# US 2023/0419046 A1 · Systems and methods for generating real-time dynamic conversational responses during conversational interactions using machine learning models

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

- Number: US 2023/0419046 A1
- Document ID: US20230419046A1
- Status: Published application
- Kind: A1
- Application number: 17/850,282
- Filed: 2022-06-27
- Published: 2023-12-28
- Earliest claimed: 2022-06-27
- Assignees: Capital One Services, LLC
- Applicants: lCapital One Services, LLC
- Inventors: Md Arafat Hossain KHAN; Isha Chaturvedi; Arturo Hernandez Zeledon; Mohammad Sorower
- Web page: https://patentgrep.com/patent/US20230419046A1

## CPC

- G06F 40/35: Discourse or dialogue representation
- G06F 18/2155: {characterised by the incorporation of unlabelled data, e.g. multiple instance learning [MIL], semi-supervised techniques using expectation-maximisation [EM] or naïve labelling}
- G06N 3/04: Architecture, e.g. interconnection topology
- G06N 3/084: Backpropagation, e.g. using gradient descent
- G06N 5/01: Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound
- G06N 20/00: Machine learning
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

Methods and systems are presented for generating real-time dynamic conversational responses during conversational interactions using machine learning models based on historic intents for a plurality of users and user-specific interactions. The machine learning models comprise a neural network trained to select a first intent from a plurality of intents based on historic data accumulated prior to the conversational interaction and a neural network trained to select a first interaction-specific intent from a plurality of interaction-specific intents based on interaction-specific data for a user.

## Claims

### Claim 1

A system for generating real-time dynamic conversational responses during conversational interactions with messaging applications using machine learning models based on historic intents for a plurality of users and user-specific interactions, the system comprising:
storage circuitry configured to store:
a first machine learning model, wherein the first machine learning model comprises a neural network trained to select a first intent from a plurality of intents based on historic data accumulated prior to conversational interactions; and
a second machine learning model, wherein the second machine learning model comprises a neural network trained to select a first interaction-specific intent from a plurality of interaction-specific intents based on interaction-specific data for a first user accumulated during the conversational interactions;
control circuitry configured to:
receive, from the first user, a first user action during a conversational interaction with a user interface of a messaging application, wherein the conversational interaction comprises an interactive exchange of text messages between the first user and the messaging application;
generate a first feature input based on the first user action;
input the first feature input into the first machine learning model to determine the first intent;
receive, from the first user, a second user action during the conversational interaction;
generate a second feature input based on the second user action and the first intent;
input the second feature input into the second machine learning model to determine the first interaction-specific intent; and
input/output circuitry configured to:
generate for display, at the user interface during the conversational interaction, a first dynamic conversational response based on the first intent, wherein the first dynamic conversational response comprises a first text message; and
generate for display, at the user interface during the conversational interaction, a second dynamic conversational response based on the first interaction-specific intent, wherein the second dynamic conversational response comprises a second text message.

### Claim 2

A method for generating real-time dynamic conversational responses during conversational interactions using machine learning models based on historic intents for a plurality of users and user-specific interactions, the method comprising:
receiving, from a first user, a first user action during a conversational interaction with a user interface;
generating a first feature input based on the first user action;
inputting the first feature input into a first machine learning model, wherein the first machine learning model comprises a neural network trained to select a first intent from a plurality of intents based on historic data accumulated prior to the conversational interaction;
generating for display, at the user interface during the conversational interaction, a first dynamic conversational response based on the first intent;
receiving, from the first user, a second user action during the conversational interaction;
generating a second feature input based on the second user action and the first intent;
inputting the second feature input into a second machine learning model, wherein the second machine learning model comprises a neural network trained to select a first interaction-specific intent from a plurality of interaction-specific intents based on interaction-specific data for the first user accumulated during the conversational interaction; and
generating for display, at the user interface during the conversational interaction, a second dynamic conversational response based on the first interaction-specific intent.

### Claim 3 (depends on 2)

The method of claim 2, further comprising:
determining a first user interface position of the first user action in the user interface; and
retrieving metadata related to the first user interface position, wherein the first feature input is further based on the metadata.

### Claim 4 (depends on 3)

The method of claim 3, further comprising selecting a second user interface position for generating the first dynamic conversational response based on the first user interface position.

### Claim 5 (depends on 2)

The method of claim 2, wherein the first machine learning model comprises a graph neural network trained to predict a next node in an intent graph based on a current node.

### Claim 6 (depends on 2)

The method of claim 2, further comprising, while monitoring for the second user action, monitoring for a third user action during the conversational interaction with the user interface, wherein the third user action corresponds to a user-initiated end to the conversational interaction.

### Claim 7 (depends on 6)

The method of claim 6, further comprising:
in response to detecting the third user action, determining a result of the first user action;
generating a third feature input based on the result;
inputting the third feature input into a third machine learning model, wherein the third machine learning model is trained to determine a level of variance of the result from prior results of prior user actions; and
generating, at the user interface, a push notification based on the level of variance.

