Method and apparatus of supporting creation of classification rules

• US 20070043690A1
• Filed: 11/17/2005
• Published: 02/22/2007
• Est. Priority Date: 08/19/2005
• Status: Active Grant

• Current Assignee: Fujitsu Limited
• Original Assignee: Fujitsu Limited
• Inventors: Inakoshi, Hiroya, Uchino, Kanji, Hanno, Hirokazu, Inoue, Daigo
• Granted Patent: US 8,176,050 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06F 16/90 Details of database functions independent of the retrieved data types

Abstract

This invention is to appropriately support creation of a classification rule for a new data item. This method includes: storing a new data item and a category of the new data item into a storage; extracting feature patterns each including a condition including a feature element of the new data item stored in the storage and a corresponding category from data stored in a correct answer data storage, which stores data items and categories of the data items, and storing the feature patterns into a feature pattern storage; and grouping the feature patterns stored in the feature pattern storage into a first set whose feature pattern matches with the category of the new data item stored in the storage and a second set whose feature pattern does not match with the category of the new data item, and storing a grouping result into a group data storage. Accordingly, the feature pattern, which is a candidate of the classification rule, is automatically generated, and furthermore, it becomes possible to select a specific feature pattern as a classification rule to be applied hereinafter from the first set.

Claims

1. A method for supporting creation of a classification rule, comprising:

storing a new data item and a category of said new data item into a storage;

extracting feature patterns each including a condition including a feature element of said new data item stored in said storage and a corresponding category from data stored in a correct answer data storage, which stores data items and categories of said data items, and storing the extracted feature patterns into a feature pattern storage; and

grouping said feature patterns stored in said feature pattern storage into a first set whose feature pattern matches with said category of said new data item stored in said storage and a second set whose feature pattern does not match with said category of said new data item, and storing a grouping result into a group data storage.

2. The method as set forth in claim 1, further comprising:

extracting a classification rule whose condition part satisfies said new data item stored in said storage from a classification rule storage, which stores classification rules each being composed of a condition part for said data item and a corresponding category, and storing the extracted classification rule into an extracted classification rule storage, and

wherein said first set has a first group and a second group,

said first group is a group of said feature patterns, which are stored in said feature pattern storage, and respectively include a first specific condition and a category matching with said category of said new data item stored in said storage,

said extracted classification rule storage stores classification rules satisfying said first specific condition,

said second group is a group of said feature patterns, which are not included in any one of said first groups,

said second set includes a third group of said feature patterns, which are stored in said feature pattern storage, and respectively include a second specific condition and a category that is different from said category of said new data item stored in said storage, and

said extracted classification rule storage stores classification rules satisfying said second specific condition.

3. The method as set forth in claim 2, wherein said grouping comprises generating a fourth group of classification rules, which are stored in said extracted classification rule storage, and correspond to said first group.

4. The method as set forth in claim 2, further comprising:

detecting that a number of feature patterns included in said first group satisfies a condition concerning said number of feature patterns included in said third group associated with said first group; and

upon said detecting, changing a condition of said feature pattern included in said first group to a condition including a negative rule.

5. The method as set forth in claim 2, further comprising:

detecting that a number of feature patterns included in said first group satisfies a condition concerning said number of feature patterns included in said third group associated with said first group;

upon said detecting, changing a condition of said feature pattern included in said first group to a condition including a negative rule;

detecting that a number of feature patterns included in said third group satisfies a condition concerning said number of feature patterns included in said first group associated with said third group; and

upon said second detecting, changing a condition of said feature pattern included in said third group to a condition including a negative rule.

6. The method as set forth in claim 2, further comprising:

calculating a score of said feature pattern based on future elements for each of said feature patterns included in said first set, and storing said score into said group data storage in association with said feature pattern.

7. The method as set forth in claim 5, further comprising:

calculating a score of said feature pattern based on feature elements for each of said feature patterns included in said first set, said second set, and a third set excluding said classification rules included in said fourth group from all said classification rules stored in said extracted classification rule storage, and storing said score into said group data storage in association with said feature pattern.

8. The method as set forth in claim 5, further comprising:

removing said classification rule included in said fourth group from said classification rule storage; and

storing said feature pattern included in said third group associated with said fourth group into said classification rule storage.

