Systems and methods for item-specific keyword recommendation

Patent metadata

Field	Value
Publication number	US-12561354-B2
Application number	US-202318104083-A
Country	US
Kind code	B2
Filing date	Jan 31, 2023
Priority date	Jan 31, 2023
Publication date	Feb 24, 2026
Grant date	Feb 24, 2026

Abstract

Systems and methods of item-specific keyword recommendation are disclosed. An item data structure including an item title is received and at least one item embedding is generated by applying a first trained semantic mapping model to the item title. The first trained semantic mapping model includes a first semantic mapping framework. The at least one item embedding is compared to a set of keyword embeddings representative of a set of platform-relevant keywords and a set of item-specific recommended keywords is selected from the set of platform-relevant keywords based on a similarity between the at least one item embedding and each embedding in the set of keyword embeddings. The item title is modified to include at least one of the set of item-specific recommended keywords and an interface including the modified item title is generated.

First claim

What is claimed is:

1 . A system, comprising:
a non-transitory memory storing instructions; and a processor communicatively coupled to the non-transitory memory, wherein the instructions, when executed, cause the processor to:
receive, from a database, an item data structure including an item title;
input the item data structure to a first embedding generating tower of a first trained semantic mapping model and generate at least one item embedding, wherein the first trained semantic mapping model includes a first semantic mapping framework;
input a set of platform-relevant keywords to a second embedding generating tower of the first trained semantic mapping model and generate a set of keyword embeddings;
compare the at least one item embedding to the set of keyword embeddings representative of the set of platform-relevant keywords and determine a similarity between the at least one item embedding and each embedding in the set of keyword embeddings;
select a set of item-specific recommended keywords from the set of platform-relevant keywords based on the similarities;
determine a match between at least one item-specific recommended keyword in the set of item-specific recommended keywords and the item title;
modify the item title to include the at least one item-specific recommended keyword in the set of item-specific recommended keywords based on the determined match;
update the item data structure to include the modified item title;
receive a search query; and
generate an interface responsive to the search query, wherein the interface includes the item data structure and the modified item title.

2 . The system of claim 1 , wherein the set of keyword embeddings are generated by a second trained semantic mapping model, and wherein the second trained semantic mapping model includes the first semantic mapping framework.

3 . The system of claim 1 , wherein the at least one item embedding is compared to the set of keyword embeddings by determining a pairwise distance between the at least one item embedding and each embedding in the set of keyword embeddings in vector space.

4 . The system of claim 3 , wherein the set of item-specific recommended keywords includes keywords associated with keyword embeddings located within a predetermined pairwise distance of the at least one item embedding in vector space.

5 . The system of claim 1 , wherein the set of keyword embeddings are generated prior to receiving the item data structure.

6 . The system of claim 1 , wherein the processor is configured to read the set of instructions to filter the set of recommended augmentation keywords to remove a set of removal words.

7 . The system of claim 1 , wherein the processor is configured to read the set of instructions to deduplicate the set of item-specific recommended keywords based on the item title.

8 . The system of claim 1 , wherein the processor is configured to read the set of instructions to deduplicate the set of item-specific recommended keywords based on a prior item-specific recommended keyword.

9 . The system of claim 8 , wherein the deduplication includes:
generating a first embedding for the prior item-specific recommended keyword using a second semantic mapping model, wherein the second semantic mapping model includes a second semantic mapping framework; generating a second embedding for a next item-specific recommended keyword in the set of item-specific recommended keywords using the second semantic mapping model; determining a similarity between the first embedding and the second embedding; and deduplicating the next item-specific recommended keyword when the similarity is above a predetermined threshold.

10 . The system of claim 1 , wherein modifying the item title includes augmenting the item title with the at least one item-specific recommended keyword in the set of item-specific recommended keywords.

11 . A computer-implemented method, comprising:
receiving, from a database, an item data structure including an item title; inputting the item data structure to a first embedding generating tower of a first trained semantic mapping model and generating at least one item embedding, wherein the first trained semantic mapping model includes a first semantic mapping framework; inputting a set of platform-relevant keywords to a second embedding generating tower of the first trained semantic mapping model and generate a set of keyword embeddings; comparing the at least one item embedding to the set of keyword embeddings representative of the set of platform-relevant keywords and determining a similarity between the at least one item embedding and each embedding in the set of keyword embeddings; selecting a set of item-specific recommended keywords from the set of platform-relevant keywords based on the similarities; determining a match between at least one item-specific recommended keyword in the set of item-specific recommended keywords and the item title; modifying the item title to include at least one of the set of item-specific recommended keywords based on the determined match; updating the item data structure to include the modified item title; receiving a search query; and generating an interface responsive to the search query, wherein the interface includes the item data structure and the modified item title.

12 . The computer-implemented method of claim 11 , wherein the set of keyword embeddings are generated by a second trained semantic mapping model, and wherein the second trained semantic mapping model includes the first semantic mapping framework.

13 . The computer-implemented method of claim 11 , wherein the at least one item embedding is compared to the set of keyword embeddings by determining a pairwise distance between the at least one item embedding and each embedding in the set of keyword embeddings in vector space.

14 . The Previously Presented of claim 13 , wherein the set of item-specific recommended keywords includes keywords associated with keyword embeddings located within a predetermined pairwise distance of the at least one item embedding in vector space.

15 . The computer-implemented method of claim 11 , wherein the set of keyword embeddings are generated prior to receiving the item data structure.

16 . The computer-implemented method of claim 11 , comprising filtering the set of item-specific recommended keywords to remove a set of removal words.

17 . The computer-implemented method of claim 11 , comprising deduplicating the set of item-specific recommended keywords based on the item title.

18 . The computer-implemented method of claim 11 , comprising deduplicating the set of item-specific recommended keywords based on a prior item-specific recommended keyword.

19 . The computer-implemented method of claim 18 , wherein the deduplication includes:
generating a first embedding for the prior item-specific recommended keyword using a second semantic mapping model, wherein the second semantic mapping model includes a second semantic mapping framework; generating a second embedding for a next item-specific recommended keyword in the set of recommended augmentation keywords using the second semantic mapping model; determining a similarity between the first embedding and the second embedding; and deduplicating the next item-specific recommended keyword when the similarity is above a predetermined threshold.

20 . A non-transitory computer-readable medium having instructions stored thereon, wherein the instructions, when executed

Assignees

• Walmart Apollo Llc

Inventors

• Wang Zigeng
• Kim Jae Young
• Cai Yaotong
• Zhao Xia
• Shen Wei

Classifications

• G06F40/30 Primary

  Semantic analysis · CPC title

• G06F16/3334 Primary

  Selection or weighting of terms from queries, including natural language queries · CPC title