# US 2026/0065055 A1 · Systems and methods for generating dynamic conversational responses using deep conditional learning

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

- Number: US 2026/0065055 A1
- Document ID: US20260065055A1
- Status: Published application
- Kind: A1
- Application number: 19/384,531
- Filed: 2025-11-10
- Published: 2026-03-05
- Earliest claimed: 2021-05-11
- Assignees: Capital One Services, LLC
- Applicants: Capital One Services, LLC
- Inventors: Minh Le; Rui Zhang; Jonathan Gross
- Web page: https://patentgrep.com/patent/US20260065055A1

## CPC

- G06F 40/35: Discourse or dialogue representation
- G06N 3/08: Learning methods
- G06F 16/90332: {Natural language query formulation or dialogue systems}
- G06F 18/214: Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06F 18/24133: {Distances to prototypes}
- G06F 18/295: {Markov models or related models, e.g. semi-Markov models; Markov random fields; Networks embedding Markov models}
- G06F 40/30: Semantic analysis
- G06N 3/045: Combinations of networks
- G06N 3/0499: Feedforward networks
- G06N 3/09: Supervised learning
- G06N 3/096: Transfer learning
- G06N 5/022: Knowledge engineering; Knowledge acquisition
- G06F 40/279: Recognition of textual entities
- G06N 3/084: Backpropagation, e.g. using gradient descent
- 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 described herein for generating dynamic conversational responses. Conversational responses include communications between a user and a system that may maintain a conversational tone, cadence, or speech pattern similar to a human during an interactive exchange between the user and the system. The interactive exchange may include the system responding to one or more user actions (which may include user inactions), and/or predicting responses prior to receiving a user action

## Claims

### Claim 1

A method for generating dynamic conversational responses using conditional deep learning, the method comprising:
receiving user data in response to a user interacting with a user interface;
generating a feature input based on the user data;
inputting the feature input into a first neural network, wherein the first neural network is trained based on a first training regime and a second training regime, wherein the first training regime comprises training an initial version of the first neural network based on a first data set, wherein the second training regime comprises further training the initial version based on a second data set and using a second neural network as a constraint, wherein parameters of the second neural network remain unchanged during the second training regime, and wherein the second neural network is trained based on a third data set;
receiving an output from the first neural network; and
generating for display, on the user interface, a first dynamic conversational response based on the output, wherein the first dynamic conversational response corresponds to a first probable intent of the user.

### Claim 2 (depends on 1)

The method of claim 1, wherein the first neural network probabilistic programming comprises a programming paradigm in which probabilistic models are specified and inference for these models is performed automatically.

### Claim 3 (depends on 2)

The method of claim 2, wherein receiving the output from the first neural network further comprises:
determining interdependencies between variables in the first neural network using stochastic variable inference; and
generating a unified probability distribution based on the output.

### Claim 4 (depends on 2)

The method of claim 2, wherein receiving the output from the first neural network further comprises:
generating a softmax output comprising a multivariate distribution of probable intents of the user; and
applying random changes to weight parameters of the first neural network based on the softmax output.

### Claim 5 (depends on 2)

The method of claim 2, wherein class probabilities of the second neural network are used as targets while training the first neural network.

### Claim 6 (depends on 2)

The method of claim 2, wherein the first data set comprises data over a first time period and having a first size, the second data set comprises data over a second time period and having a second size, and wherein the second time period and the second size are greater than the first time period and the first size, respectively.

### Claim 7 (depends on 2)

The method of claim 2, wherein training the initial version of the first neural network based on the first data set and using knowledge distillation further comprises:
training an original model using the first data set; and
training the initial version of the first neural network to mimic predictions made by the original model by minimizing a loss function in which a target is a distribution of class probabilities predicted by the original model.

### Claim 8 (depends on 2)

The method of claim 2, wherein the output comprises a multivariate distribution of probable intents of the user, and wherein generating the first dynamic conversational response based on the output comprises selecting the first dynamic conversational response from a plurality of dynamic conversational responses.

### Claim 9 (depends on 2)

The method of claim 2, further comprising:
generating for display, on the user interface, simultaneously with the first dynamic conversational response, a second dynamic conversational response;
receiving a user selection of the first dynamic conversational response; and
in response to the user selection, updating a multivariate distribution of probable intents of the user to reflect a probability that the user has a second probable intent, of the probable intents, based on the feature input, is zero.

### Claim 10 (depends on 2)

The method of claim 2, wherein the first data set comprises data on a plurality of users, and wherein the second data set comprises feedback data on actual intents of the user.

### Claim 11 (depends on 2)

The method of claim 2, wherein the first data set comprises data on a plurality of users having a first characteristic, and wherein inputting the feature input into the first neural network is done in response to:
determining, based on the user data, that the user has the first characteristic; and
selecting the first neural network, from a plurality of neural networks, based on determining that the user has the first characteristic.

### Claim 12 (depends on 2)

The method of claim 2, wherein the second neural network comprises an output layer for the initial version during the second training regime.

### Claim 13

One or more non-transitory, computer-readable media for generating dynamic conversational responses using conditional deep learning, comprising instructions that, when executed on one or more processors, cause operations comprising:
receiving user data in response to a user interacting with a user interface;
generating a feature input based on the user data;
inputting the feature input into a first neural network, wherein the first neural network is trained based on a first training regime and a second training regime, wherein the first training regime comprises training an initial version of the first neural network based on a first data set and using knowledge distillation to mimic predictions made by a first model by minimizing a loss function in which a target is a distribution of class probabilities predicted by the first model;
receiving an output from the first neural network; and
generating for display, on the user interface, a first dynamic conversational response based on the output, wherein the first dynamic conversational response corresponds to a first probable intent of the user.

### Claim 14 (depends on 13)

The one or more non-transitory, computer-readable media of claim 13, wherein the first data set comprises data over a first time period and having a first size.

### Claim 15 (depends on 13)

The one or more non-transitory, computer-readable media of claim 13, wherein training the initial version of the first neural network based on the first data set and the using knowledge distillation further comprises:
training an original model using the first data set; and
training the initial version of the first neural network to mimic predictions made by the original model.

### Claim 16 (depends on 13)

The one or more non-transitory, computer-readable media of claim 13, wherein the output comprises a multivariate distribution of probable intents of the user, and wherein generating the first dynamic conversational response based on the output comprises selecting the first dynamic conversational response from a plurality of dynamic conversational responses.

### Claim 17

The one or more non-transitory, computer-readable media of 13, wherein the instructions further cause operations comprising:
generating for display, on the user interface, simultaneously with the first dynamic conversational response, a second dynamic conversational response;
receiving a user selection of the first dynamic conversational response; and
in response to the user selection, updating a multivariate distribution of probable intents of the user to reflect a probability of the user having a second probable intent, of the probable intents, based on the feature input, is zero.

### Claim 18 (depends on 13)

The one or more non-transitory, computer-readable media of claim 13, wherein first data set comprises data on a plurality of users.

### Claim 19 (depends on 13)

The one or more non-transitory, computer-readable media of claim 13, wherein the first data set comprises data on a plurality of users having a first characteristic, and wherein inputting the feature input into the first neural network is done in response to:
determining, based on the user data, that the user has the first characteristic; and
selecting the first neural network, from a plurality of neural networks, based on determining that the user has the first characteristic.

### Claim 20 (depends on 13)

The one or more non-transitory, computer-readable media of claim 13, wherein the first neural network comprises a second neural network, wherein the second neural network comprises an output layer for the initial version during the second training regime.
