Deploying a document classification system

• US 20070100812A1
• Filed: 10/27/2005
• Published: 05/03/2007
• Est. Priority Date: 10/27/2005
• Status: Active Grant

• Current Assignee: Hewlett-Packard Development Company, L.P. (HP Inc.)
• Original Assignee: Hewlett-Packard Development Company, L.P. (HP Inc.)
• Inventors: Simske, Steven, Sturgill, Margaret, Dyer, Daniel
• Granted Patent: US 7,734,554 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06F 16/93 Document management systems

Abstract

A system and method for deploying a document classification system include deploying a document classification system with a first classification engine configuration, and thereafter, automatically redeploying the document classification system with a second classification engine configuration in response to a cost function.

Claims

1. A method for deploying a document classification system comprising:

deploying a document classification system with a first classification engine configuration;

automatically redeploying the document classification system with a second classification engine configuration in response to a cost function.

2. A method as recited in claim 1, wherein automatically redeploying the document classification system comprises:

testing a plurality of classification engine configurations against the cost function;

determining which of the plurality of classification engine configurations performs best against the cost function; and

replacing the first classification engine configuration with another classification engine configuration if the first classification engine configuration does not perform best against the cost function.

3. A method as recited in claim 2, wherein the determining comprises determining that the second classification engine configuration performs best against the cost function, and the replacing comprises replacing the first classification engine configuration with the second classification engine configuration.

4. A method as recited in claim 1, further comprising deriving the cost function based on one or more cost factors.

5. A method as recited in claim 4, wherein deriving the cost function comprises deriving the cost function based on a user preference regarding a throughput-performance cost factor, the throughput-performance cost factor indicating a number of documents processed by the document classification system in a given amount of time.

6. A method as recited in claim 4, wherein deriving the cost function comprises deriving the cost function based on a user preference regarding a minimum distance cost factor, the minimum distance cost factor indicating a minimum distance that a document classification is from the classified document'"'"'s assigned class.

7. A method as recited in claim 4, wherein deriving the cost function comprises deriving the cost function based on a user preference regarding a severity of errors cost factor, the severity of errors cost factor indicating a severity of an error made in classifying and indexing a document.

8. A method as recited in claim 4, wherein deriving the cost function comprises deriving the cost function based on a user preference regarding a licensing cost factor, the licensing cost factor indicating a licensing cost for using a classification engine.

9. A method as recited in claim 4, wherein deriving the cost function comprises deriving the cost function based on a user preference regarding a differential training cost factor, the differential training cost factor indicating a particular type of ground truth document being analyzed.

10. A method as recited in claim 4, wherein deriving the cost function comprises deriving the cost function based on a user preference regarding a retraining cost factor, the retraining cost factor indicating retraining of the document classification system based on additional documents added to a set of ground truth documents.

11. A method as recited in claim 4, wherein deriving the cost function comprises deriving the cost function based on a user preference regarding a system resource cost factor, the system resource cost factor indicating a relative amount of resources available for document classification on a front end of the document classification system and document indexing on a back end of the system.

12. A method as recited in claim 1, wherein the first classification engine configuration is based on an initial document training set and the second classification engine configuration is based on an updated document training set.

13. A method as recited in claim 1, wherein the first classification engine configuration employs a plurality of classification engines, and automatically redeploying the document classification system comprises:

incorporating into the cost function, an increased licensing cost associated with one of the plurality of classification engines; and

determining from the cost function, that the second classification engine configuration is preferable at least in part, because it does not employ the one of the plurality of classification engines.

14. A method as recited in claim 1, wherein automatically redeploying the document classification system comprises:

receiving a document;

determining that the document is of a special class; and

associating the second classification engine configuration with the special class.

15. One or more computer-readable media having computer-executable instructions configured for:

determining a default classification engine configuration based on a document training set;

deploying the default classification engine configuration in a document classification system;

classifying one or more documents with the document classification system;

updating the document training set based on the one or more documents; and

deploying a different classification engine configuration in the document classification system based on the updated document training set.

16. One or more computer-readable media as recited in claim 15, wherein deploying the different classification engine configuration comprises:

testing the updated document training set against a plurality of classification engine configurations; and

determining that the different classification engine configuration provides more desirable document classification results when tested against the updated document training set than others of the plurality of classification engine configurations.

17. One or more computer-readable media as recited in claim 16, wherein the testing includes testing each of the plurality of classification engine configurations against a current cost function.

18. One or more computer-readable media as recited in claim 16, having further computer-executable instructions configured for deriving the cost function based on current user preferences regarding one or more cost factors.

19. One or more computer-readable media as recited in claim 18, wherein the one or more cost factors are selected from the group comprising:

a throughput-performance cost factor indicating a number of documents that are processed by the document classification system in a given amount of time;

a minimum distance cost factor indicating a minimum distance that a document classification is from the classified document'"'"'s assigned class;

a severity of errors cost factor indicating a severity of an error made in classifying and indexing a document;

a licensing cost factor indicating a licensing cost for using a classification engine;

a differential training cost factor indicating a particular type of ground truth document being analyzed;

a retraining cost factor indicating retraining of the document classification system based on additional documents added to a set of ground truth documents; and

a system resource cost factor indicating a relative amount of resources available for document classification on a front end of the document classification system and document indexing on a back end of the system.

20. A document classification and indexing system comprising:

a document classifier having a first classification engine configuration to classify documents based on an initial document ground truth set;

a document indexer configured to index documents classified by the document classifier and to update the initial document ground truth set; and

a cost function configured to incorporate the updated document ground truth set and test a plurality of classification engine configurations against the updated document ground truth set to determine whether the first classification engine configuration should be replaced.

21. A document classification and indexing system as recited in claim 20, further comprising a plurality of cost factors configured to influence the cost function in determining whether the first classification engine configuration should be replaced.

22. A document classification and indexing system as recited in claim 20, wherein the cost factors include associated values and user preferences that are alterable to further influence the cost function in determining whether the first classification engine configuration should be replaced.

23. A document classification and indexing system as recited in claim 20, further comprising an optical character recognition module configured to convert a scanned document into a text document.

24. A document classification and indexing system as recited in claim 20, further comprising a classification result generated by the first classification engine, wherein the document indexer is configured to index a document based on the classification result.

25. A document classification and indexing system as recited in claim 24, wherein the classification result comprises a ranked classification list, and wherein the document indexer is further configured to attempt to index the document based on each classification in the ranked classification list in an order from a most probable classification to a least probable classification.

26. A document classification and indexing system as recited in claim 20, further comprising a workflow processor configured to receive from the document indexer, indices extracted from the documents and to apply the indices to an appropriate workflow.

27. A document classification and indexing system comprising:

means for deploying a classification system with a default engine pattern that is determined based on initial document training set; and

cost function means for automatically redeploying the classification system with an updated engine pattern.

28. A document classification and indexing system as recited in claim 27, wherein the cost function means incorporate cost factors and ground truth data, the document classification and indexing system further comprising:

means for updating the ground truth data;

means for updating the cost factors; and

means for evaluating a plurality of classification engine patterns against the updated ground truth and the updated cost factors to determine whether the default engine pattern should be redeployed in the classification system.