IP Library › Granted Patent US 12,688,373

Granted Patent B2

US 12,688,373 · App. 18/050,285 · Granted Jul 21, 2026

Generating subject lines from keywords utilizing a machine-learning model

Inventors: Suofei Wu (Fremont, CA); Jun He (Fremont, CA); Zhenyu Yan (Cupertino, CA)

Assignee: Adobe Inc.

G06F40/40 G06F40/295 G06N3/08

View Patent ↗

Loading inventors, assignments & file history…

Monitor This Case

Get email alerts when status or documents change.

Order Certified Copies

Most orders are placed with the USPTO same day — all within 24 business hours.

Order via The Patent Place →

Pre-filled with this patent's details

Quick Facts

Patent No.: US 12,688,373
App. No.: 18/050,285
Granted: Jul 21, 2026
Kind: B2

External Links

USPTO Patent Center↗ Full Patent PDF↗

Abstract

The present disclosure relates to systems, methods, and non-transitory computer readable media that utilize machine learning to generate subject lines from subject line keywords. In one or more embodiments, the disclosed systems receive, from a client device, one or more subject line keywords. Additionally, the disclosed systems generate, utilizing a subject generation machine-learning model having learned parameters, a subject line by selecting one or more words for the subject line from a word distribution based on the one or more subject line keywords. The disclosed systems further provide, for display on the client device, the subject line.

Claims (53)

1 . A system comprising:

one or more memory components; and

one or more processing devices coupled to the one or more memory components, the one or more processing devices to perform operations comprising:

receiving training subject line keywords extracted from training subject lines by a named entity recognition model that includes a transformer-based bidirectional neural network;

determining, using an emotion classification neural network, one or more training emotions from the training subject lines;

generating, utilizing a sequence-to-sequence neural network having an encoder-decoder architecture, predicted subject lines from the training subject line keywords extracted by the named entity recognition model and the one or more training emotions determined by the emotion classification neural network; and

training, based on comparisons of the predicted subject lines and the training subject lines, the sequence-to-sequence neural network to generate a plurality of subject lines in response to receiving user input including a creativity score, a number of subject lines to create, and one or more subject line keywords for creating the number of subject lines.

2 . The system of claim 1 , wherein the operations further comprise determining parameters of the named entity recognition model by:

generating, utilizing the named entity recognition model, predicted keywords from one or more training subject lines; and

modifying the parameters of the named entity recognition model based on comparisons of the predicted keywords with labels associated with the one or more training subject lines.

3 . The system of claim 1 , wherein:

training the sequence-to-sequence neural network to generate the plurality of subject lines based on the comparisons of the predicted subject lines and the training subject lines comprises iteratively modifying parameters of the sequence-to-sequence neural network based on the comparisons of the predicted subject lines and a first subset of training subject lines from the training subject lines; and

the operations further comprise training the emotion classification neural network to determine the one or more training emotions uses a second subset of training subject lines from the training subject lines that differs from the first subset of training subject lines.

4 . The system of claim 1 , wherein:

receiving the training subject line keywords extracted from the training subject lines by the named entity recognition model comprises receiving custom subject line keywords extracted from the training subject lines by the named entity recognition model; and

generating, utilizing the sequence-to-sequence neural network, the predicted subject lines from the training subject line keywords extracted by the named entity recognition model comprises generating, utilizing the sequence-to-sequence neural network, the predicted subject lines from the custom subject line keywords extracted by the named entity recognition model.

5 . The system of claim 1 , wherein the operations further comprise implementing the sequence-to-sequence neural network trained using the training subject line keywords extracted by the named entity recognition model by:

receiving, from a client device, user input including at least one creativity score, at least one number of subject lines to create, and at least one subject line keyword for creating the at least one number of subject lines; and

generating, utilizing the sequence-to-sequence neural network, a plurality of subject lines that corresponds to the at least one number of subject lines by selecting one or more words for each subject line from a word distribution based on the at least one creativity score and the at least one subject line keyword.

6 . A method comprising:

receiving training subject line keywords extracted from training subject lines by a named entity recognition model that includes a transformer-based bidirectional neural network;

determining, using an emotion classification neural network, one or more training emotions from the training subject lines;

7 . The method of claim 6 , further comprising determining parameters of the named entity recognition model by:

generating, utilizing the named entity recognition model, predicted keywords from one or more training subject lines; and

modifying the parameters of the named entity recognition model based on comparisons of the predicted keywords with labels associated with the one or more training subject lines.

8 . The method of claim 6 ,

wherein training the sequence-to-sequence neural network to generate the plurality of subject lines based on the comparisons of the predicted subject lines and the training subject lines comprises iteratively modifying parameters of the sequence-to-sequence neural network based on the comparisons of the predicted subject lines and a first subset of training subject lines from the training subject lines; and

further comprising training the emotion classification neural network to determine the one or more training emotions uses a second subset of training subject lines from the training subject lines that differs from the first subset of training subject lines.

9 . The method of claim 6 , wherein:

10 . The method of claim 6 , further comprising implementing the sequence-to-sequence neural network trained using the training subject line keywords extracted by the named entity recognition model by:

11 . A non-transitory computer-readable medium storing executable instructions, which when executed by a processing device, cause the processing device to perform operations comprising:

receiving training subject line keywords extracted from training subject lines by a named entity recognition model that includes a transformer-based bidirectional neural network;

determining, using an emotion classification neural network, one or more training emotions from the training subject lines;

12 . The non-transitory computer-readable medium of claim 11 , wherein the operations further comprise determining parameters of the named entity recognition model by:

generating, utilizing the named entity recognition model, predicted keywords from one or more training subject lines; and

modifying the parameters of the named entity recognition model based on comparisons of the predicted keywords with labels associated with the one or more training subject lines.

13 . The non-transitory computer-readable medium of claim 11 ,

14 . The non-transitory computer-readable medium of claim 11 , wherein:

15 . The non-transitory computer-readable medium of claim 11 , wherein the operations further comprise implementing the sequence-to-sequence neural network trained using the training subject line keywords extracted by the named entity recognition model by: