Granted Patent B2

US 12,743,706 · App. 18/240,301 · Granted Sep 22, 2026

Inventors: Ruoruo Zhang (Santa Clara, CA); Christopher Benjamin (Larchmont, NY); Michael Sandare (New York, NY); Lin Gao (Miami, FL); Graham Beauregard (Burlington, CA)

Assignee: Maplebear Inc

G06Q30/0185G06Q20/208

Abstract

Claims (77)

1 . A method comprising, at a computer system comprising a processor and a computer-readable medium:

gathering, by the computer system, data associated with a physical activity of a user, the user being physically present at a location of a retailer and adding a set of items to a physical receptacle, the computer system in communication with the physical receptacle, the data including information about one or more weight changes associated with one or more items of the set of items added to the physical receptacle without a scanning performed for the one or more items by the user;

accessing a machine-learning model trained to detect a fraudulent behavior of the user associated with the physical activity;

applying the machine-learning model to the gathered data including the information about the one or more weight changes associated with the one or more items added to the physical receptacle without the scanning to:

detect one or more anomalies during the physical activity of the user, the one or more anomalies being indicative of the fraudulent behavior, and

generate a score for each of the one or more detected anomalies that is indicative of a likelihood that each of the one or more detected anomalies is associated with the fraudulent behavior;

generating, using the score for each of the one or more detected anomalies, a total score for the one or more detected anomalies;

determining, based on the total score being higher than a threshold score, that the one or more detected anomalies satisfy a threshold condition for the fraudulent behavior;

in response to determining that the one or more detected anomalies satisfy the threshold condition for the fraudulent behavior, sending a communication to a management system of the retailer to cause at least one remedial action to be performed before the user physically leaves the location of the retailer;

determining whether the one or more anomalies detected by the machine-learning model were correct;

generating training data based on information whether the one or more detected anomalies were correct; and

retraining the machine-learning model using the generated training data.

2 . The method of claim 1 , wherein gathering the data further comprises gathering at least one of: information about a location of the physical receptacle at the location of the retailer when each item of the set of items is added to the physical receptacle, video data associated with an activity in the physical receptacle, or video data associated with an activity around the physical receptacle.

3 . The method of claim 1 , wherein gathering the data further comprises measuring each item of the set of items when being added to the physical receptacle using at least one sensor mounted to the physical receptacle.

4 . The method of claim 1 , wherein applying the machine-learning model comprises:

applying the machine-learning model further to information about a weight mismatch between a weight change of the physical receptacle and a weight of an item being scanned by the user to detect the one or more anomalies and generate the score for each of the one or more detected anomalies.

5 . The method of claim 1 , wherein applying the machine-learning model further comprises:

detecting an intent of the user based on the gathered data; and

detecting the one or more anomalies based on the detected intent.

6 . The method of claim 1 , further comprising:

identifying at least one item of the set of items added to the physical receptacle or at least one item removed from the physical receptacle,

wherein applying the machine-learning model further comprises applying the machine-learning model to detect the one or more anomalies based on identifying at least one of the at least one added item or the at least one removed item.

7 . The method of claim 1 , further comprising:

generating a message for the user, in response to the machine-learning model detecting the one or more anomalies associated with an item added to the physical receptacle;

responsive to generating the message, causing a device associated with the user to display a user interface with the message; and

tracking a weight for the item added to the physical receptacle, based on the user responding to the message.

8 . The method of claim 1 , further comprising:

setting the threshold condition for the fraudulent behavior based on at least one of:

information about the user, a likelihood of the fraudulent behavior associated with the location of the retailer, a total monetary value of the one or more items added to the physical receptacle, or a defined threshold condition set by the retailer.

9 . The method of claim 1 , wherein sending the communication to the management system comprises:

flagging the user for an audit process, based on determining that the one or more detected anomalies satisfy the threshold condition for the fraudulent behavior.

10 . The method of claim 9 , wherein sending the communication to the management system further comprises:

determining, during the audit process, that a first set of items the user paid for does not match a second set of items the user is attempting to remove from the location of the retailer; and

sending the communication to the management system to cause the at least one remedial action to be performed, based on determining that the first set of items does not match the second set of items.

11 . The method of claim 9 , wherein sending the communication to the management system further comprises:

triggering the audit process over a network, in response to flagging the user for the audit process.

12 . The method of claim 9 , wherein sending the communication to the management system further comprises:

sending a signal to a human reviewer at the location of the retailer to conduct the audit process at the location of the retailer, in response to flagging the user for the audit process.

