible to generate the association rules using data item of the highest detail level of convenience for.

The preprocessing unit reads a data item of each record stored in the first database, and outputs a frequency addition information for the corresponding data item each time the data item is read. A data item processing unit for calculating and storing an appearance frequency of a corresponding data item each time information is input, comparing the appearance frequency with a set reference value, and outputting an identification display instruction when the frequency is higher than the reference value; A display unit for displaying the data items of the record read via the read data interpretation unit and identifying and displaying the corresponding data items based on the identification display instruction, so that the data items to be mined for the user. Can be easily identified.

Further, the preprocessing section is based on an input section for receiving a selection operation input used for mining, and item conversion information input for a data item of a record input via the read data interpretation section. A data conversion unit that outputs a record obtained by converting the data item to the data mining engine, wherein the data item processing unit performs the second data conversion on the data item of the record input via the read data interpretation unit. A hierarchical data structure of a corresponding data item is input from the database and displayed on the display unit, and mining is performed from the data structure of the hierarchical data item based on a selection operation input input via the input unit. Is selected, and an instruction to identify and display the selected data item is output to the display unit and the selected data item is selected. Since the data conversion information of data items is output to the data conversion unit, the hierarchical structure of data items appearing in the database is displayed in an intuitive and easy-to-understand manner so that the user can display the displayed data. By selecting data items to be used for mining from the item structure and making the selected data items at a glance, it is possible to interactively specify a data item group to be used for mining in a manner that is easy for the user to understand.

Further, the data item processing section is configured to simultaneously display the frequency information of the appearance frequency calculated and stored for the data item displayed on the display section, so that the user can use any frequency information. It can be determined whether to use for mining.

Further, the data item processing unit is configured to select a plurality of data items input through the input unit with respect to the data structure of the hierarchized data items displayed on the display unit. A plurality of data items to be used for mining are selected from the data structure of the hierarchized data items, and item conversion of the plurality of data items selected to output records of the plurality of data items having a hierarchical relationship to the data mining engine. Since the information is output to the data conversion unit, it is possible to interactively select a plurality of data items having a hierarchical relationship used for mining.

The data item processing unit reads the attribute value of each data item stored in the second database with respect to the data structure of the hierarchized data items displayed on the display unit, and reads the attribute value. The attribute value is also displayed in the corresponding data item displayed in the section, and the data item or attribute value to which the attribute value is added based on the attribute value selection operation input input through the input section is changed. Since the item conversion information is output to the data conversion unit in order to output the record to which the data item has been added, it is possible to specify which attribute value data item is to be designated as a mining target.

In addition, the data item processing unit may be configured to generate an attribute for each attribute name corresponding to each data item stored in the second database with respect to the data structure of the hierarchical data item displayed on the display unit. A data item or attribute value obtained by reading the hierarchical relationship of values, displaying the hierarchical attribute values on the display unit, and adding the attribute values based on the attribute value selection operation input input via the input unit. Since the item conversion information is output to the data conversion unit in order to output a record in which the data item in which the data item is changed is added, which level of the attribute value is specified as the mining target from the hierarchical relationship of the attribute value for each attribute name Can be determined.

The data item processing section stores all operation inputs input through the input section as operation histories in the second database, and stores the operation histories input through the input section. The operation history is read from the second database based on the display operation input, and the result of applying the operation history is displayed on the display unit by replacing the operation history with the user operation. It can be reused.

Further, the data item processing section stores the operation history to which the identifier input through the input section is added in the second database, and is selected by the identifier input through the input section. The operation history
The result of applying the operation history instead of the user operation is displayed on the display unit after reading from the database, so that the next time it is executed, it can be reused by selecting it from the various histories as appropriate by using the identifier. Can be made possible.

Further, the post-processing section includes a mining result holding section for temporarily holding a mining result output from the data mining engine, a display section for displaying the mining result, and a post-processing process for the displayed mining result. An input unit for receiving an operation input for processing, a mining result processing unit for processing and outputting the mining result held in the mining result holding unit based on the operation input input via the input unit, An operation history holding unit that temporarily holds all operation inputs input via the input unit as an operation history, and all operation logs temporarily held by the operation history holding unit outside the post-processing unit. Is provided, so that highly convenient ones can be selectively selected from a large number of association rules that are necessarily generated in large quantities. The high select value rules it is possible to deal with priority, it can be reflected in the mining results to reuse the stored operation history.

Further, the mining result processing section deletes a rule specified to be deleted from the mining result displayed on the display section based on an operation input for specifying rule deletion input through the input section. Therefore, it is possible to obtain a simplified correlation rule by deleting unnecessary rules.

The mining result processing unit refers to the hierarchical data structure of the data items stored in the second database based on the operation input for specifying the hierarchical distance input via the input unit. And similar rules composed of synonyms corresponding to the hierarchical distance are also deleted at once, so all rules containing data items of synonyms corresponding to the specified hierarchical distance are also deleted at once. The complexity of specifying each item can be eliminated.

