# US 2022/0044256 A1 · Utilizing machine learning models, predictive analytics, and data mining to identify a vehicle insurance fraud ring

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2022/0044256 A1
- Document ID: US20220044256A1
- Status: Published application
- Kind: A1
- Application number: 16/986,608
- Filed: 2020-08-06
- Published: 2022-02-10
- Earliest claimed: 2020-08-06
- Assignees: Accenture Global Solutions Limited
- Applicants: Accenture Global Solutions Limited
- Inventors: Subhashini LAKSHMINARAYANAN; Gopali Raval CONTRACTOR; Shantanu SS
- Web page: https://patentgrep.com/patent/US20220044256A1

## CPC

- G06Q 10/20: Administration of product repair or maintenance
- G06F 40/279: Recognition of textual entities
- G06F 40/30: Semantic analysis
- G06F 40/40: Processing or translation of natural language (natural language analysis G06F40/20; semantic analysis G06F40/30)
- G06N 3/044: Recurrent networks, e.g. Hopfield networks
- G06N 3/0442: characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
- G06N 3/045: Combinations of networks
- G06N 3/0455: Auto-encoder networks; Encoder-decoder networks
- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G06N 3/047: Probabilistic or stochastic networks
- G06N 3/088: Non-supervised learning, e.g. competitive learning
- G06N 3/09: Supervised learning
- G06N 5/025: Extracting rules from data
- G06Q 10/10: Office automation; Time management
- G06Q 10/40: Business processes related to social networking or social networking services
- G06Q 30/0185: {Product, service or business identity fraud}
- G06Q 30/0205: {based on location or geographical consideration}
- G06Q 40/08: Insurance
- G06Q 50/265: {Personal security, identity or safety}
- G06Q: INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

A device may consolidate the input data associated with vehicle insurance claims to generate processed input data. The device may process claims data, treatment data, and repair shop data to resolve ambiguities in the processed input data and to generate resolved data. The device may process the resolved data to generate related data identifying relations between persons and vehicle accidents. The device may process notes of claims adjusters and vehicle accident descriptions to extract feature data identifying features. The device may process the feature data to determine contradiction data identifying contradictions in the feature data. The device may process weather data, location data, and telematics data to determine weather conditions and locations associated with the accidents. The device may process the related data, the contradiction data, the weather conditions, and the locations to generate a knowledge graph. The device may identify a fraud ring based on the knowledge graph.

## Claims

### Claim 1

A method, comprising:
receiving, by a device, input data associated with vehicle insurance claims;
consolidating and pre-processing, by the device, the input data to generate processed input data;
processing, by the device, claims data, social media data, treatment data, and repair shop data, of the processed input data, with a long short-term memory model and a conditional random field model, to resolve ambiguities in the processed input data and to generate resolved data;
processing, by the device, the resolved data, with a neural network model, to generate related data identifying relations between persons and vehicle accidents;
performing, by the device, natural language processing on notes of claims adjusters and vehicle accident descriptions, of the resolved data, to extract feature data identifying features;
processing, by the device, the feature data, with a convolutional neural network (CNN) model, to determine contradiction data identifying contradictions in the feature data;
processing, by the device, weather data, location data, and telematics data, of the processed input data, with a graph-based entropy model, to determine actual weather conditions and actual locations associated with the vehicle accidents;
processing, by the device, the related data, the contradiction data, and data identifying the actual weather conditions and the actual locations associated with the vehicle accidents, with a graph CNN model and a graph-based attention network model, to generate a knowledge graph;
identifying, by the device and based on the knowledge graph, a fraud ring associated with two or more persons; and
performing, by the device, one or more actions based on identifying the fraud ring.

### Claim 2 (depends on 1)

The method of claim 1, wherein performing the one or more actions includes one or more of:
utilizing one or more convolutional autoencoders to validate an identification of the fraud ring;
causing one or more insurance claims associated with the fraud ring to be denied; or
notifying one or more law enforcement agencies about the fraud ring.

### Claim 3 (depends on 1)

The method of claim 1, wherein performing the one or more actions includes one or more of:
notifying one or more vehicle insurance companies about the fraud ring;
causing one or more vehicles associated with the two or more persons of the fraud ring to be disabled; or
retraining one or more of the long short-term memory model, the conditional random field model, the neural network model, the CNN model, the graph-based entropy model, the graph CNN model, or the graph-based attention network model based on data associated with the fraud ring.

