# US 12,182,829 B2 · Generating concise and common user representations for edge systems from event sequence data stored on hub systems

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 12,182,829 B2
- Document ID: US12182829B2
- Status: Granted
- Kind: B2
- Application number: 17/849,320
- Filed: 2022-06-24
- Published: 2023-12-28
- Granted: 2024-12-31
- Earliest claimed: 2022-06-24
- Assignees: Adobe Inc.
- Applicants: Adobe Inc.
- Inventors: Sarthak Chakraborty; Sunav Choudhary; Atanu R. Sinha; Sapthotharan Krishnan Nair; Manoj Ghuhan Arivazhagan; Yuvraj; Atharva Anand Joshi; Atharv Tyagi; Shivi Gupta
- Web page: https://patentgrep.com/patent/US12182829B2

## CPC

- G06Q 30/0201: Market modelling; Market analysis; Collecting market data
- G06N 3/04: Architecture, e.g. interconnection topology
- G06N 3/0442: characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
- G06N 3/0455: Auto-encoder networks; Encoder-decoder networks
- G06N 3/08: Learning methods
- G06Q 30/0255: {based on user history}
- G06Q 30/0269: {based on user profile or attribute}
- G06Q: INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

A system includes a representation generator subsystem configured to execute a user representation model and a task prediction model to generate a user representation for a user. The user representation model receives user event sequence data comprises a sequence of user interactions with the system. The task prediction model is configured to train the user representation model. The user representation includes a vector of a predetermined size that represents the user event sequence data and is generated by applying the trained user representation model to the user event sequence data. A storage requirement of the user representation is less than a storage space requirement of the user event sequence data. The system includes a data store configured for storing the user representation in a user profile associated with the user.

## Claims

### Claim 1

A system comprising:
a representation generator subsystem configured to execute a user representation model and a task prediction model to generate a user representation for a user, wherein:
the user representation model comprises a machine learning model trained to generate a user representation based on user event sequence data, the user event sequence data comprising high-dimensional data representing sequences of user interactions by the user with the system,
the user representation comprises one of a task specific embedding or a multitask embedding, each of which comprises a low-dimensional representation of the user event sequence data expressed as a vector of a predetermined size,
the task prediction model is used to train the user representation model with a training data set comprising a subset of the user event sequence data, wherein the training comprises applying a loss function to an output of the user representation model to assess a prediction power of the user representation, and
if a loss determined by the loss function exceeds a predetermined threshold, one or more parameters of the user representation model are iteratively modified until the loss determined by the loss function does not exceed the threshold; and
a data store configured for storing the user representation in a user profile associated with the user, wherein a storage space requirement of the user representation in the data store is less than a storage space requirement of the user event sequence data.

### Claim 2 (depends on 1)

The system of claim 1, wherein the user representation model comprises an autoencoder network including a single-layer LSTM encoder and a single-layer LSTM decoder.

### Claim 3 (depends on 1)

The system of claim 1, wherein the user representation model comprises a two-layered long short-term memory (“LSTM”) network.

### Claim 4 (depends on 1)

The system of claim 1, wherein the vector has a predefined number of dimensions.

### Claim 5 (depends on 1)

The system of claim 1, wherein the user interactions with the system comprise one or more of conversions or user interface interactions.

### Claim 6

A non-transitory computer-readable medium storing executable instructions, which when executed by a processing device, cause the processing device to:
execute a user representation model and a task prediction model to generate a user representation for a user, wherein:
the user representation model comprises a machine learning model trained to generate a user representation based on user event sequence data, the user event sequence data comprising high-dimensional data representing sequences of user interactions by the user with a system;
the user representation comprises one of a task specific embedding or a multitask embedding, each of which comprises a low-dimensional representation of the user event sequence data expressed as a vector of a predetermined size,
the task prediction model is used to train the user representation model with a training data set comprising a subset of the user event sequence data, wherein the training comprises applying a loss function to an output of the user representation model to assess a prediction power of the user representation; and
if a loss determined by the loss function exceeds a predetermined threshold, one or more parameters of the user representation model are iteratively modified until the loss determined by the loss function does not exceed the threshold; and
store the user representation in a user profile associated with the user, wherein a storage space requirement of the user representation is less than a storage space requirement of the user event sequence data.

### Claim 7 (depends on 6)

The non-transitory computer readable medium of claim 6, wherein the user representation model comprises a two-layered long short-term memory (“LSTM”) network.

### Claim 8 (depends on 6)

The non-transitory computer readable medium of claim 6, wherein the user representation model comprises an autoencoder network including a single-layer LSTM encoder and a single-layer LSTM decoder.

### Claim 9

A method, comprising:
executing a user representation model and a task prediction model to generate a user representation for a user, wherein;
the user representation model comprises a machine learning model trained to generate a user representation based on user event sequence data, the user event sequence data comprising high-dimensional data representing sequences of user interactions by the user with a system;
the user representation comprises one of a task specific embedding or a multitask embedding, each of which comprises a low-dimensional representation of the user event sequence data expressed as a vector of a predetermined size
the task prediction model is used to train the user representation model with a training data set comprising a subset of the user event sequence data, wherein the training comprises applying a loss function to an output of the user representation model to assess a prediction power of the user representation; and
if a loss determined by the loss function exceeds a predetermined threshold, one or more parameters of the user representation model are iteratively modified until the loss determined by the loss function does not exceed the threshold; and
storing the user representation in a user profile associated with the user, wherein a storage space requirement of the user representation is less than a storage space requirement of the user event sequence data.

### Claim 10 (depends on 9)

The method of claim 9, wherein the user representation model comprises a two-layered long short-term memory (“LSTM”) network.

### Claim 11 (depends on 9)

The method of claim 9, wherein the user representation model comprises an autoencoder network including a single-layer LSTM encoder and a single-layer LSTM decoder.