Providing training information for training a categorizer

• US 9,792,359 B2
• Filed: 04/29/2005
• Issued: 10/17/2017
• Est. Priority Date: 04/29/2005
• Status: Active Grant

• Current Assignee: Micro Focus LLC (Open Text Corporation)
• Original Assignee: entIT Software LLC (Open Text Corporation)
• Inventors: Kirshenbaum, Evan R., Suermondt, Henri J., Forman, George H., Chiocchetti, Stephane
• Primary Examiner(s): Chbouki, Tarek
• Application Number:
  US11/118,178
• Publication Number:
  US 20060248054A1
• Time in Patent Office: 4,554 Days
• Field of Search: 707 3
• CPC Class Codes: G06F 16/35 Clustering; Classification

Abstract

A method and system of providing training information for training a categorizer includes receiving a query relating to at least one category and identifying at least one case within a data set that matches the query. The method and system receives one of a first indication that the identified at least one case belongs to the category, and a second indication that the identified at least one case does not belong to the category. Training information is modified based on receiving one of the first indication and second indication.

Claims

1. A system, comprising:

at least one processor;

a data set comprising a plurality of cases;

a search engine executable on the at least one processor to receive a query relating to at least one category and to identify cases within the data set that match the query, the identified cases being unlabeled with respect to the category, wherein the search engine is to identify the cases that match the query without using a categorizer that determines whether or not the cases belong to the category;

a confirmation module executable on the at least one processor to;

receive a first user indication in a user interface that a first case of the identified cases belongs to the category, and a second user indication in the user interface that a second case of the identified cases does not belong to the category,

in response to receiving the first and second user indications in the user interface, modify training information for training the categorizer, the confirmation module modifying the training information by adding the first case to a positive training set of cases, and adding the second case to a negative training set of cases; and

a training module executable on the at least one processor to modify the categorizer based on the positive and negative training sets.

2. The system of claim 1, further comprising a display monitor configured to display information associated with the identified cases, wherein the display monitor is configured to display selection boxes for the respective identified cases, and wherein the confirmation module is executable on the at least one processor to receive the first indication for the first case in response to user selection of a respective one of the selection boxes, and the confirmation module is executable on the at least one processor to receive the second indication for the second case in response to user deselection of a respective one of the selection boxes.

3. The system of claim 1, wherein the training module is executable on the at least one processor to modify the categorizer concurrently with execution of the search engine and the confirmation module.

4. The system of claim 3, wherein the training module is executable on the at least one processor to train the categorizer in a background during execution of the search engine and the confirmation module.

5. The system of claim 1, wherein the query specifies a term and the search engine is executable on the at least one processor to match the term with information associated with the identified cases.

6. The system of claim 5, wherein the term comprises one of a string expression, a regular expression, a glob expression, a substring expression, and an expression containing non-text data.

7. The system of claim 5, wherein the term is associated with a collection of terms, and wherein the search engine is executable on the at least one processor to further match the collection of terms with information associated with the identified cases.

8. The system of claim 7, wherein the term specified by the query comprises a word, and wherein the collection of terms comprises synonyms of the word.

9. The system of claim 1, wherein the search engine is executable on the at least one processor to receive a second query and to identify cases in response to the second query by performing at least one of the following:

determining a membership status of the cases with respect to a set of cases associated with a second category;

determining whether or not the categorizer has labeled the cases with the second category; and

determining a value provided by the categorizer with respect to the cases and the second category.

10. The system of claim 1, wherein the categorizer is executable on the at least one processor to compute a confidence level with respect to a decision whether to label each of the identified cases as belonging to the category.

11. The system of claim 1, wherein the category is part of a hierarchy of categories, the system further comprising a category editor executable on the at least one processor to update the hierarchy of categories based on user inputs by adding at least one more category to the hierarchy.

12. The system of claim 11, wherein the search engine and confirmation module are executable on the at least one processor to enable identification of cases in the data set with respective categories.

13. The system of claim 11, wherein the category editor is executable on the at least one processor to enable creation of subcategories of at least one of the categories.

14. The system of claim 1, wherein the category is part of an initial hierarchy of categories, and wherein the initial hierarchy of categories provides a starting point, the system further comprising a category editor executable on the at least one processor to update the initial hierarchy of categories based on user inputs.

15. The system of claim 1, further comprising a hierarchy inference module executable on the at least one processor to examine the cases in the data set and to construct a hierarchy of categories in response to examining the cases.

16. The system of claim 1, wherein the search engine is executable on the at least one processor to receive another query not relating to the category and to identify at least one case within the data set that matches the other query, the system further comprising a display monitor to display information associated with the identified cases.

17. The system of claim 1, wherein the positive training set is a positive training set for the category, and the negative training set is a negative training set for the category.

18. The system of claim 1, wherein the search engine is executable on the at least one processor to receive the query from the user interface.

19. A method, comprising:

receiving, by a system having a processor, a query relating to at least a first category to search cases stored in a data set;

identifying, by the system, a first group of cases in the data set matching the query, the first group of cases being unlabeled with respect to the first category, wherein the first group of cases that are unlabeled with respect to the first category have not been labeled by a categorizer for determining whether or not cases belong to the first category;

receiving, by the system, a first user indication in a user interface that at least a first case in the first group belongs to the first category, and a second user indication in the user interface that at least a second case in the first group does not belong to the first category;

modifying, by the system, training information for training the categorizer in response to receiving the first and second user indications, wherein modifying the training information comprises adding the first case to a positive training set of cases and adding the second case to a negative training set of cases; and

modifying, by the system, the categorizer based on the positive and negative training sets.