Further, the mining result processing unit sets a rule corresponding to the priority designation among the rule group displayed on the display unit based on a priority designation operation input input via the input unit, to a head position. Since the user is moved and identified and displayed, it is possible to identify the rule preferentially specified by the user.

Further, the mining result processing section refers to a hierarchical data structure of data items stored in the second database based on an operation input for specifying a hierarchical distance input through the input section. The similarity rule composed of synonyms corresponding to the hierarchical distance is also displayed on the display unit based on the priority designation, so that priority can be easily designated for similar rules composed of synonyms. can do.

Further, the mining result processing unit refers to the hierarchical data structure of the data items stored in the second database and refers to a rule group having a large number of condition part items in a rule group. , A rule with a large number of items in the condition part was deleted or given priority processing for a rule with a small number of items in the condition part where the conclusion part matches or is similar to the rule and all the items in the condition part match or are similar In some cases, similar processing is executed, so that deletion or priority processing can be easily performed on rules with matching or similar data items.

Further, the mining result processing section causes the operation history to which the identifier input through the input section is added to be stored in the fourth database via the operation history holding section, and the mining result processing section stores the operation history through the input section. The operation history selected by the input identifier is read out from the fourth database and a mining result reflecting the operation history is displayed on the display unit, so that the operation held for each user in the next operation is performed. The history or the operation history to which the identification name is assigned can be appropriately used by the user.

Further, a fifth database storing an externally created selection rule to which an identifier is attached is further provided, and the post-processing unit stores the identifier stored in the fifth database. An externally created selection rule holding unit that reads and temporarily holds the attached selection rule is further provided, and the mining result processing unit stores a mining result reflecting the selection rule held in the externally created selection rule holding unit. Since the display rule is displayed on the display unit, the selection rule created externally can be stored as an identification name in the same manner as the operation history, and can be appropriately used by the user.

[Brief description of the drawings]

FIG. 1 is a configuration diagram showing a data mining device according to the present invention.

FIG. 2 is an internal configuration diagram of a preprocessing unit in FIG. 1;

FIG. 3 is an internal configuration diagram of a post-processing unit in FIG. 1;

[Explanation of symbols]

101 data mining engine, 102 preprocessing unit, 102a read data interpretation unit, 102b data item processing unit, 102c display unit, 102d input unit,
02e Data conversion unit, 103 Post-processing unit, 103a
Mining result holding unit, 103b display unit, 103c
Input unit, 103d Mining result processing unit, 103e
User operation history storage unit, 103f Externally created selection rule storage unit, 104-108 database.

Claims (17)

