Granted Patent B2

US 12,488,368 · App. 18/582,839 · Granted Dec 2, 2025

Inventor: Michael Montero (Palo Alto, CA)

Assignee: Maplebear Inc.

G06Q30/0244G06Q30/0201G06Q30/0239G06Q30/0271G06N20/00

Abstract

Claims (54)

1 . A method comprising:

accessing, by one or more processors, a first set of candidate content items, wherein each candidate content item in the first set of candidate content items includes a variable value for each of a plurality of variables;

generating, by one or more processors, a content item vector for each content item in the first set of candidate content items, wherein each content item vector includes the variable values for the corresponding candidate content item of the first set of candidate content items, and wherein each content item vector includes a success metric for the candidate content item;

generating, by one or more processors, a derivative table based on the generated content item vectors, wherein the derivative table comprises a plurality of differential vectors, wherein each differential vector comprises pair-wise differences between variable values for a pair of the generated content item vectors and a differential success score representing a difference of the success metrics of the pair of generated content item vectors;

training, by one or more processors, a machine-learning model based on the derivative table;

receiving a content item from a client device, wherein the content item includes a variable value for each of the plurality of variables;

generating, by one or more processors, a content item vector for the content item, wherein the content item vector includes the variable values for the content item;

computing, by one or more processors, a score for the content item by applying the machine-learning model to the content item vector for the content item; and

responsive to the computed score exceeding a threshold, transmitting, by one or more processors, the content item to a plurality of users.

2 . The method of claim 1 , further comprising:

selecting an additional set of content items by computing an orthogonality of the content item vector to the content item vectors of the additional set of content items; and

transmitting the additional set of content items to the plurality of users.

3 . The method of claim 2 , wherein computing an orthogonality of the content item vectors comprises:

generating a matrix comprising variable values corresponding to the content item and variable values corresponding to the additional set of content items; and

optimizing for a maximum determinant of the matrix by selecting the additional set of content items to maximize the determinant of the matrix.

4 . The method of claim 1 , further comprising:

computing a score for a plurality of content items by applying the machine-learning model to content item vectors for the plurality of content items;

ranking the plurality of content items based on the computed scores; and

selecting the content item based on the ranking.

5 . The method of claim 1 , wherein transmitting the content item to a plurality of users comprises transmitting the content item to a set of segmented users.

6 . The method of claim 1 , wherein the machine-learning model comprises a neural network or a decision tree.

7 . The method of claim 1 , wherein the machine-learning model is trained by:

receiving an initial set of content items; and

generating a set of content items for training the machine-learning model by applying a set of heuristics to filter the initial set of content items.

8 . The method of claim 7 , wherein the initial set of content items comprises a content item for each possible combination of variable values for the plurality of variables.

9 . The method of claim 7 , wherein the set of heuristics comprises a user-defined heuristic.

10 . The method of claim 1 , wherein the success metric for a content item is based on a clip rate of the content item.

11 . A non-transitory computer-readable medium storing instructions that, when executed by a processor, cause the processor to perform operations comprising:

accessing, by one or more processors, a first set of candidate content items, wherein each candidate content item in the first set of candidate content items includes a variable value for each of a plurality of variables;

generating, by one or more processors, a content item vector for each content item in the first set of candidate content items, wherein each content item vector includes the variable values for the corresponding candidate content item of the first set of candidate content items, and wherein each content item vector includes a success metric for the candidate content item;

generating, by one or more processors, a derivative table based on the generated content item vectors, wherein the derivative table comprises a plurality of differential vectors, wherein each differential vector comprises pair-wise differences between variable values for a pair of the generated content item vectors and a differential success score representing a difference of the success metrics of the pair of generated content item vectors;

training, by one or more processors, a machine-learning model based on the derivative table;

receiving a content item from a client device, wherein the content item includes a variable value for each of the plurality of variables;

generating, by one or more processors, a content item vector for the content item, wherein the content item vector includes the variable values for the content item;

computing, by one or more processors, a score for the content item by applying the machine-learning model to the content item vector for the content item; and

responsive to the computed score exceeding a threshold, transmitting, by one or more processors, the content item to a plurality of users.

12 . The computer-readable medium of claim 11 , the operations further comprising:

selecting an additional set of content items by computing an orthogonality of the content item vector to the content item vectors of the additional set of content items; and

transmitting the additional set of content items to the plurality of users.

13 . The computer-readable medium of claim 12 , wherein computing an orthogonality of the content item vectors comprises:

generating a matrix comprising variable values corresponding to the content item and variable values corresponding to the additional set of content items; and

optimizing for a maximum determinant of the matrix by selecting the additional set of content items to maximize the determinant of the matrix.

14 . The computer-readable medium of claim 11 , the operations further comprising:

computing a score for a plurality of content items by applying the machine-learning model to content item vectors for the plurality of content items;

ranking the plurality of content items based on the computed scores; and

selecting the content item based on the ranking.

15 . The computer-readable medium of claim 11 , wherein transmitting the content item to a plurality of users comprises transmitting the content item to a set of segmented users.

16 . The computer-readable medium of claim 11 , wherein the machine-learning model comprises a neural network or a decision tree.

17 . The computer-readable medium of claim 11 , wherein the machine-learning model is trained by:

receiving an initial set of content items; and

generating a set of content items for training the machine-learning model by applying a set of heuristics to filter the initial set of content items.

18 . The computer-readable medium of claim 17 , wherein the initial set of content items comprises a content item for each possible combination of variable values for the plurality of variables.

19 . The computer-readable medium of claim 17 , wherein the set of heuristics comprises a user-defined heuristic.

20 . The computer-readable medium of claim 11 , wherein the success metric for a content item is based on a clip rate of the content item.

Assignments (2)

Continuity (4)

References Cited (307)