### Claim 4 (depends on 1)

The method of claim 1, wherein the input data includes one or more of:
the claims data identifying the vehicle insurance claims for the vehicle accidents,
the social media data identifying social media interactions of parties associated with the vehicle accidents,
the treatment data identifying treatment of the vehicle insurance claims,
the repair shop data identifying repair shop costs associated with the vehicle insurance claims,
the data identifying the actual weather conditions and the actual locations associated with the vehicle accidents,
the weather data identifying weather associated with the vehicle accidents,
the location data identifying geographical locations associated with the vehicle accidents, or
the telematics data identifying telematics associated with the vehicle accidents.

### Claim 5 (depends on 1)

The method of claim 1, further comprising one or more of:
performing data mining on the weather data to identify weather associated with the vehicle accidents;
performing data mining on the location data to identify geographical locations associated with the vehicle accidents; or
performing data mining on the telematics data to identify telematics associated with the vehicle accidents.

### Claim 6 (depends on 1)

The method of claim 1, wherein processing the claims data, the social media data, the treatment data, and the repair shop data, of the processed input data, with the long short-term memory model and the conditional random field model, to resolve the ambiguities in the processed input data and to generate the resolved data comprises:
processing the claims data, the social media data, the treatment data, and the repair shop data, of the processed input data, with the long short-term memory model and the conditional random field model, to identify persons involved in vehicle insurance claims; and
processing data identifying the persons involved in the vehicle insurance claims, with a Levenshtein distance model and a bidirectional long short-term memory model, to resolve the ambiguities in the processed input data and to generate the resolved data.

### Claim 7 (depends on 1)

The method of claim 1, wherein the neural network model includes a graph neural network model based on an incidence matrix, an adjacency matrix, a degree matrix, and a Laplacian matrix.

### Claim 8

A device, comprising:
one or more memories; and
one or more processors, communicatively coupled to the one or more memories, configured to:
receive input data associated with vehicle insurance claims;
process claims data, social media data, treatment data, and repair shop data, of the input data, with a first model and a second model, to resolve ambiguities in the input data and to generate resolved data;
process the resolved data, with a third model, to generate related data identifying relations between persons and vehicle accidents;
perform natural language processing on notes of claims adjusters and vehicle accident descriptions, of the resolved data, to extract feature data identifying features;
process the feature data, with a fourth model, to determine contradiction data identifying contradictions in the feature data;
process weather data, location data, and telematics data, of the input data, with a fifth model, to determine actual weather conditions and actual locations associated with the vehicle accidents;
process the related data, the contradiction data, and data identifying the actual weather conditions and the actual locations associated with the vehicle accidents, with a sixth model and a seventh model, to generate a knowledge graph;
determine, based on the knowledge graph, a fraud ring associated with two or more persons; and
perform one or more actions based on the fraud ring.

### Claim 9 (depends on 8)

The device of claim 8, wherein the one or more processors are further configured to:
perform natural language processing on the social media data to extract additional features for the feature data.

### Claim 10 (depends on 8)

The device of claim 8, wherein the feature data includes one or more of:
nouns connected to a query context,
verbs connected to the query context,
adverbs connected to the query context,
adjectives connected to the query context, or
phrases connected to the query context.

### Claim 11 (depends on 8)

The device of claim 8, wherein the one or more processors, when processing the feature data, with the fourth model, to determine the contradiction data identifying the contradictions in the feature data, are configured to:
process the feature data associated with the notes of the claims adjusters, with the fourth model, to determine first meanings associated with the notes of the claims adjusters;
process the feature data associated with the vehicle accident descriptions, with the fourth model, to determine second meanings associated with the vehicle accident descriptions; and
compare the first meanings and the second meanings to determine the contradiction data identifying the contradictions in the feature data.

### Claim 12 (depends on 8)

The device of claim 8, wherein the one or more processors, when processing the weather data, the location data, and the telematics data, with the fifth model, to determine the actual weather conditions and the actual locations associated with the vehicle accidents, are configured to:
process the weather data, the location data, and the telematics data, with the fifth model, to generate a co-occurrence matrix;
determine the actual weather conditions and the actual locations associated with the vehicle accidents based on the co-occurrence matrix; and
determine frequencies of the vehicle accidents based on the actual weather conditions and the actual locations.

### Claim 13 (depends on 8)

The device of claim 8, wherein the telematics data includes data identifying one or more of:
vehicle accelerations exceeding an acceleration threshold,
vehicle braking conditions exceeding a braking threshold, or
vehicle cornering conditions exceeding a cornering threshold.

