US 12,597,074 B1Granted

Method and system for identifying overpayment of claims

AssigneeEspereo Inc.

InventorsSandipan Gangopadhyay, Vijayaraj Chakravarthy, Manoj Prasad, Timothy Kearney, Priyanka Moorthy, Dheeraj Jeevagasami, Ryan Britt, Suchisrit Gangopadhyay, Shushanth Masilamani, Vallabh Vijayaraj, Parag Karnik, Suchisman Gangopadhyay, Rachael Patel

Life of this patent

Estimating the term…

patentgrep

Loading patent…

Abstract

A computer-based tool such as computer-implemented methods and systems identify and provide identification of anomalies in insurance claim payments. The computer-based tool receives in memory of a processor, raw claim data from a claim, and reconciles the received raw claim data with at least one entity of a knowledge graph to form reconciled claim data. The tool defines domain specific identities and relationships using the knowledge graph, applies a heuristic rule on the knowledge graph using the reconciled claim data, and creates a prediction of a presence of an anomaly in the reconciled claim data based on the heuristic rule applied to the knowledge graph. Responsive to the prediction of the presence of an anomaly resulting in no anomaly, the tool executes (closes) the claim, or, responsive to the prediction of an anomaly resulting in an anomaly, the tool flags the claim.

Claims20

Dependent claims

A computer-implemented method of automatically identifying anomalies in insurance claims, the method comprising, by a processor:

automatically accessing, in memory of the processor, a knowledge graph comprised of (i) a plurality of nodes representing rules and (ii) a plurality of nodes representing relationships;
auditing a given claim for an anomaly, including automatically producing an initial indication of presence or absence of an anomaly in enriched claim data by:
(a) receiving, in the memory of the processor, textual claim data from the given claim;
(b) normalizing, based on the knowledge graph, the received textual claim data to form normalized claim data;
(c) producing a context vector based on the knowledge graph and the normalized claim data, wherein the context vector is specific to the given claim and indicates a relationship between two or more elements of the normalized claim data;
(d) applying the plurality of nodes representing relationships to the normalized claim data to produce enriched claim data;
and
(e) applying the plurality of nodes representing rules to the enriched claim data and producing an initial indication of presence or absence of an anomaly in the enriched claim data;
generating a risk score representing probability of an anomaly in the enriched claim data by responsively executing a supervised learning model trained on a set of historic claim data indicating presence and absence of anomalies in the historic claim data and-trained on a set of synthetic claim data indicating presence of anomalies in the synthetic claim data, the synthetic claim data being formed from sparse data enhanced in a manner that balances heuristic data and focuses on case data representing relatively high risk of an anomaly;
processing by a meta-model the enriched claim data, the initial indication of presence or absence of an anomaly, and the generated risk score to produce an anomaly score;
and
generating an outcome of the audit of the given claim by:
(i) responsive to the produced anomaly score indicating no anomaly in the given claim, automatically closing the given claim as the outcome of the audit and (ii) responsive to the produced anomaly score indicating presence of an anomaly in the given claim, automatically flagging the given claim as the outcome of the audit.

10 more elements

A computer-based system of automatically identifying anomalies in insurance claims, the system comprising:

a processor;
and
a memory with computer code instructions stored thereon, the processor and the memory, with the computer code instructions, being configured to cause the system to:
automatically access, in the memory, a knowledge graph comprised of (i) a plurality of nodes representing rules and (ii) a plurality of nodes representing relationships;
audit a given claim for an anomaly, including automatically produce an initial indication of presence or absence of an anomaly in enriched claim data by:
(a) receiving, in the memory of the processor, textual claim data from the given claim
(b) normalizing, based on the knowledge graph, the received textual claim data to form normalized claim data;
(c) producing a context vector based on the knowledge graph and the normalized claim data, wherein the context vector is specific to the given claim and indicates a relationship between two or more elements of the normalized claim data;
(d) applying the plurality of nodes representing relationships to the normalized claim data to produce enriched claim data;
and
(e) applying the plurality of nodes representing rules to the enriched claim data and producing an indication of presence or absence of an anomaly in the enriched claim data;
generating a risk score representing probability of an anomaly in the enriched claim data by responsively executing a supervised learning model trained on a set of historic claim data indicating presence and absence of anomalies in the historic claim data and trained on a set of synthetic claim data indicating presence of anomalies in the synthetic claim data, the synthetic claim data being formed from sparse data enhanced in a manner that balances heuristic data and focuses on case data representing relatively high risk of an anomaly;
processing by a meta-model the enriched claim data, the initial indication of presence or absence of an anomaly, and the generated risk score to produce an anomaly score;
and
generating an outcome of the audit of the given claim by:
(i) responsive to the produced anomaly score indicating no anomaly in the given claim, automatically closing the given claim as the outcome of the audit and (ii) responsive to the produced anomaly score indicating presence of an anomaly in the given claim, automatically flagging the given claim as the outcome of the audit.

13 more elements

Dependencies are inferred from the claim text.