US 2021/0233519 A1Published application

Adversarial learning framework for persona-based dialogue modeling

AssigneeCapital One Services, LLC

InventorsOluwatobi OLABIYI, Alan SALIMOV, Anish KHAZANE, Erik MUELLER

Life of this patent

Estimating the term…

Loading patent…

Abstract

Various embodiments may be generally directed to the use of an adversarial learning framework for persona-based dialogue modeling. In some embodiments, automated multi-turn dialogue response generation may be performed using a persona-based hierarchical recurrent encoder-decoder-based generative adversarial network (phredGAN). Such a phredGAN may feature a persona-based hierarchical recurrent encoder-decoder (PHRED) generator and a conditional discriminator. In some embodiments, the conditional discriminator may include an adversarial discriminator that is provided with attribute representations as inputs. In some other embodiments, the conditional discriminator may include an attribute discriminator, and attribute representations may be handled as targets of the attribute discriminator. The embodiments are not limited in this context.

Claims20

Dependent claims

A system, comprising:

a generative adversarial network having generator logic, the generator logic including:
a context recurrent neural network (cRNN),
an encoder recurrent neural network (eRNN),
an attention recurrent neural network (aRNN), and
and a decoder recurrent neural network (dRNN);
and
discriminator logic coupled to the generator logic;
a network interface coupled to the generative adversarial network, wherein the network interface is configured to receive a dialogue utterance from a device external to the system;
processing circuitry configured to implement the generator logic and the discriminator logic and coupled to the network interface, wherein the generator logic and the discriminator logic that when executed cause the processing circuitry to:
receive, via the network interface, the dialogue utterance;
generate, using the generator logic, response candidates responsive to the dialogue utterance, wherein:
the cRNN is applied to concatenate a source attribute with an output of the eRNN to generate an initial state for the dRNN, and
the dRNN generates the response candidates based on the initial state;
determine, using the discriminator logic, a response to the dialogue utterance from the response candidates;
and
output, via the network interface, the response to the dialogue utterance to a user device.

13 more elements

A non-transitory, computer-readable medium storing instructions that, when executed, cause one or more processors of one or more servers to implement generator logic and discriminator logic of a generative adversarial network:

generate, using the generator logic of the generative adversarial network, response candidates responsive to a dialogue utterance, wherein the generator logic of the generative adversarial network includes a context recurrent neural network (cRNN), an encoder recurrent neural network (eRNN), an attention recurrent neural network (aRNN), and a decoder recurrent neural network (dRNN), the cRNN utilizes an output of the eRNN to generate an initial state for the dRNN, and the dRNN is applied to generate the response candidates based on the initial state;
determine, via the discriminator logic, a response to the dialogue utterance from the response candidates;
and
output, via the network interface, the response to the dialogue utterance to a user device.

1 more element

A computer-implemented method utilizing a generative adversarial network, the method comprising:

receiving, via a network interface, a dialogue utterance;
generating, using generator logic of the generative adversarial network, response candidates responsive to the dialogue utterance, wherein the generator logic of the generative adversarial network includes a context recurrent neural network (cRNN), an encoder recurrent neural network (eRNN), an attention recurrent neural network (aRNN), and a decoder recurrent neural network (dRNN), the cRNN is configured to concatenate a source attribute with an output of the eRNN to generate an initial state for the dRNN, and the dRNN is configured to use an output of the aRNN to generate the response candidates based on the initial state;
determining, by using discriminator logic, a response to the dialogue utterance from the response candidates;
and
outputting, via the network interface, the response to the dialogue utterance to a user device.

2 more elements

Dependencies are inferred from the claim text.

Facts

Application: 17/228,158Filed 12 Apr 2021
Priority: 26 Sep 2018
Published: 29 Jul 2021
Term ends: Not estimatedThe source records don’t support an estimate.

Classification20 codes

G06F 40/30
Handling natural language data: semantic analysis
G10L 15/16
Speech classification or search: using artificial neural networks
G06F 40/216
Parsing: using statistical methods
G06F 40/35
Semantic analysis: discourse or dialogue representation

Family

Application62/737,089Us provisional application · filed 26 Sep 2018
Application16/560,571Continuation · filed 4 Sep 2019
Application17/228,158Continuation
US 10,978,051 B2Related patent
US 11,705,112 B2Related patent
US 2020/0098353 A1Published application
US 2021/0233519 A1This patent

From the US filing history. Foreign family members aren’t listed.

Timeline

Citations

Open elsewhere

Building an agent or pulling many patents? Use the API or MCP server instead of scraping pages. Get an API key