Predictive model-based discriminator

• US 10,176,528 B2
• Filed: 06/24/2016
• Issued: 01/08/2019
• Est. Priority Date: 04/08/2008
• Status: Active Grant

• Current Assignee: Hartford Fire Insurance Company (The Hartford Financial Services Group, Inc.)
• Original Assignee: Hartford Fire Insurance Company (The Hartford Financial Services Group, Inc.)
• Inventors: Adams, Jonathan Kaleb, McGoldrick, John Albert, Ruotolo, James Dominic
• Primary Examiner(s): Wong, Eric T
• Application Number:
  US15/192,470
• Publication Number:
  US 20160307200A1
• Time in Patent Office: 928 Days
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 5/04 Inference or reasoning models
• CPC Class Codes: G06N 99/00 Subject matter not provided for in other groups of this subclass
• CPC Class Codes: G06Q 10/04 Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem" market predictions or forecasting for commercial activities G06Q30/0202
• CPC Class Codes: G06Q 20/4016 involving fraud or risk level assessment in transaction processing
• CPC Class Codes: G06Q 30/0202 Market predictions or forecasting for commercial activities
• CPC Class Codes: G06Q 40/08 Insurance

Abstract

A computer system includes a data storage module which receives, stores, and provides access to determinate data, raw indeterminate data, and extracted indeterminate data captured by an indeterminate data capture module. The computer system also includes a computer processor, a model training component, and a screening module. The model training component generates the predictive model based upon historical determinate and indeterminate data, and continuously adapts the predictive model with new historical data. The screening module categorizes current claims according to whether they are suitable for predictive analysis by the predictive model. The predictive model is applied to the current claims suitable for predictive analysis to determine a value for each claim indicative of whether the current claim transaction is questionable. The system also includes an output device which outputs the determined value for each claim, and a routing modules which routes claim workflow based on the outputted values.

Claims

1. A computer system comprising:

an indeterminate data capture component, including a natural language processing application, configured to;

receive raw indeterminate data corresponding to historical claim transactions and current claim transactions; and

extract indeterminate data from the raw indeterminate data, comprising text data, indicative of the historical claim transactions and the current claim transactions, and to mine opinion data from the text data;

a predictive model component, configured to;

train the predictive model component, based on determinate data and the extracted indeterminate data corresponding to the historical claim transactions, to generate a value indicative of whether a claim is questionable; and

adapt the predictive model component to changing patterns of questionable claim transactions by periodically updating the predictive model component based on new determinate data and new extracted indeterminate data;

a screening module configured to compare each of the current claim transactions to the data indicative of claim categories that are highly unlikely to be questionable;

a computer processor configured to;

categorize, based on application of the screening module, each of said current claim transactions as one of (1) a current claim suitable for analysis by a computerized predictive model when the current claim transaction does not correspond to a plurality of claim categories that are highly unlikely to be questionable, and (2) a current claim not suitable for analysis by the computerized predictive model when the current claim transaction corresponds to at least one of the claim categories that are highly unlikely to be questionable;

responsive to a categorization of one or more current claim transactions as not suitable for analysis by the computerized predictive model, transmit an electronic claim file to a claims processing computer; and

responsive to a categorization of one or more current claim transactions as suitable for analysis by the computerized predictive model;

determine, by application of the predictive model component to each of the one or more current claim transactions categorized as suitable for analysis by the computerized predictive model, a value indicative of whether the current claim transaction is questionable;

output, by an output device coupled to the computer processor, for each of the current claim transactions categorized as suitable for analysis by the computerized predictive model, the determined value indicative of whether the current claim transaction is questionable;

a routing module configured to;

responsive to a determination that the outputted value exceeds a threshold value, transmit data indicative of an electronic case file for the one of the current claim transactions to an investigation unit computer; and

responsive to a determination that the outputted value does not exceed the threshold value, transmit data indicative of an electronic case file to a claims processing computer; and

a data storage unit configured to receive, store, and provide access to the indeterminate data and determinate data corresponding to the historical claim transactions and to the current claim transactions, and the data indicative of claim categories that are highly unlikely to be questionable.

2. The computer system of claim 1, wherein the indeterminate data capture component includes an optical character reader for converting paper documents to text files;

and

wherein the natural language processing application identifies opinion data from narrative text data extracted by the optical character reader from narrative text notes.

3. The computer system of claim 1, wherein the natural language processing application is configured to determine whether the raw indeterminate data contains one or more predetermined statements.