### Claim 8 (depends on 6)

The method of claim 6, further comprising:
in response to detecting the third user action, monitoring for a fourth user action, wherein the fourth user action is not received via the user interface, and wherein the fourth user action corresponds to the first user contacting a second user; and
in response to detecting the fourth user action, generating, at a second user interface corresponding to the second user, a recommendation based on the first dynamic conversational response.

### Claim 9 (depends on 8)

The method of claim 8, further comprising:
determining user information corresponding to the recommendation;
determining a network location of the user information; and
generating a network pathway to the user information.

### Claim 10 (depends on 9)

The method of claim 9, further comprising:
automatically retrieving the user information from the network location in response to the fourth user action; and
generating for display the user information on the second user interface.

### Claim 11 (depends on 2)

The method of claim 2, further comprising:
receiving a first labeled feature input, wherein the first labeled feature input is labeled with a first known dynamic conversational response for the first labeled feature input, and wherein the first known dynamic conversational response is based on a known intent for a plurality of users prior to beginning conversational interactions;
training the first machine learning model to classify the first labeled feature input with the first known dynamic conversational response;
receiving a second labeled feature input, wherein the second labeled feature input is labeled with a second known dynamic conversational response for the second labeled feature input, and wherein the second known dynamic conversational response is based on a known interaction-specific intent for a plurality of users after beginning the conversational interactions; and
training the second machine learning model to classify the second labeled feature input with the second known dynamic conversational response.

### Claim 12

A non-transitory, computer-readable medium for generating dynamic conversational responses using machine learning models, comprising instructions that, when executed by one or more processors, cause operations comprising:
receiving, from a first user, a first user action during a conversational interaction with a user interface;
generating a first feature input based on the first user action;
inputting the first feature input into a first machine learning model, wherein the first machine learning model comprises a neural network trained to select a first intent from a plurality of intents based on historic data accumulated prior to the conversational interaction;
generating for display, at the user interface during the conversational interaction, a first dynamic conversational response based on the first intent;
receiving, from the first user, a second user action during the conversational interaction;
generating a second feature input based on the second user action and the first intent;
inputting the second feature input into a second machine learning model, wherein the second machine learning model comprises a neural network trained to select a first interaction-specific intent from a plurality of interaction-specific intents based on interaction-specific data for the first user accumulated during the conversational interaction; and
generating for display, at the user interface during the conversational interaction, a second dynamic conversational response based on the first interaction-specific intent.

### Claim 13 (depends on 12)

The non-transitory, computer-readable medium of claim 12, wherein the instructions further cause operations comprising:
determining a first user interface position of the first user action in the user interface; and
retrieving metadata related to the first user interface position, wherein the first feature input is further based on the metadata.

### Claim 14 (depends on 13)

The non-transitory, computer-readable medium of claim 13, wherein the instructions further cause operations comprising selecting a second user interface position for generating the first dynamic conversational response based on the first user interface position.

### Claim 15 (depends on 12)

The non-transitory, computer-readable medium of claim 12, wherein the first machine learning model comprises a graph neural network trained to predict a next node in an intent graph based on a current node.

### Claim 16 (depends on 12)

The non-transitory, computer-readable medium of claim 12, wherein the instructions further cause operations comprising, while monitoring for the second user action, monitoring for a third user action during the conversational interaction with the user interface, wherein the third user action corresponds to a user-initiated end to the conversational interaction.

### Claim 17 (depends on 16)

The non-transitory, computer-readable medium of claim 16, wherein the instructions further cause operations comprising:
in response to detecting the third user action, determining a result of the first user action;
generating a third feature input based on the result;
inputting the third feature input into a third machine learning model, wherein the third machine learning model is trained to determine a level of variance of the result from prior results of prior user actions; and
generating, at the user interface, a push notification based on the level of variance.

### Claim 18 (depends on 16)

The non-transitory, computer-readable medium of claim 16, wherein the instructions further cause operations comprising:
in response to detecting the third user action, monitoring for a fourth user action, wherein the fourth user action is not received via the user interface, and wherein the fourth user action corresponds to the first user contacting a second user; and
in response to detecting the fourth user action, generating, at a second user interface corresponding to the second user, a recommendation based on the first dynamic conversational response.

### Claim 19 (depends on 18)

The non-transitory, computer-readable medium of claim 18, wherein the instructions further cause operations comprising:
determining user information corresponding to the recommendation;
determining a network location of the user information;
generating a network pathway to the user information;
automatically retrieving the user information from the network location in response to the fourth user action; and
generating for display the user information on the second user interface.

### Claim 20 (depends on 12)

The non-transitory, computer-readable medium of claim 12, wherein the instructions further cause operations comprising:
receiving a first labeled feature input, wherein the first labeled feature input is labeled with a first known dynamic conversational response for the first labeled feature input, and wherein the first known dynamic conversational response is based on a known intent for a plurality of users prior to beginning conversational interactions;
training the first machine learning model to classify the first labeled feature input with the first known dynamic conversational response;
receiving a second labeled feature input, wherein the second labeled feature input is labeled with a second known dynamic conversational response for the second labeled feature input, and wherein the second known dynamic conversational response is based on a known interaction-specific intent for a plurality of users after beginning the conversational interactions; and
training the second machine learning model to classify the second labeled feature input with the second known dynamic conversational response.
