Granted Patent B1

US 12,561,383 · App. 18/192,958 · Granted Feb 24, 2026

Inventors: Yu Wang (Chicago, IL); Zhengyang Wang (Sammamish, WA); Qingyu Yin (Mountain View, CA); Xianfeng Tang (San Jose, CA); Xiao Cheng (Palo Alto, CA); Bing Yin (Palo Alto, CA)

Assignee: AMAZON TECHNOLOGIES, INC.

G06F16/9532G06F40/30

Abstract

Claims (70)

1 . A system comprising:

one or more non-transitory memories storing computer-executable instructions; and

one or more hardware processors to execute the computer-executable instructions to:

determine a sequence of search queries associated with a session, wherein the sequence of search queries includes at least: a first search query comprising first text, a second search query comprising second text, and a third search query comprising third text;

use the first search query and the second search query as inputs to a first machine learning model to determine a first query intent associated with the second search query, wherein the first machine learning model is trained to determine query intents associated with queries using sequences of queries as training data;

use the second search query and the third search query as inputs to the first machine learning model to determine a second query intent associated with the third search query;

determine a sequence of intents that includes at least the first query intent and the second query intent;

use the sequence of intents as inputs to the first machine learning model to determine an output that indicates a predicted intent for a subsequent query, wherein the first machine learning model is further trained to determine a subsequent query intent based on a sequence of previous query intents;

use the sequence of search queries and the predicted intent as inputs to a second machine learning model to determine a predicted search query, wherein the second machine learning model is trained to determine subsequent queries based on a sequence of previous queries and predicted intents for subsequent queries, and wherein the predicted search query corresponds to the predicted intent; and

generate output data indicative of the predicted search query.

2 . The system of claim 1 , further comprising computer-executable instructions to:

determine a user interaction associated with at least one item presented in a response to the first search query; and

use the user interaction as an input to the first machine learning model, wherein the first machine learning model further determines the predicted intent based on the user interaction.

3 . The system of claim 1 , further comprising computer-executable instructions to:

determine a total count of queries associated with at least a subset of the sequence of search queries;

determine a relationship between a count associated with a subsequent query and the total count; and

use the relationship between the count associated with the subsequent query and the total count as an input to the first machine learning model, wherein the first machine learning model further determines the predicted intent based on the relationship.

4 . A system comprising:

one or more non-transitory memories storing computer-executable instructions; and

one or more hardware processors to execute the computer-executable instructions to:

determine a sequence of search queries that includes at least a first search query, a second search query, and a third search Query;

use the first search query and the second search query as inputs to a first machine learning model to determine a first query intent associated with the second search query;

use the second search query and the third search query as inputs to the first machine learning model to determine a second query intent associated with the third search query;

use the first machine learning model and a sequence of intents that includes at least the first query intent and the second query intent to determine a predicted intent for a subsequent query;

use the sequence of search queries and the predicted intent as inputs to a second machine learning model to determine a predicted search query; and

determine output indicative of the predicted search query.

5 . The system of claim 4 , further comprising computer-executable instructions to:

use data indicative of the sequence of search queries as an input to the first machine learning model, wherein the first machine learning model further determines the predicted intent based on the sequence of search queries.

6 . The system of claim 4 , further comprising computer-executable instructions to:

determine a count of intents associated with the sequence of intents, wherein the sequence of intents and the count of intents are used by one or more of the first machine learning model or the second machine learning model to determine the predicted intent.

7 . The system of claim 4 , further comprising computer-executable instructions to:

determine a user interaction associated with a response to the first search query; and

use data indicative of the user interaction as an input to the first machine learning model, wherein the first machine learning model further determines the predicted intent based on the user interaction.

8 . The system of claim 4 , further comprising computer-executable instructions to:

determine session data indicative of input provided to one or more interfaces that are not associated with the sequence of search queries; and

use the session data as an input to the first machine learning model, wherein the first machine learning model further determines the predicted intent based on the session data.

9 . The system of claim 4 , further comprising computer-executable instructions to:

determine session data indicative of input provided to one or more computing devices external to the system; and

determine the first search query based on the input.

10 . The system of claim 4 , further comprising computer-executable instructions to:

determine, based on a first type of intent associated with one or more search queries of the sequence of search queries, one or more of a total count of queries associated with at least a subset of the sequence of search queries or a normalized count of queries associated with the at least a subset of the sequence of search queries; and

use the one or more of the total count or the normalized count as an input to the first machine learning model, wherein the first machine learning model further determines the predicted intent based on the one or more of the total count or the normalized count.

11 . The system of claim 4 , further comprising computer-executable instructions to:

determine a subset of the sequence of intents that is associated with a first type of intent;

determine one or more of a total count of intents or a normalized count of intents associated with the subset; and

use the one or more of the total count or the normalized count as an input to the first machine learning model, wherein the first machine learning model further determines the predicted intent based on the one or more of the total count or the normalized count, and wherein the predicted intent is associated with a second type of intent that differs from the first type.

12 . The system of claim 4 , further comprising computer executable instructions to:

use a third machine learning model to determine one or more first feature embeddings based on first text of the first search query and one or more second feature embeddings based on second text of the second search query;

wherein the one or more first feature embeddings and the one or more second feature embeddings are used as inputs to the second machine learning model to determine the predicted search query.

13 . A system comprising:

one or more non-transitory memories storing computer-executable instructions; and

one or more hardware processors to execute the computer-executable instructions to:

determine a sequence of inputs that includes at least a first input, a second input; and a third input;

use the first input, the second input, and a first machine learning model to determine a first intent associated with the second input;

use the second input, the third input, and the first machine learning model to determine a second intent associated with the third input;

use the first intent, the second intent, and one or more of the first machine learning model or a second machine learning model to determine a predicted intent for a subsequent input;

use the predicted intent, the sequence of inputs, and one or more of the first machine learning model, the second machine learning model, or a third machine learning model to determine a predicted input; and

determine output indicative of the predicted input.

14 . The system of claim 13 , wherein one or more of the first input, the second input, or the third input comprises a search query that is input using an interface that is associated with search queries for items available for transactions.

15 . The system of claim 13 , wherein one or more of the first input, the second input, or the third input comprises a user interaction associated with an item available for transactions.

16 . The system of claim 15 , wherein the user interaction includes one or more of: navigation to an interface that includes information associated with the item, addition of the item to an electronic list, or initiation of a transaction associated with the item.

17 . The system of claim 13 , further comprising computer-executable instructions to:

determine session data associated with the sequence of inputs, wherein the session data is indicative of input provided to one or more interfaces that are associated with the sequence of inputs; and

determine the first input based on the session data.

18 . The system of claim 13 , wherein the sequence of inputs is used by the one or more of the first machine learning model or the second machine learning model to determine the predicted intent.

19 . The system of claim 13 , wherein the sequence of inputs further includes a third input, the system further comprising computer-executable instructions to:

determine a count of intents associated with the sequence of intents, wherein the sequence of intents and the count of intents are used by the one or more of the first machine learning model or the second machine learning model to determine the predicted intent.

20 . The system of claim 13 , wherein the output includes one or more of:

a list of one or more search queries that include the predicted input presented in association with one or more search results that are presented in response to the third input; or

a list of one or more search queries that include the predicted input presented in association with an interface for receiving search queries.

Assignments (2)

References Cited (64)