Personalized ranking using deep attribute extraction and attentive user interest embeddings

Patent metadata

Field	Value
Publication number	US-11797624-B2
Application number	US-202217696461-A
Country	US
Kind code	B2
Filing date	Mar 16, 2022
Priority date	Jul 31, 2019
Publication date	Oct 24, 2023
Grant date	Oct 24, 2023

Abstract

In some examples, a system may be configured to generate one or more query attributes for a search query received from a computing device of a user. Additionally, the system may be configured to, based at least in part on historical data of the user including data characterizing one or more items associated with the user, generate relevant item data. In various examples, the relevant item data characterizing a set of relevant items. Moreover, the system may be configured to, based on the relevant item data, the historical data of the user and the one or more query attributes, implement a set of operations that generate a set of personalized search results associated with the search query.

First claim

What is claimed is:

1. A system, comprising:
a memory storing instructions; and
one or more processors coupled to the memory, the one or more processors being configured to execute the instructions to:
generate one or more query attributes for a search query received from a computing device of a user by implementing an attribute extraction module configured to:
generate a set of combined word representations;
implement a trained bidirectional recurrent neural network configured to receive the set of combined word representations and generate an output; and
implement a trained conditional random field layer configured to receive the output of the trained bidirectional recurrent neural network and generate attribute values for the one or more query attributes;

based at least in part on historical data of the user including data characterizing one or more items associated with the user, generate relevant item data, the relevant item data characterizing a set of relevant items;
based on the relevant item data, the historical data of the user and the one or more query attributes implement a set of operations that generate a set of personalized search results associated with the search query, the set of operations including:
generating one or more user interest embeddings based at least on the relevant item data and the historical data of the user; and
based on the one or more user interest embeddings and the one or more query attributes, generate data identifying each item of the set of relevant items and corresponding rank.

2. The system of claim 1 , wherein generating the one or more user interest embeddings includes combining each of the one or more user interest embeddings with one or more associated attributes.

3. The system of claim 2 , wherein the one or more user interest embeddings are each multimodal.

4. The system of claim 2 , wherein the one or more processors being configured to execute the instructions further to _generate data characterizing a likelihood of a purchase event associated with one or more items identified in the relevant item data at a future time interval by utilizing at least the one or more user interest embeddings.

5. The system of claim 2 , wherein the one or more processors being configured to execute the instructions further to _generate data characterizing a likelihood of a selection event associated with one or more items identified in the relevant item data at a future time interval by utilizing at least the one or more user interest embeddings.

6. The system of claim 1 , wherein generating the one or more user interest embeddings includes _generating, for each of the one or more user interest embeddings, data indicating a probability score.

7. The system of claim 1 , wherein the one or more processors being configured to execute the instructions further to _generate data identifying each item of the set of relevant items and corresponding rank by utilizing a ranking network.

8. The system of claim 7 , wherein the ranking network comprises a learning to rank (LeTOR) framework.

9. The system of claim 1 , wherein the historical data includes data identifying one or more items that the user had previously interacted with on an online platform.

10. The system of claim 1 , wherein the historical data includes data identifying one or more non-purchasing user interactions of the user with one or more items.

11. A computer-implemented method comprising:
generating one or more query attributes for a search query received from a computing device of a user by implementing an attribute extraction module, wherein the attribute extraction module is configured to:
generate a set of combined word representations;
implement a trained bidirectional recurrent neural network configured to receive the set of combined word representations and generate an output; and
implement a trained conditional random field layer configured to receive the output of the trained bidirectional recurrent neural network and generate attribute values for the one or more query attributes;

based at least in part on historical data of the user including data characterizing one or more items associated with the user, generating relevant item data, the relevant item data characterizing a set of relevant items;
based on the relevant item data, the historical data of the user and the one or more query attributes, implementing a set of operations that generate a set of personalized search results associated with the search query, the set of operations including:
generating one or more user interest embeddings based at least on the relevant item data and the historical data of the user; and
based on the one or more user interest embeddings and the one or more query attributes, generate data identifying each item of the set of relevant items and corresponding rank.

12. The computer-implemented method of claim 11 , wherein generating the one or more user interest embeddings includes combining each of the one or more user interest embeddings with one or more associated attributes.

13. The computer-implemented method of claim 12 , wherein the one or more user interest embeddings are each multimodal.

14. The computer-implemented method of claim 12 , further comprising _generating data characterizing a likelihood of a purchase event associated with one or more items identified in the relevant item data at a future time interval by utilizing at least the one or more user interest embeddings.

15. The computer-implemented method of claim 12 , further comprising _generating data characterizing a likelihood of a selection event associated with one or more items identified in the relevant item data at a future time interval by utilizing at least the one or more user interest embeddings.

16. The computer-implemented method of claim 11 , wherein generating the one or more user interest embeddings includes _generating, for each of the one or more user interest embeddings, data indicating a probability score.

17. The computer-implemented method of claim 11 , further comprising _generating data identifying each item of the set of relevant items and corresponding rank by utilizing a ranking network.

18. The computer-implemented method of claim 17 , wherein the ranking network comprises a learning to rank (LeTOR) framework.

19. The computer-implemented method of claim 11 , wherein the historical data includes data identifying one or more items that the user had previously interacted with on an online platform.

20. A non-transitory computer-readable medium storing instructions that when executed by one or more processors, cause a computing system to:
generate one or more query attributes for a search query received from a computing device of a user by implementing an attribute extraction module, wherein the attribute extraction module is configured to:
generate a set of combined word representations;
implement a trained bidirectional recurrent neural network configured to receive the set of combined word representations and generate an output; and
implement a trained conditional random field layer configured to receive the output of the trained bidirectional recurrent neural network and generate attribute values for the one or more query attributes;

based at least in part on historical data of the user including data characterizing one or more items associated with the user, generate relevant item data, the relevant item data characterizing a set of relevant items;
based on the relevant item data, the historical data of the user

Assignees

• Walmart Apollo Llc

Inventors

• Iyer Rahul
• Wadhwa Soumya
• Kumar Surya Prasanna
• Kanumala Praveenkumar
• Guo Stephen Dean
• Achan Kannan
• Ramkumar Rahul

Classifications

• G06N3/0442

  characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU] · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06F16/9535 Primary

  Search customisation based on user profiles and personalisation · CPC title