20. The method of claim 19, further comprising adding the first group of cases to a baseline set.

21. The method of claim 19, further comprising:

receiving a second query relating to a second category to search the cases stored in the data set;

identifying a second group of cases in the data set matching the second query, the second group of cases unlabeled with respect to the second category;

receiving user indications in the user interface of which cases in the second group belong to the second category, the first and second categories being part of a hierarchy of categories; and

modifying the categorizer in response to receiving the user indications of which cases in the second group belong to the second category.

22. The method of claim 21, wherein the positive training set is a first positive training set, the method further comprising:

adding cases in the first group indicated as belonging to the first category to the first positive training set; and

adding cases in the second group indicated as belonging to the second category to a second positive training set of cases,

wherein modifying the categorizer is based on the first and second positive training sets.

23. The method of claim 19, further comprising displaying indicators of which cases in the first group belong to the first category and indicators of which cases in the first group do not belong to the first category.

24. The method of claim 23, wherein displaying the indicators is based on at least one of (1) whether a user has labeled cases as belonging to the first category, and (2) whether a user has labeled cases as not belonging to the first category.

25. The method of claim 19, further comprising:

receiving an indication that a second category is to be added to a hierarchy of categories, the hierarchy further including the first category.

26. The method of claim 25, wherein receiving the indication that the second category is to be added comprises receiving an indication that the second category is to be added as a child of the first category.

27. The method of claim 19, further comprising:

receiving an indication that a second category is to be added to an initial hierarchy of categories, the initial hierarchy further including the first category, wherein the initial hierarchy of categories provides a starting point.

28. The method of claim 19, wherein the first category is part of a hierarchy of categories, the method further comprising:

receiving an indication to delete a second one of the categories from the hierarchy; and

deleting the second category from the hierarchy in response to receiving the indication to delete the second category.

29. The method of claim 19, wherein the first category is part of a hierarchy of categories that further includes a second category and a third category, wherein the second category is a child of the first category, the method further comprising:

receiving an indication that the second category should be a child of the third category instead of the first category.

30. The method of claim 19, wherein the first category is part of a set of categories, the method further comprising:

performing at least one of a plurality of tasks, the plurality of tasks including;

identifying training cases for the first category; and

identifying subcategories for the first category.

31. The method of claim 30, further comprising indicating a desirability of switching from one of the plurality of tasks to another of the plurality of tasks.

32. The method of claim 19, further comprising displaying information regarding a performance of the categorizer.

33. The method of claim 32, wherein displaying the information regarding the performance of the categorizer comprises displaying at least one of a false positive rate, a true positive rate, a true negative rate, an accuracy measure, a recall measure, a precision measure, a binormal separation measure, an information gain measure, a lift measure, a stability under cross-validation measure, a measure for an area under a receiver operating characteristic curve, a number of training cases, a percentage of a target training size, an f-measure, a total cost, and an average cost.

34. The method of claim 19, wherein receiving the first user indication comprises receiving first user selection of a first user-selectable item in the user interface, and wherein receiving the second user indication comprises receiving second user selection of a second user-selectable item in the user interface.

35. The method of claim 19, wherein the positive training set is a positive training set for the first category, and the negative training set is a negative training set for the first category.

36. The method of claim 19, wherein the query is received from the user interface.

37. An article comprising at least one non-transitory storage medium containing instructions that when executed cause a computer to:

store a data set comprising a plurality of cases not labeled with respect to a category;

receive a first query relating to at least the category;

identify cases within the data set that match the first query, the identified cases unlabeled with respect to the category, wherein the identified cases that are unlabeled with respect to the category have not been labeled by a categorizer for determining whether or not the cases belong to the category;

receive a first user indication in a user interface that a first case of the identified cases belongs to the category, and a second user indication in the user interface that a second case of the identified cases does not belong to the category;

modify training information for training the categorizer in response to receiving the first user indication and the second user indication, wherein modifying the training information comprises adding the first case to a positive training set of cases and adding the second case to a negative training set of cases; and

modifying the categorizer based on the positive and negative training sets.

38. The article of claim 37, wherein the instructions when executed cause the computer to further:

receive a second query; and

identify at least another case within the data set that matches the second query, wherein the identified at least another case has been processed by the categorizer.

39. The article of claim 37, wherein the instructions when executed cause the computer to:

display the cases matching the first query, the first query specifying a limit of the number of cases to display.

40. The article of claim 39, wherein the limit is based on an amount of information that is viewable without any further action from a user.

41. The article of claim 37, wherein the instructions when executed cause the computer to display a data item associated with at least one of the identified cases in at least one of a table and a graph.

42. The article of claim 37, wherein receiving the first user indication comprises receiving first user selection of a first user-selectable item in the user interface, and wherein receiving the second user indication comprises receiving second user selection of a second user-selectable item in the user interface.

43. The article of claim 37, wherein the positive training set is a positive training set for the category, and the negative training set is a negative training set for the category.

44. The article of claim 37, wherein the first query is received based on user input in the user interface.