### Claim 14 (depends on 8)

The device of claim 8, wherein the one or more processors, when performing the one or more actions, are configured to:
utilize convolutional autoencoders to determine a reconstructed fraud ring;
compare data identifying the fraud ring and data identifying the reconstructed fraud ring; and
determine whether the fraud ring is valid based on comparing the data identifying the fraud ring and the data identifying the reconstructed fraud ring.

### Claim 15

A non-transitory computer-readable medium storing instructions, the instructions comprising:
one or more instructions that, when executed by one or more processors, cause the one or more processors to:
receive input data associated with vehicle insurance claims;
process claims data, social media data, treatment data, and repair shop data, of the input data, with a long short-term memory model and a conditional random field model, to resolve ambiguities in the input data and to generate resolved data;
process the resolved data, with a neural network model, to generate related data identifying relations between persons and vehicle accidents;
perform natural language processing on notes of claims adjusters and vehicle accident descriptions, of the resolved data, to extract feature data identifying features;
process the feature data, with a convolutional neural network (CNN) model, to determine contradiction data identifying contradictions in the feature data;
process weather data, location data, and telematics data, of the input data, with a graph-based entropy model, to determine actual weather conditions and actual locations associated with the vehicle accidents;
process the related data, the contradiction data, and data identifying the actual weather conditions and the actual locations associated with the vehicle accidents, with a graph CNN model and a graph-based attention network model, to generate a knowledge graph;
identify, based on the knowledge graph, a fraud ring associated with two or more persons;
utilize convolutional autoencoders to determine a reconstructed fraud ring;
compare data identifying the fraud ring and data identifying the reconstructed fraud ring; and
determine whether the fraud ring is valid based on comparing the data identifying the fraud ring and the data identifying the reconstructed fraud ring.

### Claim 16 (depends on 15)

The non-transitory computer-readable medium of claim 15, wherein the one or more instructions, when executed by the one or more processors, further cause the one or more processors to one or more of:
cause one or more insurance claims associated with the fraud ring to be denied;
notify one or more law enforcement agencies about the fraud ring;
notify one or more vehicle insurance companies about the fraud ring;
cause one or more vehicles associated with the two or more persons of the fraud ring to be disabled; or
retrain one or more of the long short-term memory model, the conditional random field model, the neural network model, the CNN model, the graph-based entropy model, the graph CNN model, or the graph-based attention network model based on the data identifying the fraud ring.

### Claim 17 (depends on 15)

The non-transitory computer-readable medium of claim 15, wherein the one or more instructions, that cause the one or more processors to process the claims data, the social media data, the treatment data, and the repair shop data, of the input data, with the long short-term memory model and the conditional random field model, to resolve the ambiguities in the input data and to generate the resolved data, cause the one or more processors to:
process the claims data, the social media data, the treatment data, and the repair shop data, of the input data, with the long short-term memory model and the conditional random field model, to identify persons involved in vehicle insurance claims; and
process data identifying the persons involved in the vehicle insurance claims, with a Levenshtein distance model and a bidirectional long short-term memory model, to resolve the ambiguities in the input data and to generate the resolved data.

### Claim 18 (depends on 15)

The non-transitory computer-readable medium of claim 15, wherein the one or more instructions, when executed by the one or more processors, further cause the one or more processors to:
perform natural language processing on the social media data to extract additional features for the feature data.

### Claim 19 (depends on 15)

The non-transitory computer-readable medium of claim 15, wherein the one or more instructions, that cause the one or more processors to process the feature data, with the CNN model, to determine the contradiction data identifying the contradictions in the feature data, cause the one or more processors to:
process the feature data associated with the notes of the claims adjusters, with the CNN model, to determine first meanings associated with the notes of the claims adjusters;
process the feature data associated with the vehicle accident descriptions, with the CNN model, to determine second meanings associated with the vehicle accident descriptions; and
compare the first meanings and the second meanings to determine the contradiction data identifying the contradictions in the feature data.

### Claim 20 (depends on 15)

The non-transitory computer-readable medium of claim 15, wherein the one or more instructions, that cause the one or more processors to process the weather data, the location data, and the telematics data, with the graph-based entropy model, to determine the actual weather conditions and the actual locations associated with the vehicle accidents, cause the one or more processors to:
process the weather data, the location data, and the telematics data, with the graph-based entropy model, to generate a co-occurrence matrix;
determine the actual weather conditions and the actual locations associated with the vehicle accidents based on the co-occurrence matrix; and
determine frequencies of the vehicle accidents based on the actual weather conditions and the actual locations.
