Systems and methods for accelerating a disposition of digital dispute events in a machine learning-based digital threat mitigation platform

Patent metadata

Field	Value
Publication number	US-12143402-B2
Application number	US-202418416883-A
Country	US
Kind code	B2
Filing date	Jan 18, 2024
Priority date	Nov 9, 2021
Publication date	Nov 12, 2024
Grant date	Nov 12, 2024

Abstract

A system and method for accelerating a disposition of a digital dispute event includes routing a digital dispute event to one of a plurality of distinct machine learning-based dispute scoring models; computing, by the one of the plurality of distinct machine learning-based dispute scoring models, a preliminary machine learning-based dispute inference based on one or more features extracted from the digital dispute event, wherein the preliminary machine learning-based dispute inference relates to a probability of the subscriber prevailing against the digital dispute event based on each piece of evidence data of a service-proposed corpus of evidence data being available to include in a dispute response artifact; and generating the dispute response artifact based on the digital dispute event, wherein the generating includes installing one or more obtainable pieces of evidence data associated with the digital event into one or more distinct sections of the dispute response artifact.

First claim

We claim:

1. A computer-implemented method comprising:
obtaining, via one or more computers, a digital dispute event associated with a subscriber;
classifying, using a dispute classification machine learning model, the digital dispute event to a digital dispute event type of a plurality of predetermined digital dispute event types based on one or more features extracted from the digital dispute event;
computing, via the one or more computers, a historical success rate of the subscriber in historical digital dispute events of the subscriber that correspond to the digital dispute event type;
routing, via the one or more computers, the digital dispute event to one of (i) a subscriber-specific machine learning model, (ii) a subscriber-agnostic machine learning model based on the historical success rate of the subscriber and (iii) a target review queue, wherein:
the digital dispute event is routed to the subscriber-specific machine learning model when the historical success rate of the subscriber indicates the subscriber historically prevails in the historical digital dispute events of the digital dispute event type,
the digital dispute event is routed to the subscriber-agnostic machine learning model when the historical success rate of the subscriber indicates the subscriber historically underperforms in the historical digital dispute events of the digital dispute event type; and
the digital dispute event is routed to the target user review queue based on a dispute score value of the dispute response inference failing to satisfy a predetermined minimum dispute score threshold;

computing, by the one of the subscriber-specific machine learning model and the subscriber-agnostic machine learning model, a dispute response evidence recommendation inference in response to routing the digital dispute event;
generating, by the one or more computers, a dispute response artifact based on each piece of recommended evidence included in the dispute response evidence recommendation inference; and
transmitting, by the one or more computers, the dispute response artifact to a target entity based on the dispute response artifact satisfying a dispute response submittal criterion.

2. The computer-implemented method according to claim 1 , wherein:
the subscriber-specific machine learning model is trained on historical dispute response data of the subscriber, and
the subscriber-agnostic machine learning model is trained on historical dispute response data of a plurality of distinct subscribers to a digital threat mitigation service.

3. The computer-implemented method according to claim 2 , wherein:
if the digital dispute event is routed to the subscriber-specific machine learning model, the dispute response evidence recommendation inference includes a first set of recommended evidence to include in the dispute response artifact,
if the digital dispute event is routed to the subscriber-agnostic machine learning model, the dispute response evidence recommendation inference includes a second set of recommended evidence to include in the dispute response artifact, and
the second set of recommended evidence includes different evidence recommendations than the first set of recommended evidence.

4. The computer-implemented method according to claim 3 , wherein:
generating the dispute response artifact includes:
automatically generating a partially completed dispute response artifact using a set of available evidence automatically sourced by the digital threat mitigation service,
automatically surfacing, via a graphical user interface, the partially completed dispute response artifact to the subscriber; and
visually highlighting, via the graphical user interface, one or more dispute response sections of the partially completed dispute response artifact that is missing one or more pieces of evidence recommended by the digital threat mitigation service.

5. The computer-implemented method according to claim 2 , wherein:
the subscriber-specific machine learning model is one of a plurality of distinct subscriber-specific machine learning models of the digital threat mitigation service, wherein each of the plurality of distinct subscriber-specific machine learning models is trained on historical dispute response data of a distinct subscriber, and
the subscriber-agnostic machine learning model is one of a plurality of distinct subscriber-agnostic machine learning models of the digital threat mitigation service, wherein each of the plurality of distinct subscriber-agnostic machine learning models is further trained on historical dispute response data that involves a distinct digital event processor.

6. The computer-implemented method according to claim 1 , further comprising:
displaying, on a web-based user interface, the dispute response artifact and a dispute response insights user interface element, wherein the dispute response insight user interface element indicates one or more probative pieces of evidence data missing in the dispute response artifact that, if included in the dispute response artifact, increases a likelihood of the subscriber prevailing against the digital dispute event.

7. The computer-implemented method according to claim 6 , further comprising:
visually emphasizing, on the web-based user interface, a selective subset of distinct dispute response sections of the dispute response artifact that map to the one or more probative pieces of evidence data missing in the dispute response artifact, and
wherein visually emphasizing the selective subset of distinct dispute response sections indicates to the subscriber where to include one or more additional pieces of evidence data.

8. The computer-implemented method according to claim 1 , further comprising:
computing, by a target one of the subscriber-specific machine learning model and the subscriber-agnostic machine learning model, an updated dispute inference based on a current state of evidence data included in the dispute response artifact.

9. The computer-implemented method according to claim 8 , wherein:
the digital dispute event is associated with a digital transaction event that occurred between an online user and the subscriber,
computing the updated dispute inference further includes:
(a) using the subscriber-specific machine learning model when the subscriber-specific machine learning model is trained on a bank identification number (BIN) used in the digital transaction event,
(b) using the subscriber-specific machine learning model when the subscriber-specific machine learning model is trained on a card type used in the digital transaction event,
(c) using the subscriber-agnostic machine learning model when (c-i) the subscriber-specific machine learning model is not trained on the bank identification number (BIN) used in the digital transaction event and (c-ii) the subscriber-agnostic machine learning model is trained on the bank identification number (BIN) used in the digital transaction event, and
(d) using the subscriber-agnostic machine learning model when (d-i) the subscriber-specific machine learning model is not trained on the card type used in the digital transaction event and (d-ii) the subscriber-agnostic machine learning model is trained on the card type used in the digital transaction event.

10. A computer-implemented method comprising:
obtaining, via one or more computers, a digital dispute event associated with a subscriber;
classifying, using a dispute classification machine learning model, the digital dispute event to a digital dispute event type of a plurality of predetermined digital dispute event types based on a first set of features extracted from the digital dispute event;
computing, via the one or more computers, a

Assignees

• Sift Science Inc

Inventors

• St Pierre Eric
• Forbess Alex

Classifications

• G06Q20/4016

  involving fraud or risk level assessment in transaction processing · CPC title

• G06Q20/409

  Device specific authentication in transaction processing · CPC title

• H04L63/1408 Primary

  by monitoring network traffic (monitoring network traffic per se H04L43/00) · CPC title