CLASSIFICATION RULE GENERATION DEVICE, CLASSIFICATION RULE GENERATION METHOD, CLASSIFICATION RULE GENERATION PROGRAM, AND RECORDING MEDIUM

• US 20130275433A1
• Filed: 01/13/2011
• Published: 10/17/2013
• Est. Priority Date: 01/13/2011
• Status: Active Grant

• Current Assignee: Mitsubishi Electric Corporation
• Original Assignee: Mitsubishi Electric Corporation
• Inventors: Kato, Mamoru, Kori, Mitsunori, Shibata, Hideya
• Granted Patent: US 9,323,839 B2
• Time in Patent Office: Days
• US Class Current: 707/740
• CPC Class Codes: G06F 16/35 Clustering; Classification
• CPC Class Codes: G06F 16/355 Class or cluster creation or modification
• CPC Class Codes: G06F 16/93 Document management systems

Abstract

In a document classification device 100, a sample document extraction condition storage unit 160 stores sample document extraction conditions 160-1 set for each of classification categories for extracting partial text according to the classification categories from an input document 301 input by a document input unit 110. A document matching unit 120 matches the input document 301 against the sample document extraction conditions 160-1. Based on a result of matching by the document matching unit 120, a document extraction unit 130 extracts the partial text from the input document 301 according to the classification categories. A learning unit 140 performs predetermined machine learning using as a sample document the partial text extracted by the document extraction unit 120, and thereby generates classification rules 150-1.

Claims

1. A classification rule generation device comprising:

an input unit that inputs a document as a sample target document;

a storage unit that stores extraction conditions for extracting partial text which is a portion of the sample target document and which is used for generating classification rules for classifying a classification target document to be classified into one of classification categories, the partial text being extracted from the sample target document according to the classification categories, the extraction conditions being set for each of the classification categories;

a matching unit that matches the sample target document input by the input unit against the extraction conditions stored in the storage unit;

an extraction unit that attempts to perform partial text extraction to extract the partial text from the sample target document according to the classification categories, based on a result of matching by the matching unit; and

a learning unit that, when the partial text corresponding to one of the classification categories is extracted by the partial text extraction by the extraction unit, performs predetermined machine learning using the partial text extracted, and generates the classification rules.

2. The classification rule generation device according to claim 1,

wherein the extraction conditions set for each of the classification categories include at least one appropriate keyword that is appropriate for the each of the classification categories,

wherein the matching unit includes a position identification unit that identifies an existing position of the appropriate keyword for the each of the classification categories in the sample target document, and

wherein the extraction unit extracts a portion around and including the appropriate keyword as the partial text from the sample target document, based on the existing position of the appropriate keyword identified by the position identification unit.

3. The classification rule generation device according to claim 2,

wherein the extraction conditions set for the each of the classification categories are set such that type information indicating a type of the appropriate keyword is set for at least one of the appropriate keyword, and

wherein the extraction unit, in attempting to extract the partial text corresponding to the each of the classification categories from the sample target document, attempts to perform the partial text extraction based on the type information indicated by the appropriate keyword identified by the position identification unit.

4. The classification rule generation device according to claim 3,

wherein the extraction unit, in attempting to extract the partial text corresponding to the each of the classification categories from the sample target document, does not extract the partial text corresponding to the each of the classification categories from the sample target document when a count of type information indicating a specified type out of the type information of the appropriate keyword identified by the position identification unit is less than a set value.

5. The classification rule generation device according to claim 3,

wherein the extraction unit, in attempting to extract the partial text corresponding to the each of the classification categories from the sample target document, uses the sample target document as a target of extraction to extract the partial text corresponding to the each of the classification categories only when the type information of the appropriate keyword identified by the position identification unit includes the type information indicating a specified type.

6. The classification rule generation device according to claim 3,

wherein the extraction unit, in attempting to extract the partial text corresponding to the each of the classification categories from the sample target document, changes a size of the partial text corresponding to the each of the classification categories in accordance with the type indicated by the type information of the appropriate keyword identified by the position identification unit.

7. The classification rule generation device according to claim 1,

wherein the extraction conditions set for the each of the classification categories include at least one inappropriate keyword that is not appropriate for the each of the classification categories,

wherein the matching unit includes a position identification unit that identifies an existing position of the inappropriate keyword for the each of the classification categories in the sample target document, and

wherein the extraction unit excludes a portion around and including the inappropriate keyword from the sample target document, based on the existing position of the inappropriate keyword identified by the position identification unit.

8. The classification rule generation device according to claim 7,

wherein the extraction conditions set for the each of the classification categories are set such that type information indicating a type of the inappropriate keyword is set for at least one of the inappropriate keyword, and

wherein the extraction unit, in attempting to extract the partial text corresponding to the each of the classification categories from the sample target document, attempts to perform the partial text extraction based on the type information indicated by the inappropriate keyword identified by the position identification unit.

9. The classification rule generation device according to claim 8,

wherein the extraction unit, in attempting to extract the partial text corresponding to the each of the classification categories from the sample target document, does not extract the partial text corresponding to the each of the classification categories from the sample target document when a count of type information indicating a specified type out of the type information of the inappropriate keyword identified by the position identification unit is equal to or more than a set value.

10. The classification rule generation device according to claim 8,

wherein the extraction unit, in attempting to extract the partial text corresponding to the each of the classification categories from the sample target document, uses the sample target document as the target of extraction to extract the partial text corresponding to the each of the classification categories only when the type information of the inappropriate keyword identified by the position identification unit does not coincide with the type information indicating a specified type.