[Claims] A first database storing a plurality of related data items for each record; and a first database storing data including an attribute name and an attribute value hierarchized for each data item. And a hierarchical structure for each data item of each record stored in the first database and each data item stored in the second database. In response to the data item of each record stored in the first database, a record obtained by converting the data item with reference to a hierarchical data structure for each data item stored in the second database is output. Pre-processing unit, and picks up data items that frequently appear in the same record for all the records input through the pre-processing unit. A data mining engine that generates an association rule that satisfies the thresholds of appearance frequency and reliability between items and outputs the result as a mining result; and the mining result output from the data mining engine is selected based on a selection operation input of selection. A post-processing unit that outputs a mining result that is deleted or rearranged based on priority designation with reference to a hierarchical data structure for each data item stored in the second database; and a post-processing unit that outputs the mining result. A data mining apparatus comprising: a third database that stores a mining result. 2. The data mining device according to claim 1, wherein the preprocessing unit reads data items of each record stored in the first database, and outputs additional frequency information for the corresponding data items each time the data items are read. A read data interpreting unit that calculates and stores the appearance frequency of the corresponding data item each time the above-mentioned additional frequency information is input, compares the appearance frequency with a set reference value, and indicates an identification display when the frequency is higher than the reference value. And a display unit for displaying the data item of the record read via the read data interpretation unit and for identifying and displaying the corresponding data item based on the identification display instruction. A data mining device characterized by the following. 3. The data mining apparatus according to claim 2, wherein the pre-processing unit includes an input unit for receiving a selection operation input used for mining, and a data item of a record input via the read data interpretation unit. A data conversion unit that outputs a record obtained by converting a data item based on the input item conversion information to the data mining engine, wherein the data item processing unit is input via the read data interpretation unit. For the data item of the record, the hierarchical data structure of the corresponding data item is input from the second database and displayed on the display unit, and the hierarchical structure is selected based on the selection operation input input via the input unit. Data item to be used for mining is selected from the data structure of the coded data item, and an instruction to identify and display the selected data item A data mining device, which outputs to the display unit and the item conversion information of the selected data item to the data conversion unit. 4. The data mining device according to claim 3, wherein the data item processing unit simultaneously displays frequency information of an appearance frequency calculated and stored for the data item displayed on the display unit. Data mining equipment. 5. The data mining apparatus according to claim 4, wherein the data item processing unit is configured to input a plurality of data structures of hierarchical data items displayed on the display unit via the input unit. In order to select a plurality of data items to be used for mining from the data structure of the hierarchized data items based on the input operation of the data items of the above, and to output records of the plurality of data items having a hierarchical relationship to the data mining engine A data mining device, which outputs item conversion information of a plurality of selected data items to the data conversion unit. 6. The data mining device according to claim 3, wherein the data item processing unit is configured to execute the second data item processing on the data structure of the hierarchical data items displayed on the display unit. An attribute value for each data item stored in the database is read, the attribute value is also displayed on the corresponding data item displayed on the display unit, and an attribute value selection operation input through the input unit is input. A data mining device, which outputs item conversion information to the data conversion unit so as to output a record to which a data item to which an attribute value has been added or a data item to which an attribute value has been changed has been added. 7. The data mining device according to claim 3, wherein the data item processing unit is configured to perform the second data item processing on the data structure of the hierarchical data items displayed on the display unit. The hierarchical relationship of attribute values for each attribute name corresponding to each data item stored in the database is read and the display unit also displays the hierarchized attribute values, and the attribute values input through the input unit are displayed. A data mining device, which outputs item conversion information to the data conversion unit to output a record to which a data item to which an attribute value has been added or a data item to which an attribute value has been changed has been added based on a selection operation input. 8. The data mining apparatus according to claim 6, wherein the data item processing unit stores all operation inputs input via the input unit as operation histories in the second database, The operation history is read from the second database based on the operation history input input via the input unit, and the result obtained by applying the operation history to the display unit instead of the user operation is displayed on the display unit. Data mining equipment. 9. The data mining device according to claim 8, wherein the data item processing unit causes the second database to store an operation history to which an identifier input via the input unit is added, and stores the operation history in the second database. A data mining device that reads an operation history selected by an identifier input through the second database from the second database, replaces the operation history with a user operation on the display unit, and displays a result of applying the operation history. 10. The data mining apparatus according to claim 1, wherein the post-processing unit temporarily stores a mining result output from the data mining engine, and the mining result. Display unit, an input unit for receiving an operation input for post-processing for the displayed mining result, and the mining result storage unit based on the operation input input via the input unit. A mining result processing unit that processes and outputs a mining result, and an operation history holding unit that temporarily holds all operation inputs input via the input unit as an operation history,
A data mining apparatus, comprising: a fourth database that stores all operation histories temporarily held by the operation history holding unit, outside the post-processing unit. 11. The data mining device according to claim 10, wherein the mining result processing unit is configured to perform processing on the mining result displayed on the display unit based on an operation input for designating rule deletion input through the input unit. A data mining device for deleting a rule specified to be deleted. 12. The data mining apparatus according to claim 11, wherein the mining result processing unit stores data stored in the second database based on a hierarchical distance designation operation input input via the input unit. A data mining apparatus characterized in that similar rules formed by synonyms corresponding to the hierarchical distance are collectively deleted with reference to a hierarchical data structure of items. 13. The data mining device according to claim 10, wherein the mining result processing unit displays a table on the display unit based on a priority designation operation input input via the input unit. A data mining apparatus characterized in that a rule corresponding to a priority designation among a group of rules to be moved is moved to a head position and displayed for identification. 14. The data mining apparatus according to claim 13, wherein the mining result processing unit stores data stored in the second database based on a hierarchical distance designation operation input input via the input unit. A data mining apparatus characterized in that a similar rule composed of synonyms corresponding to the hierarchical distance is displayed on the display unit based on the priority designation with reference to a hierarchical data structure of the items. 15. The data mining device according to claim 11, wherein the mining result processing unit refers to a hierarchical data structure of data items stored in the second database. , Of a rule group, a rule with a large number of items in the condition part, a rule whose conclusion part matches or is similar to the rule, and a rule with a small number of items in the condition part where all items of the condition part match or are similar, A data mining apparatus, wherein a similar process is executed when a rule having a large number of items in the condition section is deleted or subjected to priority processing. 16. The data mining device according to claim 11, wherein the mining result processing unit stores the operation history to which the identifier input via the input unit is added, in the operation history holding unit. Through the fourth
The operation history selected by the identifier input through the input unit is read from the fourth database, and the display unit displays a mining result reflecting the operation history. Mining equipment. 17. The data mining device according to claim 16, further comprising a fifth database storing a selection rule to which an identifier created externally is added, and wherein the post-processing unit includes the fifth database. Further, an externally created selection rule holding unit for reading out and temporarily storing the selection rule with the identifier stored in the database of No. 4 is further provided, and the mining result processing unit is held in the externally created selection rule storage unit. A data mining apparatus for displaying a mining result reflecting the selection rule on the display unit.