Predictive selection of content transformation in predictive modeling systems
First Claim
Patent Images
1. A method for selecting transformation rules for application to unstructured text content in customer accounts, comprising:
- storing a plurality of customer accounts, each customer account comprising;
a structure content record of financial and personal information associated with a customer;
unstructured text content derived from an interaction with the customer; and
an actual outcome of an event related to the customer;
providing a set of source tokens from the unstructured text content of the customer accounts, each source token associated with at least one of the structured content records;
applying candidate transformation rules to a set of source tokens to selectively produce tokens in response to the transformation rules;
determining for each candidate transformation rule a statistical measure of accuracy of a predictive model for predicting outcomes of events related to the customers based on the actual outcomes of events in the customer accounts associated with the produced tokens; and
selecting transformation rules that improve the measure of accuracy of the predictive model.
12 Assignments
0 Petitions
Accused Products
Abstract
A predictive modeling system and methodology makes predictions using unstructured content as an input, either alone or in conjunction with structured content. Content transformation rules are selected for application to the unstructured content, such as emails, call center notes, and other forms of human communication, by identifying the rules that are likely to improve the performance of a predictive modeling system.
79 Citations
52 Claims
-
1. A method for selecting transformation rules for application to unstructured text content in customer accounts, comprising:
-
storing a plurality of customer accounts, each customer account comprising; a structure content record of financial and personal information associated with a customer; unstructured text content derived from an interaction with the customer; and an actual outcome of an event related to the customer; providing a set of source tokens from the unstructured text content of the customer accounts, each source token associated with at least one of the structured content records; applying candidate transformation rules to a set of source tokens to selectively produce tokens in response to the transformation rules; determining for each candidate transformation rule a statistical measure of accuracy of a predictive model for predicting outcomes of events related to the customers based on the actual outcomes of events in the customer accounts associated with the produced tokens; and selecting transformation rules that improve the measure of accuracy of the predictive model. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17)
-
-
18. A method for selecting transformation rules for application to unstructured text content in customer accounts, comprising:
-
providing a plurality of customer accounts, each customer account comprising; a structure content record of financial and personal information associated with a customer; unstructured text content derived from an interaction with the customer; and a predicted outcome from a predictive model, wherein the predictive model predicts outcomes of events in customer accounts providing an index of source tokens from the unstructured text content, each source token associated with at least one of the structured content records; applying candidate transformation rules to the source tokens to selectively produce tokens in response to the transformation rules, associating each token produced by a transformation rule with the structured content records associated with a source token; determining for each transformation rule a statistical measure of the accuracy of the predicted outcomes from the structured content records associated with the tokens produced by the transformation rule; and selecting transformation rules that improve the statistical measure of accuracy of the predicted model.
-
-
19. A computer implemented software system for selection of content transformation rules for application to unstructured text content in customer accounts, the system comprising:
-
a database of customer accounts, each customer account comprising a structure content record of financial and personal information associated with a customer; unstructured text content derived from an interaction with the customer; and a predicted outcome of an event related to the customer; an index of source tokens derived from the unstructured text content of the customer accounts, each source token associated with at least one of the structured content records; a database of content transformation rules, each transformation rule adapted to produce a token in response to a source token; a predictive model, adapted to generate the predicted outcomes of events related to the customers using the structured content records and tokens derived from the unstructured text content using the content transformation rules; and a rules selection process, adapted to apply selected transformation rules to the index to produce tokens from the source tokens, and identify transformation rules that improve the accuracy of the predictive model. - View Dependent Claims (20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35)
-
-
36. A computer program product, for selecting transformation rules for application to unstructured text content in customer accounts, and storing program instructions on a computer readable medium, the instructions causing a processor to perform the operations comprising:
-
storing a plurality of customer accounts, each customer account comprising; a structure content record of financial and personal information associated with a customer; unstructured text content derived from an interaction with the customer; and an actual outcome of an event related to the customer; providing a set of source tokens from the unstructured text content of the customer accounts, each source token associated with at least one of the structured content records; applying candidate transformation rules to a set of source tokens to selectively produce tokens in response to the transformation rules; determining for each candidate transformation rule a statistical measure of accuracy of a predictive model for predicting outcomes of events related to the customers based on the actual outcomes of events in the customer accounts associated with the produced tokens; and selecting transformation rules that to improve the measure of accuracy of the predictive model. - View Dependent Claims (37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52)
-
Specification