4. The computer system of claim 1, wherein:

the raw indeterminate data includes a recording of a claimant'"'"'s voice; and

the indeterminate data capture component applies a speech recognition application to the recording of the claimant'"'"'s voice to generate one or both of extracted indeterminate data comprising claimant voice text data and data corresponding to the claimant'"'"'s state of mind.

5. The computer system of claim 1, wherein the extracted indeterminate data includes at least one of:

(a) text data captured by optical character reading;

(b) information extracted from text data; and

(c) text data converted from a voice signal.

6. The computer system of claim 1, wherein the routing module being configured to transmit the data indicative of the electronic case file for the one of the current claim transactions to the investigation unit computer comprises the routing module being configured to:

transmit the data indicative of the case file for the one of the current claim transactions, and data indicative of the outputted value indicative of whether each of the current claim transactions is questionable, to an investigation unit computer queue which is configured to place data indicative of case files in order according to the outputted value corresponding to each of the case files.

7. A computer-implemented method comprising:

receiving determinate data and raw indeterminate data corresponding to historical claim transactions and current claim transactions;

extracting, by one or more computer processors, from the raw indeterminate data by an indeterminate data capture module, indeterminate data comprising text data, the indeterminate data capture module including a natural language processing application to mine opinion data from the text data;

storing, in a data storage device, the determinate data and the extracted indeterminate data;

training, by a model training component, a predictive model component for evaluating claims for referral to an investigation unit based on the determinate data and the extracted indeterminate data related to the historical claim transactions, to generate a value indicative of whether a claim is questionable;

adapting the predictive model component to changing patterns of questionable claim transactions by periodically updating the predictive model component based on new determinate data and new extracted indeterminate data;

comparing, by a screening module, each of the current claim transactions to the data indicative of claim categories that are highly unlikely to be questionable;

categorizing, based on application of the screening module, each of said current claim transactions as one of (1) a current claim suitable for analysis by a computerized predictive model when the current claim transaction does not correspond to the claim categories that are highly unlikely to be questionable, and (2) a current claim not suitable for analysis by a computerized predictive model when the current claim transaction corresponds to at least one of the claim categories that are highly unlikely to be questionable;

responsive to a categorization of one or more current claim transactions as not suitable for analysis by the computerized predictive model, transmitting an electronic claim file to a claims processing computer;

responsive to a categorization of one or more current claim transactions as suitable for analysis by the computerized predictive model;

determining, by application of the predictive model component to each of the one or more current claim transactions categorized as suitable for analysis by the computerized predictive model, a value indicative of whether the current claim transaction is questionable;

outputting, by an output device coupled to the one or more computer processors, for each of the current claim transactions categorized as suitable for analysis by the computerized predictive model, the determined value indicative of whether the current claim transaction is questionable; and

directing, by a routing module, workflow including;

determining whether the outputted value for each of said current claim transactions categorized as suitable for predictive analysis exceeds a threshold value indicative of a claim being advisable for referral to an investigation unit computer;

responsive to a determination that the outputted value exceeds the threshold value, transmitting data indicative of an electronic case file for the one of the current claim transactions to an investigation unit computer; and

responsive to a determination that the outputted value does not exceed the threshold value, transmitting data indicative of an electronic case file to a claims processing computer.

8. The method of claim 7, wherein extracting the extracted indeterminate data from the raw indeterminate data includes:

applying optical character recognition to paper documents to generate text files; and

identifying, by the natural language processing application, opinion data from narrative text data extracted by application of optical character recognition to narrative text notes.

9. The method of claim 7, wherein:

the raw indeterminate data includes a recording of a claimant'"'"'s voice; and

wherein extracting the extracted indeterminate data from the raw indeterminate data includes applying a speech recognition application to the recording of the claimant'"'"'s voice to generate claimant voice text data.

10. The method of claim 7, wherein a workers'"'"' compensation claim that arises from loss of life comprises a claim category that is highly unlikely to be questionable.

11. The method of claim 7, wherein the historical claim transactions and the current claim transactions comprise one of worker'"'"'s compensation claims, bodily injury claims;

and property damage claims.

12. The method of claim 7, wherein extracting the extracted indeterminate data includes detecting if the extracted indeterminate data includes one or more of:

(a) an indication that one of said current claim transactions is disputed;

(b) an indication that a claimant engaged in physical activity; and

(c) an indication that said claimant made inconsistent statements.