# US 12,217,296 B2 · Systems and methods using deep joint variational autoencoders

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

- Number: US 12,217,296 B2
- Document ID: US12217296B2
- Status: Granted
- Kind: B2
- Application number: 17/589,229
- Filed: 2022-01-31
- Published: 2023-08-03
- Granted: 2025-02-04
- Earliest claimed: 2022-01-31
- Assignees: Walmart Apollo, LLC
- Applicants: Walmart Apollo, LLC
- Inventors: Venugopal Mani; Jianpeng Xu; Hyun Duk Cho; Sushant Kumar; Kannan Achan; Aysenur Inan
- Web page: https://patentgrep.com/patent/US12217296B2

## CPC

- G06Q 30/0631: {Recommending goods or services}
- G06N 3/044: Recurrent networks, e.g. Hopfield networks
- G06N 3/045: Combinations of networks
- G06N 3/047: Probabilistic or stochastic networks
- G06N 3/08: Learning methods
- G06N 3/084: Backpropagation, e.g. using gradient descent
- G06N 3/088: Non-supervised learning, e.g. competitive learning
- G06N 7/01: Probabilistic graphical models, e.g. probabilistic networks
- 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

Systems and methods for generating top-k recommendation using latent space representations generated by deep joint variational autoencoder processes are disclosed. A user identifier is received and a set of prior interactions associated with the user identifier is obtained. A set of latent space representations of the set of prior interactions is generated using a trained inference model. The trained inference model includes a joint variational autoencoder model. A set of k-recommended items is generated based on a comparison of the set of latent space representations of the set of prior interactions and a set of latent space representations of one or more items. A user interface including the set of k-recommended items is generated.

## Claims

### Claim 1

A system, comprising:
a processor; and
a non-transitory memory storing instructions, that when executed, cause the processor to:
receive a user identifier corresponding to a set of prior interactions;
generate a set of candidate items;
generate a set of latent space representations of the set of prior interactions and the candidate items using a trained inference model, wherein the trained inference model generates distribution likelihoods for hidden variables and includes a joint variational autoencoder model;
generate a set of k-recommended items based on a comparison of the set of latent space representations of the set of prior interactions and the set of latent space representations of the candidate items;
rank the set of k-recommended items based on the candidate items and a set of all available items; and
generate a user interface including the ranked set of k-recommended items.

### Claim 2 (depends on 1)

The system of claim 1, wherein the joint variational autoencoder model comprises a user variational autoencoder model and an item variational autoencoder model.

### Claim 3 (depends on 2)

The system of claim 2, wherein each latent space representation in the set of latent space representations of the set of prior interactions is generated by combining a first user-item value generated by the user variational autoencoder model and a second user-item value generated by the item variational autoencoder model.

### Claim 4 (depends on 3)

The system of claim 3, wherein each latent space representation in the set of latent space representations is an average value of the first user-item value and the second user-item.

### Claim 5 (depends on 1)

The system of claim 1, wherein the trained inference model is configured to generate a plurality of probability distributions for a set of features identified during training of the joint variational autoencoder model.

### Claim 6 (depends on 1)

The system of claim 1, wherein the joint variational autoencoder is configured to estimate a probability distribution according to
min KL(q(z|x)∥p(z|x))
where KL is Kullback-Leibler divergence, p(z|x) is an intractable distribution, and q(z|x) is a tractable distribution.

### Claim 7 (depends on 1)

The system of claim 1, wherein the set of latent space representations of one or more items is generated by a second joint variational autoencoder model prior to receiving the user identifier.

### Claim 8

A non-transitory computer readable medium having instructions stored thereon, wherein the instructions, when executed by a processor cause a device to perform operations comprising:
receiving a user identifier corresponding to a set of prior interactions;
generating a set of candidate items;
generating a set of latent space representations of the set of prior interactions and the candidate items using a trained inference model, wherein the trained inference model generates distribution likelihoods for hidden variables and includes a joint variational autoencoder model;
generating a set of k-recommended items based on a comparison of the set of latent space representations of the set of prior interactions and the set of latent space representations of the candidate items;
ranking the set of k-recommended items based on the candidate items and a set of all available items; and
generating a user interface including the ranked set of k-recommended items.

### Claim 9 (depends on 8)

The non-transitory computer readable medium of claim 8, wherein the joint variational autoencoder model comprises a user variational autoencoder model and an item variational autoencoder model.

### Claim 10 (depends on 9)

The non-transitory computer readable medium of claim 9, wherein the one or more latent space representations of the set of prior interactions is generated by combining a first user-item value generated by the user variational autoencoder model and a second user-item value generated by the item variational autoencoder model.

### Claim 11 (depends on 10)

The non-transitory computer readable medium of claim 10, wherein the one or more latent space representations of the set of prior interactions are an average value of the first user-item value and the second user-item value.

### Claim 12 (depends on 8)

The non-transitory computer readable medium of claim 8, wherein the trained inference model is configured to generate a plurality of probability distributions for a set of features identified during training of the joint variational autoencoder model.

### Claim 13 (depends on 8)

The non-transitory computer readable medium of claim 8, wherein the joint variational autoencoder is configured to estimate a probability distribution according to
min KL(q(z|x)∥p(z|x))
where KL is Kullback-Leibler divergence, p(z|x) is an intractable distribution, and q(z|x) is a tractable distribution.

### Claim 14 (depends on 8)

The non-transitory computer readable medium of claim 8, wherein the set of latent space representations of one or more items is generated by a second joint variational autoencoder model prior to receiving the user identifier.

### Claim 15

A method, comprising:
receiving a user identifier corresponding to a set of prior interactions;
generating a set of candidate items;
generating a set of latent space representations of the set of prior interactions and the candidate items using a trained inference model, wherein the trained inference model generates distribution likelihoods for hidden variables and includes a joint variational autoencoder model;
generating a set of k-recommended items based on a comparison of the set of latent space representations of the set of prior interactions and the set of latent space representations of the candidate items;
ranking the set of k-recommended items based on the candidate items and a set of all available items; and
generating a user interface including the ranked set of k-recommended items.

### Claim 16 (depends on 15)

The method of claim 15, wherein the joint variational autoencoder model comprises a user variational autoencoder model and an item variational autoencoder model.

### Claim 17 (depends on 16)

The method of claim 16, wherein each latent space representation is generated by combining a first user-item value generated by the user variational autoencoder model and a second user-item value generated by the item variational autoencoder model.

### Claim 18 (depends on 17)

The method of claim 17, wherein each latent space representation is an average value of the first user-item value and the second user-item.

### Claim 19 (depends on 15)

The method of claim 15, wherein the trained inference model is configured to generate a plurality of probability distributions for a set of features identified during training of the joint variational autoencoder model.

### Claim 20 (depends on 15)

The method of claim 15, wherein the joint variational autoencoder is configured to estimate a probability distribution according to
min KL(q(z|x)∥p(z|x))
where KL is Kullback-Leibler divergence, p(z|x) is an intractable distribution, and q(z|x) is a tractable distribution.