Personalized item recommendations through large-scale deep-embedding architecture with real-time inferencing

Patent metadata

Field	Value
Publication number	US-11836782-B2
Application number	US-202117466277-A
Country	US
Kind code	B2
Filing date	Sep 3, 2021
Priority date	Jan 30, 2020
Publication date	Dec 5, 2023
Grant date	Dec 5, 2023

Abstract

A method being implemented via execution of computing instructions configured to run at one or more processors and stored at one or more non-transitory computer-readable media. The method can include training two sets of item embeddings for items in an item catalog and a set of user embeddings for users, using a triple embeddings model, with triplets. The triplets each include a respective first user of the users, a respective first item from the item catalog, and a respective second item from the item catalog, in which the respective first user selected the respective first item and the respective second item in a respective same basket. The method also can include randomly sampling an anchor item from a category of items selected by a user. The method additionally can include generating a list of complementary items using a query vector associated with the user and the anchor item. The query vector is generated for the user and the anchor item using the two sets of item embeddings and the set of user embeddings. Other embodiments are disclosed.

First claim

What is claimed is:

1. A system comprising:
one or more processors; and
one or more non-transitory computer-readable media storing computing instructions configured to run on the one or more processors and perform:
training two sets of item embeddings for items in an item catalog and a set of user embeddings for users, using a triple embeddings model, with triplets, wherein the triplets each comprise a respective first user of the users, a respective first item from the item catalog, and a respective second item from the item catalog, in which the respective first user selected the respective first item and the respective second item in a respective same basket;
randomly sampling an anchor item from a category of items selected by a user; and
generating a list of complementary items using a query vector associated with the user and the anchor item, wherein the query vector is generated for the user and the anchor item using the two sets of item embeddings and the set of user embeddings.

2. The system of claim 1 , wherein the query vector is generated for the user and the anchor item using the two sets of item embeddings and the set of user embeddings.

3. The system of claim 1 , wherein generating the list of complementary items further comprises using an approximate nearest neighbor index that is periodically precomputed using a similarity search library.

4. The system of claim 1 , wherein generating the list of complementary items further comprises making a lookup call to an approximate nearest neighbor index.

5. The system of claim 4 , wherein the set of user embeddings are loaded into a memory cache before the lookup call is made.

6. The system of claim 1 , wherein the computing instructions are further configured to perform:
building a list of personalized recommended items for the user based at least in part on the list of complementary items.

7. The system of claim 6 , wherein building the list of personalized recommended items further comprises:
filtering the list of the complementary items based on complementary subcategories; and
performing a weighted sampling of the list of the complementary items to generate a sampled sub-list of the list of the complementary items.

8. The system of claim 7 , wherein performing the weighted sampling of the list of the complementary items comprises:
sampling a quantity of items from the list of the complementary items proportional to a quantity of basket items in the category with respect to a total quantity of the basket items.

9. The system of claim 6 , wherein:
at least a portion of the list of personalized recommended items is displayed to the user on a checkout page; and
the checkout page appears on a user interface of a user device of the user.

10. The system of claim 1 , wherein the triple embeddings model is trained with the triplets using an adaptive moment estimation optimizer to optimize a co-occurrence log-likelihood of each of the triplets.

11. A method being implemented via execution of computing instructions configured to run at one or more processors and stored at one or more non-transitory computer-readable media, the method comprising:
training two sets of item embeddings for items in an item catalog and a set of user embeddings for users, using a triple embeddings model, with triplets, wherein the triplets each comprise a respective first user of the users, a respective first item from the item catalog, and a respective second item from the item catalog, in which the respective first user selected the respective first item and the respective second item in a respective same basket;
randomly sampling an anchor item from a category of items selected by a user; and
generating a list of complementary items using a query vector associated with the user and the anchor item, wherein the query vector is generated for the user and the anchor item using the two sets of item embeddings and the set of user embeddings.

12. The method of claim 11 , wherein the query vector is generated for the user and the anchor item using the two sets of item embeddings and the set of user embeddings.

13. The method of claim 11 , wherein generating the list of complementary items further comprises using an approximate nearest neighbor index that is periodically precomputed using a similarity search library.

14. The method of claim 11 , wherein generating the list of complementary items further comprises making a lookup call to an approximate nearest neighbor index.

15. The method of claim 14 , wherein the set of user embeddings are loaded into a memory cache before the lookup call is made.

16. The method of claim 11 further comprising:
building a list of personalized recommended items for the user based at least in part on the list of complementary items.

17. The method of claim 16 , wherein building the list of personalized recommended items further comprises:
filtering the list of the complementary items based on complementary subcategories; and
performing a weighted sampling of the list of the complementary items to generate a sampled sub-list of the list of the complementary items.

18. The method of claim 17 , wherein performing the weighted sampling of the list of the complementary items comprises:
sampling a quantity of items from the list of the complementary items proportional to a quantity of basket items in the category with respect to a total quantity of the basket items.

19. The method of claim 16 , wherein:
at least a portion of the list of personalized recommended items is displayed to the user on a checkout page; and
the checkout page appears on a user interface of a user device of the user.

20. The method of claim 11 , wherein the triple embeddings model is trained with the triplets using an adaptive moment estimation optimizer to optimize a co-occurrence log-likelihood of each of the triplets.

Assignees

• Walmart Apollo Llc

Inventors

• Mantha Aditya
• Arora Yokila
• Gupta Shubham
• Kanumala Praveenkumar
• Guo Stephen Dean
• Achan Kannan

Classifications

• G06Q30/0631 Primary

  Recommending goods or services · CPC title

• G06F16/9035

  Filtering based on additional data, e.g. user or group profiles · CPC title

• G06Q30/0635

  replenishment orders; recurring orders · CPC title

• G06Q30/0641

  utilising user interfaces specially adapted for shopping · CPC title