13 . A computer program product comprising a non-transitory computer readable storage medium having instructions encoded thereon that, when executed by a processor, cause the processor to perform steps comprising:

gathering, by a computer system, data associated with a physical activity of a user, the user being physically present at a location of a retailer and adding a set of items to a physical receptacle, the computer system in communication with the physical receptacle, the data including information about one or more weight changes associated with one or more items of the set of items added to the physical receptacle without a scanning performed for the one or more items by the user;

accessing a machine-learning model trained to detect a fraudulent behavior of the user associated with the physical activity;

applying the machine-learning model to the gathered data including the information about the one or more weight changes associated with the one or more items added to the physical receptacle without the scanning to:

detect one or more anomalies during the physical activity of the user, the one or more anomalies being indicative of the fraudulent behavior, and

generate a score for each of the one or more detected anomalies that is indicative of a likelihood that each of the one or more detected anomalies is associated with the fraudulent behavior;

generating, using the score for each of the one or more detected anomalies, a total score for the one or more detected anomalies;

determining, based on the total score being higher than a threshold score, that the one or more detected anomalies satisfy a threshold condition for the fraudulent behavior;

in response to determining that the one or more detected anomalies satisfy the threshold condition for the fraudulent behavior, sending a communication to a management system of the retailer to cause at least one remedial action to be performed before the user physically leaves the location of the retailer;

determining whether the one or more anomalies detected by the machine-learning model were correct;

generating training data based on information whether the one or more detected anomalies were correct; and

retraining the machine-learning model using the generated training data.

14 . The computer program product of claim 13 , wherein the instructions further cause the processor to perform steps comprising:

applying the machine-learning model further to information about a weight mismatch between a weight change of the physical receptacle and a weight of an item being scanned by the user to detect the one or more anomalies and generate the score for each of the one or more detected anomalies.

15 . The computer program product of claim 13 , wherein the instructions further cause the processor to perform steps comprising:

generating a message for the user, in response to the machine-learning model detecting the one or more anomalies associated with an item added to the physical receptacle;

responsive to generating the message, causing a device associated with the user to display a user interface with the message; and

tracking a weight for the item added to the physical receptacle, based on the user responding to the message.

16 . The computer program product of claim 13 , wherein the instructions further cause the processor to perform steps comprising:

applying the machine-learning model to detect an intent of the user based on the gathered data; and

applying the machine-learning model to detect the one or more anomalies based on the detected intent.

17 . The computer program product of claim 13 , wherein the instructions further cause the processor to perform steps comprising:

flagging the user for an audit process, based on determining that the one or more detected anomalies satisfy the threshold condition for the fraudulent behavior;

determining, during the audit process, that a first set of items the user paid for does not match a second set of items the user is attempting to remove from the location of the retailer; and

sending the communication to the management system to cause the at least one remedial action to be performed, based on determining that the first set of items does not match the second set of items.

18 . A computer system comprising:

a processor; and

a non-transitory computer-readable storage medium having instructions that, when executed by the processor, cause the computer system to perform steps comprising:

gathering, by the computer system, data associated with a physical activity of a user, the user being physically present at a location of a retailer and adding a set of items to a physical receptacle, the computer system in communication with the physical receptacle, the data including information about one or more weight changes associated with one or more items of the set of items added to the physical receptacle without a scanning performed for the one or more items by the user;

accessing a machine-learning model trained to detect a fraudulent behavior of the user associated with the physical activity;

applying the machine-learning model to the gathered data including the information about the one or more weight changes associated with the one or more items added to the physical receptacle without the scanning to:

detect one or more anomalies during the physical activity of the user, the one or more anomalies being indicative of the fraudulent behavior, and

generate a score for each of the one or more detected anomalies that is indicative of a likelihood that each of the one or more detected anomalies is associated with the fraudulent behavior;

generating, using the score for each of the one or more detected anomalies, a total score for the one or more detected anomalies;

determining, based on the total score being higher than a threshold score, that the one or more detected anomalies satisfy a threshold condition for the fraudulent behavior;

in response to determining that the one or more detected anomalies satisfy the threshold condition for the fraudulent behavior, sending a communication to a management system of the retailer to cause at least one remedial action to be performed before the user physically leaves the location of the retailer;

determining whether the one or more anomalies detected by the machine-learning model were correct;

generating training data based on information whether the one or more detected anomalies were correct; and

retraining the machine-learning model using the generated training data.

Assignments (1)

Continuity (1)

References Cited (56)