Systems and methods for improving textual descriptions using large language models

Patent metadata

Field	Value
Publication number	US-12596876-B2
Application number	US-202318447697-A
Country	US
Kind code	B2
Filing date	Aug 10, 2023
Priority date	Mar 21, 2023
Publication date	Apr 7, 2026
Grant date	Apr 7, 2026

Abstract

A textual description that includes a body of unstructured text is received. Using a rating model configured to output a rating based on a degree of similarity of the received textual description and each of a set of selected textual descriptions, a rating is generated based on the received textual description. The rating model to also used to generate a suggested modification of the received textual description that, when applied to the received textual descriptions, changes the rating of the received textual description. An indication of the suggested modification can be output to a user.

First claim

We claim:

1 . A computer-implemented method comprising:
obtaining, using a large language model, a textual description that includes a body of unstructured text; generating a rating based on the obtained textual description using a rating model, wherein the rating model comprises a machine learning model configured to output a rating based on a degree of similarity evaluated between the obtained textual description and each of a set of selected textual descriptions; generating, using the rating model, a suggested modification of the obtained textual description that, when applied to the obtained textual description, changes the rating of the obtained textual description, wherein the suggested modification is based on the rating output by the rating model; and outputting an indication of the suggested modification via a user interface, wherein generating the suggested modification comprises:
identifying a feature type that is not described in the obtained textual description but that, if added to the obtained textual description, would improve the rating of the obtained textual description;
wherein the feature type is associated with a structured data set; and
wherein the suggested modification includes a recommendation to add the identified feature type to the obtained textual description.

2 . The computer-implemented method of claim 1 , wherein the obtained textual description describes an object and the feature type is associated with the object.

3 . The computer-implemented method of claim 2 , wherein outputting the indication of the suggested modification comprises:
displaying the obtained textual description via the user interface; and displaying a control, at a selected location within the displayed obtained textual description, that is operable to change text related to the identified feature type at the selected location.

4 . The computer-implemented method of claim 1 , wherein generating the suggested modification comprises:
identifying a portion of the obtained textual description that, if replaced by a modified portion of text, would improve the rating of the obtained textual description; wherein the suggested modification includes the modified portion of text.

5 . The computer-implemented method of claim 4 , wherein outputting the indication of the suggested modification comprises:
displaying the obtained textual description via the user interface, wherein the identified portion of the obtained textual description is emphasized in the displayed obtained textual description; and displaying a control that is operable to replace the identified portion of the obtained textual description with the modified portion of text.

6 . The computer-implemented method of claim 1 , further comprising:
generating for respective selected textual descriptions in the set of selected textual descriptions, using the rating model, a uniqueness rating indicating a degree of uniqueness of the respective selected textual description relative to other selected textual descriptions in the set of selected textual description; and determining a range of the uniqueness ratings of the set of selected textual descriptions; wherein scoring the obtained textual description further comprises outputting an assessment of the rating of the obtained textual description relative to the range of the uniqueness ratings of the set of selected textual descriptions.

7 . The computer-implemented method of claim 6 , wherein a higher rating indicates the obtained textual description is more unique relative to the set of selected textual descriptions, and wherein generating the suggested modification of the obtained textual description comprises, if a size of the range of the uniqueness ratings of the set of selected textual descriptions is greater than a threshold size:
in response to determining the rating of the obtained textual description is less than a threshold rating, generating a modification that, when applied to the obtained textual description, would cause the rating of the obtained textual description to increase.

8 . The computer-implemented method of claim 6 , wherein a higher rating indicates the obtained textual description is more similar to the set of selected textual descriptions, and wherein generating the suggested modification of the obtained textual description comprises, if a size of the range of the uniqueness ratings of the set of selected textual descriptions is less than a threshold size:
in response to determining the rating of the obtained textual description is less than a threshold rating, generating a modification that, when applied to the obtained textual description, would cause the rating of the obtained textual description to increase.

9 . The computer-implemented method of claim 1 , wherein the obtained textual description describes a specified object, and wherein the set of selected textual descriptions comprises selected textual descriptions that are related to corresponding entities in a same category as the specified object.

10 . The computer-implemented method of claim 1 , wherein the obtained textual description describes a specified object that is associated with a structured body of data, and wherein the method further comprises:
determining a degree of similarity between the structured body of data associated with the specified object and a plurality of other structured bodies of data corresponding to other entities; and selecting one or more of the other entities for which the structured body of data associated with the selected entities is within a threshold similarity to the structured body of data associated with the specified object; wherein the set of selected textual descriptions comprises textual descriptions corresponding to the selected entities.

11 . The computer-implemented method of claim 1 , further comprising:
measuring performance of a plurality of textual descriptions; and selecting, as the set of selected textual descriptions, one or more of the plurality of textual descriptions based on the measured performance.

12 . The computer-implemented method of claim 1 , wherein outputting the indication of the suggested modification comprises:
automatically regenerating a portion of the obtained textual description; and displaying, via the user interface, a modified version of the obtained textual description including the regenerated portion.

13 . The computer-implemented method of claim 1 , wherein the obtained textual description is received from a user via the user interface.

14 . The computer-implemented method of claim 1 , wherein the obtained textual description is generated at least in part by the large language model.

15 . The computer-implemented method of claim 1 , wherein each of the set of selected textual descriptions include respective bodies of unstructured text.

16 . The computer-implemented method of claim 1 , wherein the textual description includes a description of a first feature of an object, and wherein the suggested modification comprises a suggestion to add a description of a second feature of the object.

17 . A non-transitory computer readable storage medium storing executable instructions, execution of which by a processor causing the processor to:
obtain, using a large language model, a textual description that includes a body of unstructured text; generate a rating based on the obtained textual description using a rating model, wherein the rating model comprises a machine learning model configured to output a rating based on a degree of similarity evaluated be

Assignees

• Shopify Inc

Inventors

• Maschmeyer Russ
• Causley Trisha
• Beauchamp Daniel

Classifications

• G06F40/40 Primary

  Processing or translation of natural language (natural language analysis G06F40/20; semantic analysis G06F40/30) · CPC title

• G06F40/166

  Editing, e.g. inserting or deleting · CPC title

• G06F40/284 Primary

  Lexical analysis, e.g. tokenisation or collocates · CPC title