System and method for modifying a training set

• US 7,797,282 B1
• Filed: 09/29/2005
• Issued: 09/14/2010
• Est. Priority Date: 09/29/2005
• Status: Expired due to Fees

• Current Assignee: Micro Focus LLC (Open Text Corporation)
• Original Assignee: Hewlett-Packard Development Company, L.P. (HP Inc.)
• Inventors: Kirshenbaum, Evan R., Suermondt, Henri J., Forman, George H.
• Primary Examiner(s): Pardo; Thuy N
• Application Number:
  US11/241,082
• Time in Patent Office: 1,811 Days
• Field of Search: 707/2, 707/3, 707/5, 707/6, 707/649, 707/651, 707/681, 707/692, 370/231, 702/179, 705/2
• US Class Current: 707/651
• CPC Class Codes: G06F 16/35 Clustering; Classification
• CPC Class Codes: G06F 16/36 Creation of semantic tools, e.g. ontology or thesauri

Abstract

Information representing a hierarchy of categories is stored. An indication is received that a first case is either a positive case or a negative case for a first category in the hierarchy of categories. The first case is identified, by inference, as a positive case or negative case for a second category in the hierarchy based on relationship of the second category to the first category. A training set for the second category is modified based on the identification by reference.

Claims

1. A method comprising:

storing information representing a hierarchy of categories;

receiving, by at least one processor, an indication that a first case is either a positive case or a negative case for a first category in the hierarchy of categories;

identifying, by inference by the at least one processor, the first case as a positive case or negative case for a second category in the hierarchy based on a relationship of the second category to the first category; and

modifying a training set for the second category based on the identifying.

2. The method of claim 1, further comprising:

modifying a categorizer for the second category based on the modified training set.

3. The method of claim 2, further comprising:

in response to receiving the indication, modifying a training set for the first category; and

modifying the categorizer for the first category based on the modified training set for the first category.

4. The method of claim 1, wherein receiving the indication that a first case is either a positive case or negative case for the first category comprises receiving the indication based on manual selection made in a user interface.

5. The method of claim 1, wherein receiving the indication that a first case is either a positive case or negative case for the first category comprises receiving the indication generated based on applying one or more rules against the first case.

6. The method of claim 1, wherein identifying the first case as a positive case or negative case for the second category in the hierarchy is based on an upward inferencing rule that provides that unless explicitly indicated otherwise, any case that is a positive for a child category of the second category will be inferred to be positive for the second category.

7. The method of claim 1, wherein identifying the first case as a positive case or negative case for the second category in the hierarchy is based on a downward inferencing rule that provides that unless there is an indication that the first case should be considered positive for the second category, the first case will be inferred to be negative if a parent category of the second category considers the first case negative.

8. The method of claim 1, wherein identifying the first case as a positive case or negative case for the second category in the hierarchy is based on a lateral inferencing rule that indicates that any positive case of children categories of a parent category of the second category will be inferred to be a negative case of the second category unless otherwise indicated as a positive case.

9. The method of claim 7, further comprising disabling the lateral inferencing rule for at least some top-level categories in the hierarchy.

10. The method of claim 1, further comprising:

in response to receiving the indication, changing a first bound collection associated with the first category; and

propagating the change of the first bound collection to a second bound collection associated with the first category that contains one of positive cases and negative cases.

11. The method of claim 10, further comprising propagating the change of the second bound collection to a third bound collection associated with the second category in the hierarchy, wherein the identifying is in response to propagating the change to the third bound collection.

12. The method of claim 10, wherein the first bound collection is associated with at least one listener, and wherein propagating the change of the first bound collection to the second bound collection comprises calling the listener.

13. The method of claim 1, wherein receiving the indication is in response to one of:

(a) manual labeling of the first case as a positive case or negative case, and (b) labeling of the first case as a positive case or negative case based on one or more predefined rules, the method further comprising;

making a change to a first set in response to receiving the indication;

propagating the change of the first set to one or more other sets to enable the identifying by inference whether the first case should be a positive case or negative case for the second category in the hierarchy.

14. The method of claim 1, wherein receiving the indication comprises receiving an indication that a first case is a positive case, a negative case, not a positive case, or not a negative case for the first category;

and

wherein identifying by inference comprises identifying by inference the first case as a positive case, a negative case, not a positive case, or not a negative case for the second category.

15. The method of claim 1, further comprising:

creating a change event in response to the received indication; and

modifying a categorizer from a previously trained state using information contained in the change event.

16. The method of claim 1, further comprising:

further modifying the training set in response to a change in the hierarchy of categories.

17. The method of claim 1, wherein storing information representing the hierarchy of categories comprises storing information regarding at least two of the categories in the hierarchy that are related to each other as a parent and child.

18. A method comprising:

storing information representing a hierarchy of categories;

receiving, by at least one processor, an indication that a first case is either a positive case or a negative case for a first category in the hierarchy of categories;

identifying, by inference by the at least one processor, the first case as a positive case or negative case for a second category in the hierarchy based on a relationship of the second category to the first category;

modifying a training set for the second category based on the identifying;

storing positive and negative cases in respective sets for the first category; and

storing positive and negative cases in respective sets for a third category marked as invisible to indicate that the third category has been deleted from the hierarchy,

wherein at least one of the sets containing positive and negative cases of the third category are inherited by at least one of the sets for the first category.

19. The method of claim 18, wherein the third category is an invisible child of the first category, the hierarchy further containing a fourth category that is a visible child of the first category, the method further comprising:

storing a visible children'"'"'s positives set associated with the first category that contains positive cases of the fourth category but does not contain positive cases of the third category.

20. A system comprising:

a storage to store information representing a hierarchy of categories;

at least one processor to;

receive an indication that a first case is one of a positive case and negative case for a first category in the hierarchy of categories;

identify, by inference, the first case as a positive case or negative case for a second category in the hierarchy based on a relationship of the second category to the first category;

modify a training set for the second category based on the identifying; and

retrain a previously trained categorizer for the second category based on the modified training set.

21. The system of claim 20, wherein at least two of the categories in the hierarchy are related to each other as a parent and child.

22. A system comprising:

a storage to store information representing a hierarchy of categories;

at least one processor to;

receive an indication that a first case is one of a positive case and negative case for a first category in the hierarchy of categories;

identify, by inference, the first case as a positive case or negative case for a second category in the hierarchy based on a relationship of the second category to the first category;

modify a training set for the second category based on the identifying; and

retrain a previously trained categorizer for the second category based on the modified training set,

wherein the first case is identified as a positive case or negative case for the second category in the hierarchy based on at least one of;

an upward inferencing rule that provides that unless explicitly indicated otherwise, any case that is a positive for a child category of the second category will be inferred to be positive for the second category;

a downward inferencing rule that provides that unless there is an indication that the first case should be considered positive for the second category, the first case will be inferred to be negative if a parent category of the second category considers the first case negative; and

a lateral inferencing rule that indicates that any positive case of children categories of a parent category of the second category will be inferred to be a negative case of the second category unless otherwise indicated as a positive case.

23. A method comprising:

storing information representing a hierarchy of categories;

receiving, by at least one processor, an indication that a first case is either a positive case or a negative case for a first category in the hierarchy of categories;

identifying, by inference by the at least one processor, the first case as a positive case or negative case for a second category in the hierarchy based on a relationship of the second category to the first category;

modifying a training set for the second category based on the identifying; and

modifying a categorizer for the second category based on the modified training set,

wherein modifying the categorizer comprises modifying the categorizer that is used to decide, based on information associated with a target case, whether the target case is positive or negative with respect to the second category.

24. A system comprising:

a storage to store information representing a hierarchy of categories;

at least one processor to;

receive an indication that a first case is one of a positive case and negative case for a first category in the hierarchy of categories;

identify, by inference, the first case as a positive case or negative case for a second category in the hierarchy based on a relationship of the second category to the first category;

modify a training set for the second category based on the identifying; and

retrain a previously trained categorizer for the second category based on the modified training set,

wherein the categorizer is configured to decide, based on information associated with a target case, whether the target case is positive or negative with respect to the second category.