11. The classification rule generation device according to claim 8,

wherein the extraction unit, in attempting to extract the partial text corresponding to the each of the classification categories from the sample target document, changes a size of an exclusion range to be excluded from the sample target document in accordance with the type indicated by the type information of the inappropriate keyword identified by the position identification unit.

12. The classification rule generation device according to claim 1,

wherein the extraction unit, when a plurality of the partial text having an identical portion are extracted for the each of the classification categories, integrates the plurality of the partial text into one piece of the partial text.

13. The classification rule generation device according to claim 2,

wherein the extraction unit, in attempting to extract the partial text corresponding to the each of the classification categories from the sample target document, does not extract the partial text corresponding to the each of the classification categories from the sample target document when a count of the appropriate keyword identified by the position identification unit is less than a set value.

14. The classification rule generation device according to claim 2,

wherein the extraction unit, in attempting to extract the partial text corresponding to the each of the classification categories from the sample target document, uses a predetermined range as the target of partial text extraction when a count of the appropriate keyword identified by the position identification unit is equal to or more than a predetermined count within the predetermined range.

15. The classification rule generation device according to claim 1,

wherein the extraction conditions set for the each of the classification categories include at least either of an appropriate keyword that is appropriate for the each of the classification categories and an inappropriate keyword that is not appropriate for the each of the classification categories,

wherein the matching unit includes a position identification unit that identifies existing positions of the appropriate keyword and the inappropriate keyword for the each of the classification categories in the sample target document, and

wherein the extraction unit extracts an area including only the appropriate keyword as the partial text from the sample target document, based on the existing positions of the appropriate keyword and the inappropriate keyword identified by the position identification unit.

16. The classification rule generation device according to claim 15,

wherein the extraction conditions set for the each of the classification categories are set such that type information indicating a type of the appropriate keyword is set for at least one of the appropriate keyword, and type information indicating a type of the inappropriate keyword is set for at least one of the inappropriate keyword, and

wherein the extraction unit, in attempting to extract the partial text corresponding to the each of the classification categories from the sample target document, uses the sample target document as the target of extraction to extract the partial text corresponding to the each of the classification categories when the type information of the appropriate keyword identified by the position identification unit coincides with the type information indicating a specified type, and the type information of the inappropriate keyword identified by the position identification unit does not coincide with the type information indicating a specified type.

17. The classification rule generation device according to claim 1,

wherein the input unit inputs an electronic mail message as the sample target document,

wherein the extraction conditions of at least one of the classification categories out of the extraction conditions set for the each of the classification categories include a specified mail address included in a header of the electronic mail message,

wherein the matching unit matches the sample target document against the mail address using the extraction conditions, and

wherein the extraction unit controls the partial text extraction in accordance with a result of matching using the mail address by the matching unit.

18. The classification rule generation device according to claim 1,

wherein the input unit inputs a Web page as the sample target document,

wherein the extraction conditions of at least one of the classification categories out of the extraction conditions set for the each of the classification categories include a URL (Uniform Resource Locator),

wherein the matching unit matches the sample target document against the URL using the extraction conditions, and

wherein the extraction unit controls the partial text extraction in accordance with a result of matching using the URL by the matching unit.

19. The classification rule generation device according to claim 7,

wherein the extraction unit, in attempting to extract the partial text corresponding to the each of the classification categories from the sample target document, does not extract the partial text corresponding to the each of the classification categories from the sample target document when a count of the inappropriate keyword identified by the position identification unit is equal to or more than a set value.

20. The classification rule generation device according to claim 1, further comprising:

a division unit that divides the classification target document according to an average size of the partial text extracted by the extraction unit.

21. A classification rule generation method that is executed by a classification rule generation device including an input unit, a storage unit, a matching unit, an extraction unit, and a learning unit, the classification rule generation method comprising:

using the input unit, inputting a sample target document;

using the storage unit, storing extraction conditions for extracting partial text which is a portion of the sample target document and which is used for generating classification rules for classifying a classification target document to be classified into one of classification categories, the partial text being extracted from the sample target document according to the classification categories, the extraction conditions being set for each of the classification categories;

by the matching unit, matching the sample target document that is input by the input unit against the extraction conditions stored in the storage unit;

by the extraction unit, attempting to perform partial text extraction to extract the partial text from the sample target document according to the classification categories;

by the learning unit, when the partial text corresponding to one of the classification categories is extracted by the partial text extraction by the extraction unit, performing predetermined machine learning using the partial text extracted, and generating the classification rules.

22. A classification rule generation program that makes a computer to function as

an input unit that inputs a document as a sample target document;

a storage unit that stores extraction conditions for extracting partial text which is a portion of the sample target document and which is used for generating classification rules for classifying a classification target document to be classified into one of classification categories, the partial text being extracted from the sample target document according to the classification categories, the extraction conditions being set for each of the classification categories;

a matching unit that matches the sample target document input by the input unit against the extraction conditions stored in the storage unit;

an extraction unit that attempts to perform partial text extraction to extract the partial text from the sample target document according to the classification categories, based on a result of matching by the matching unit; and

a learning unit that, when the partial text corresponding to one of the classification categories is extracted by the partial text extraction by the extraction unit, performs predetermined machine learning using the partial text extracted, and generates the classification rules.

23. A computer-readable recording medium recording the classification rule generation program according to claim 22.