# US 2024/0203424 A1 · Computerized-method and computerized-system for training and applying a machine learning (ml) textual behavioral identification model to authenticate an agent, in a digital multi-channel environment

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

- Number: US 2024/0203424 A1
- Document ID: US20240203424A1
- Status: Published application
- Kind: A1
- Application number: 18/084,573
- Filed: 2022-12-20
- Published: 2024-06-20
- Earliest claimed: 2022-12-20
- Assignees: Nice Ltd.
- Applicants: Nice Ltd.
- Inventors: Matan KERET; Rahul VYAS; Salil DHAWAN; Or GERSHON
- Web page: https://patentgrep.com/patent/US20240203424A1

## CPC

- G10L 17/04: Training, enrolment or model building
- G06F 21/32: using biometric data, e.g. fingerprints, iris scans or voiceprints
- G10L 17/22: Interactive procedures; Man-machine interfaces
- G10L: SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

A computerized-method for training and applying a Machine Learning (ML) textual behavioral-identification-model to authenticate an agent, in a digital multi-channel environment, is provided herein. The computerized-method may include: (i) training a ML-textual-behavioral-identification-model using retrieved textual responses of each agent in one or more historical-interactions which were conducted in a controlled environment, as a training dataset. The ML-textual-behavioral-identification-model may be configured to process the retrieved textual responses of each agent to generate a profile-identity-data for each agent to be used to authenticate identity of the agent; (ii) receiving a textual-response of each agent, when the agent starts an interaction with a customer; (iii) applying the textual-response of the agent to the ML-textual-behavioral-identification-module to authenticate an identity of the agent in real-time by calculating an imposter-probability score for the agent and; (iv) sending the imposter-probability score to a file-management-system to take one or more actions when the imposter-probability score is above a preconfigured-threshold.

## Claims

### Claim 1

A computerized-method for training and applying a Machine Learning (ML) textual behavioral identification model to authenticate an agent, in a digital multi-channel environment, said computerized-method comprising:
training a ML textual behavioral identification model using retrieved textual responses of each agent in one or more historical interactions which were conducted in a controlled environment, as a training dataset,
wherein the ML textual behavioral identification model is configured to process the retrieved textual responses of each agent in one or more historical interactions which were conducted in a controlled environment to generate a profile-identity-data for each agent to be used to authenticate identity of the agent,
wherein the profile-identity-data for each agent is stored in a pattern-identification data store;
receiving a textual response of each agent, when the agent starts an interaction with a customer;
applying the textual response of the agent to the ML textual behavioral identification module to authenticate an identity of the agent in real-time by calculating an imposter-probability score for the agent, based on the received textual response and the generated profile-identity-data of the agent,
wherein the imposter-probability score indicates a probability that the agent is an imposter;
sending the imposter-probability score to a file management system to take one or more actions when the imposter-probability score is above a preconfigured threshold.

### Claim 2 (depends on 1)

The computerized-method of claim 1, wherein the process of the retrieved textual responses of each agent in one or more historical interactions which were conducted in a controlled environment comprising: (i) creating a set of textual units,
wherein the set of textual units is comprised of vocabulary elements, and
wherein each vocabulary element of the vocabulary elements is an n-gram unit which is a contiguous sequence of ‘1’ through ‘n’ vocabulary elements from the received textual response; (ii) finding a distribution of each n-gram; (iii) embedding each vocabulary element using Natural Language Processing (NLP) embedding engine to output a unique vector representation; (iv) arranging each n-gram graph distribution of ‘1’ through ‘n’ n-gram distributions; and (v) using the unique vector representation to create a continues similarity between the textual elements.

### Claim 3 (depends on 1)

The computerized-method of claim 1, wherein a controlled environment is an environment where an agent is verified and authenticated individually.

### Claim 4 (depends on 1)

The computerized-method of claim 1, wherein the one or more historical interactions were conducted via digital text-based channels or voice channels.

