Back to results
US 10,650,306 B1Granted
User representation using a generative adversarial network
AssigneeAmazon Technologies, Inc.
InventorAnjishnu Kumar
Life of this patent
Estimating the term…
Abstract
An interactive system makes use of a concise user representation, for example, in the process of making predictions of a user's next action. In some examples, the concise user representation is computed from a larger amount of user data, which is processed using a transformation trained using a Generative Adversarial Network (GAN) approach.
Claims18
A method for computer-implemented user interaction comprising:
- processing a first user input received via a user interface of the system, including updating user data, that includes a history of interactions corresponding to a first user, according to the first user input;
- maintaining a user characteristic corresponding to the first user, the user characteristic comprising values of a plurality of quantitative or categorical variables used to represent the user, the quantitative or categorical variable forming a substantially independent set of variables determined using a Generative Adversarial Network (GAN), the maintaining of the user characteristic including
- sending the user data to a remote service, and
- receiving the user characteristic in response to the sending of the user data, wherein the user characteristic is represented using substantially fewer values than the user data, and wherein the remote service is configured according to values of a plurality inference parameters, the inference parameters being computed from user data corresponding to a plurality of users;
- predicting a further action by processing at least the user characteristic, and the history of interactions with the user;
- determining a first user output using the further action;
- and
- providing the first user output via the user interface.
5 more elements
A computer-implemented system comprising:
- a user interface for receiving a user input from a user;
- a storage for user data corresponding to the user;
- a communication interface for sending the user data to a remote network-based service via a data network, and for receiving, via the data network, a user characteristic in response to sending the user data, wherein the user characteristic is represented using fewer values than the user data;
- an action predictor configured to
- process a first user input received via the user interface of the system, including updating the user data, that includes a history of interactions corresponding to a first user, according to the first user input,
- maintain a user characteristic corresponding to the first user, the user characteristic comprising values of a plurality of quantitative or categorical variables used to represent the user, the quantitative or categorical variable forming a substantially independent set of variables determined using a Generative Adversarial Network (GAN), the maintaining of the user characteristic including
- sending the user data to the remote network-based service, and
- receiving the user characteristic in response to the sending of the user data, wherein the user characteristic is represented using substantially fewer values than the user data, and wherein the remote service is configured according to values of a plurality inference parameters, the inference parameters being computed from user data corresponding to a plurality of users, and
- predict a further action base at least on the user characteristic and the history of interactions with the user;
- and
- an interaction manager configured to
- receive the further action,
- determine a first user output using the further action, and
- provide the first user output to the user via the user interface according to the further action.
11 more elements
Dependencies are inferred from the claim text.