System and method for modifying a training set
First Claim
Patent Images
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.
9 Assignments
0 Petitions
Accused Products
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.
-
Citations
24 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. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17)
-
-
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. - View Dependent Claims (19)
-
-
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. - View Dependent Claims (21)
-
-
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.
-
Specification