Systems and methods for interface generation using explore and exploit strategies

Patent metadata

Field	Value
Publication number	US-12579155-B2
Application number	US-202318489405-A
Country	US
Kind code	B2
Filing date	Oct 18, 2023
Priority date	Oct 18, 2023
Publication date	Mar 17, 2026
Grant date	Mar 17, 2026

Abstract

Systems and methods of generating an interface are disclosed. A request for an interface including a search query is received and a set of search results are generated via a search engine. A set of initial content campaigns are selected based on a category associated with the search query and are filtered based on a similarity score generated for the search query and each content campaign to generate a set of candidate content campaigns. The set of candidate content campaigns is filtered based on an exploration content score and an exploitation content score to generate a set of final content campaigns. An interface is generated including at least one content element selected from the set of final content campaigns and at least one content element selected from the set of search results. The interface is transmitted to the requesting system, which is configured to display the interface on a display.

First claim

What is claimed is:

1 . A system, comprising:
a database configured to store a plurality of content campaigns, wherein each content campaign in the plurality of content campaigns includes at least one content element; a processor; and a non-transitory memory storing instructions that, when executed, cause the processor to: receive, from a requesting system, a request for an interface including a search query;
generate, via a search engine, a set of search results responsive to the search query;
select a set of initial content campaigns from the database based on a category associated with the search query, each of the initial content campaigns having an associated category;
generate a query embedding for the search query and a content element embedding for each content campaign in the set of initial content campaigns;
generate a first similarity sub-score based on the query embedding and each content element embedding, wherein each first similarity sub-score characterizes a similarity between the search query and the corresponding content campaign;
generate a query taxonomy embedding for each category associated with the query;
generate a content campaign taxonomy embedding for each category associated with each of the initial content campaigns;
generate a second similarity sub-score based on the query taxonomy embedding and each content campaign taxonomy embedding, wherein each second similarity sub-score characterizes a similarity between the search query and the corresponding content campaign;
generate a plurality of similarity scores based on the generated first similarity sub-scores and the generated second similarity sub-scores;
filter the set of initial content campaigns based on the generated similarity scores to generate a set of candidate content campaigns;
filter the set of candidate content campaigns based on an exploration content score and an exploitation content score to generate a set of final content campaigns;
generate an interface including at least one content element selected from the content elements associated with a content campaign in the set of final content campaigns and at least one content element selected from the set of search results; and
transmit the interface to the requesting system, wherein the requesting system is configured to display the interface on a display associated with the requesting system.

2 . The system of claim 1 , wherein the set of candidate content campaigns is filtered based on a combined content score comprising a first weighted combination of the exploration content score and the exploitation content score.

3 . The system of claim 2 , wherein the set of candidate content campaigns is filtered based on a second weighted combination of the first weighted combination and one or more of the similarity scores.

4 . The system of claim 2 , wherein the first weighted combination comprises a predicted click-through-rate for each content campaign in the set of candidate content campaigns.

5 . The system of claim 1 , wherein the exploration content score is generated by a trained model including an XGBoost framework.

6 . The system of claim 1 , wherein the exploitation content score is generated by a trained model including a Thompson Sampling framework.

7 . The system of claim 1 , wherein one or more of the similarity scores includes a cosine similarity of at least two embeddings.

8 . The system of claim 7 , wherein one or more of the similarity scores comprises a weighted combination of a first cosine similarity of a first embedding and a second embedding and a second cosine similarity of a third embedding and a fourth embedding.

9 . A computer-implemented method, comprising:
receiving, from a requesting system, a request for an interface including a search query; generating, via a search engine, a set of search results responsive to the search query; determining a category associated with the search query; selecting a set of initial content campaigns from a database based on the category associated with the search query and a category associated with each content campaign in the database, each of the initial content campaigns having an associated category; generating a query embedding for the search query and a content element embedding for each content campaign in the set of initial content campaigns; generating a first similarity sub-score based on the query embedding and each content element embedding, wherein each similarity score characterizes a similarity between the search query and the corresponding content campaign; generating a query taxonomy embedding for each category associated with the query; generating a content campaign taxonomy embedding for each category associated with each of the initial content campaigns; generating a second similarity sub-score based on the query taxonomy embedding and each content campaign taxonomy embedding, wherein each second similarity sub-score characterizes a similarity between the search query and the corresponding content campaign; generating a plurality of similarity scores based on the generated first similarity sub-scores and the generated second similarity sub-scores; selecting a set of candidate content campaigns from the set of initial content campaigns based on the similarity scores generated for the search query and each content campaign in the set of initial content campaigns; selecting a set of final content campaigns from the set of candidate content campaigns based on a combined content score comprising a weighted combination of an exploration content score and an exploitation content score; generating an interface including at least one content element selected from the content elements associated with a content campaign in the set of final content campaigns and at least one content element selected from the set of search results; and transmitting the interface to the requesting system, wherein the requesting system is configured to display the interface on a display associated with the requesting system.

10 . The computer-implemented method of claim 9 , wherein the set of final content campaigns is selected based on a weighted combination of the combined content score and one or more of the similarity scores.

11 . The computer-implemented method of claim 9 , wherein the exploration content score is generated by a trained model including an XGBoost framework.

12 . The computer-implemented method of claim 9 , wherein the exploitation content score is generated by a trained model including a Thompson Sampling framework.

13 . The computer-implemented method of claim 9 , wherein one or more of the similarity scores includes a cosine similarity of at least two embeddings.

14 . The computer-implemented method of claim 13 , wherein one or more of the similarity scores comprises a weighted combination of a first cosine similarity of a first embedding and a second embedding and a second cosine similarity of a third embedding and a fourth embedding.

15 . A non-transitory computer-readable medium having instructions stored thereon that, when executed by a processor, cause a device to:
receive, from a requesting system, a request for an interface including a search query; generate, via a search engine, a set of search results responsive to the search query; determine a category associated with the search query; select a set of initial content campaigns from a database based on the category associated with the search query and a category associated with each content campaign in the database, each of the initial content campaigns having an associated

Assignees

• Walmart Apollo Llc

Inventors

• Xu Dong
• Ai Di
• Sun Xinzi

Classifications

• G06F16/24575

  using context · CPC title

• G06F16/9538

  Presentation of query results · CPC title

• G06F16/248 Primary

  Presentation of query results · CPC title