### Claim 5 (depends on 1)

The computerized-method of claim 1, wherein the retrieved textual responses are text of text-based interactions or transcripts of voice interactions.

### Claim 6 (depends on 4)

The computerized-method of claim 4, wherein text-based interactions are processed by at least one of: (i) chat server; and (ii) digital or email channel server.

### Claim 7 (depends on 1)

The computerized-method of claim 1, wherein the one or more actions are selected from at least one of: (i) prompting the agent to authenticate; (ii) blocking the agent from handling further interactions; (iii) sending a report or alerts to a supervisor as to the agent imposter-probability score.

### Claim 8 (depends on 7)

The computerized-method of claim 7, wherein the prompted authentication is an application for authentication.

### Claim 9 (depends on 8)

The computerized-method of claim 8, wherein the prompted authentication is voice-biometric.

### Claim 10 (depends on 2)

The computerized-method of claim 2, wherein the arranging of the n-gram distribution is operated by Dijkstra algorithm using Euclidian distance.

### Claim 11 (depends on 2)

The computerized-method of claim 2, wherein the imposter-probability score is calculated by: (i) processing the received textual response to a set of textual units, wherein the set of textual units is comprised of vocabulary elements, wherein each vocabulary element of the vocabulary elements is an n-gram unit which is a contiguous sequence of ‘1’ through ‘n’ vocabulary elements from the received textual response; (ii) finding a distribution of each n-gram; (iii) embedding each vocabulary element by using Natural Language Processing (NLP) embedding engine to output a unique vector representation (iv) arranging each n-gram distribution graph of the ‘1’ through ‘n’ n-gram distributions; (v) using the unique vector representation to create a continues similarity between the textual elements; (vi) calculating an average distance between each vector of the unique vector representation of the profile-identity-data, and each vector of the unique vector representation of the received textual response to yield an imposter-probability score.

### Claim 12 (depends on 11)

The computerized-method of claim 11, wherein the calculating of the average distance is operated by an Earth Movers Distance (EMD) algorithm.

### Claim 13 (depends on 10)

The computerized-method of claim 10, wherein when the imposter-probability score is below a preconfigured threshold it indicates that the agent is authenticated and the received textual response is used for retraining or enriching the ML textual behavioral identification module.

### Claim 14 (depends on 1)

The computerized-method of claim 1, wherein every preconfigured time the ML textual behavioral identification model is retrained or enriched by a textual response of each agent during an interaction which has been conducted in a controlled environment, wherein the agent has been authenticated during the interaction by an authentication application.

### Claim 15

A computerized-system for training and applying a Machine Learning (ML) textual behavioral identification model to authenticate an agent, in a digital multi-channel environment, said computerized-system comprising:
one or more processors; and
a pattern-identification data store;
said one or more processors are configured to:
train a ML textual behavioral identification model using retrieved textual responses of each agent in one or more historical interactions which were conducted in a controlled environment, as a training dataset,
wherein the ML textual behavioral identification model is configured to process the retrieved textual responses of each agent in one or more historical interactions which were conducted in a controlled environment to generate a profile-identity-data for each agent to be used to authenticate identity of the agent,
wherein the profile-identity-data for each agent is stored in a pattern-identification data store;
receive a textual response of each agent when the agent starts an interaction with a customer and forward the textual response to the ML textual behavioral identification module;
the ML textual behavioral identification module is configured to apply the textual response of the agent to authenticate an identity of agent in real-time by calculating an imposter-probability score for the agent, based on the received textual response and the generated profile-identity-data of the agent that is stored in the pattern-identification data store, and forward the imposter-probability score to a file management system;
wherein the imposter-probability score indicates a probability that the agent is an imposter, and wherein one or more processors in
said file management system are configured to take one or more actions when the imposter-probability score is above a preconfigured threshold.

### Claim 16 (depends on 15)

The computerized-system of claim 15, wherein the file management system is further configured to detect aged training data of an agent for disposal and communicate with the ML textual behavioral identification module to retrain based on recent data of the agent.