9. The method as set forth in claim 3, further comprising:

presenting a user with said feature patterns included in said first set;

accepting designation of said feature pattern from said user;

storing the designated feature pattern into said classification rule storage;

identifying said third group associated with said first group including said designated feature pattern, and storing said feature pattern included in the identified third group into said classification rule storage; and

identifying said fourth group corresponding to said first group including said designated feature pattern, and removing said classification rule included in said identified fourth group from said classification rule storage.

10. The method as set forth in claim 3, further comprising:

removing said classification rule included in said fourth group from said classification rule storage;

identifying said third group associated with said fourth group, and storing said feature pattern included in the identified third group into said classification rule storage;

presenting a user with said feature pattern included in said first set;

accepting designation of said feature pattern from said user; and

storing the designated feature pattern into said classification rule storage.

11. The method as set forth in claim 7, further comprising:

registering said feature patterns stored in said group data storage into said classification rule storage as a new classification rule in descending order of said scores of said feature patterns, which are stored in said group data storage, until an error rate which changes according to a number of occurrences of misclassification for said data items stored in a correct answer data storage and for which an equation is defined in advance, increases.

12. A method for supporting creation of a classification rule, comprising:

storing a new data item and a category of said new data item into a storage;

extracting feature patterns each including a condition including a feature element of said new data item stored in said storage and a corresponding category from data stored in a correct answer data storage, which stores data items and categories of said data items, and storing said feature patterns into a group data storage; and

registering said feature patterns stored in said group data storage into said classification rule storage as a new classification rule in descending order of said scores of said feature patterns, which are stored in said group data storage, until an error rate which changes according to a number of occurrences of misclassification for said data items stored in a correct answer data storage and for which an equation is defined in advance, increases.

13. A program embodied on a medium, for supporting creation of a classification rule, said program comprising:

storing a new data item and a category of said new data item into a storage;

extracting feature patterns each including a condition including a feature element of said new data item stored in said storage and a corresponding category from data stored in a correct answer data storage, which stores data items and categories of said data items, and storing the extracted feature patterns into a feature pattern storage; and

grouping said feature patterns stored in said feature pattern storage into a first set whose feature pattern matches with said category of said new data item stored in said storage and a second set whose feature pattern does not match with said category of said new data item, and storing a grouping result into a group data storage.

14. A program embodied on a medium, for supporting creation of a classification rule, said program comprising:

storing a new data item and a category of said new data item into a storage;

extracting feature patterns each including a condition including a feature element of said new data item stored in said storage and a corresponding category from data stored in a correct answer data storage, which stores data items and categories of said data items, and storing said feature patterns into a group data storage; and

registering said feature patterns stored in said group data storage into said classification rule storage as a new classification rule in descending order of said scores of said feature patterns, which are stored in said group data storage, until an error rate which changes according to a number of occurrences of misclassification for said data items stored in a correct answer data storage and for which an equation is defined in advance, increases.

15. An apparatus for supporting creation of a classification rule, comprising:

a unit that stores a new data item and a category of said new data item into a storage;

an extractor that extracts feature patterns each including a condition including a feature element of said new data item stored in said storage and a corresponding category from data stored in a correct answer data storage, which stores data items and categories of said data items, and stores the extracted feature patterns into a feature pattern storage; and

a grouping unit that groups said feature patterns stored in said feature pattern storage into a first set whose feature pattern matches with said category of said new data item stored in said storage and a second set whose feature pattern does not match with said category of said new data item, and stores a grouping result into a group data storage.

16. An apparatus for supporting creation of a classification rule, comprising:

a unit that stores a new data item and a category of said new data item into a storage;

an extractor that extracts feature patterns each including a condition including a feature element of said new data item stored in said storage and a corresponding category from data stored in a correct answer data storage, which stores data items and categories of said data items, and stores said feature patterns into a group data storage; and

a unit that registers said feature patterns stored in said group data storage into said classification rule storage as a new classification rule in descending order of said scores of said feature patterns, which are stored in said group data storage, until an error rate which changes according to a number of occurrences of misclassification for said data items stored in a correct answer data storage and for which an equation is defined in advance, increases.