Result generation method, generation model training method, electronic device, and storage medium

Patent metadata

Field	Value
Publication number	US-2025384070-A1
Application number	US-202418933349-A
Country	US
Kind code	A1
Filing date	Oct 31, 2024
Priority date	Jun 18, 2024
Publication date	Dec 18, 2025
Grant date	—

Abstract

Provided is a result generation method, a generation model training method, an electronic device and a storage medium, relating to the field of computer technologies, and in particular, to the field of search and generative model technologies. The result generation method includes: acquiring a change query corresponding to an input query; obtaining a reference result by searching according to the input query and the change query corresponding to the input query; and generating an output result corresponding to the input query according to the input query, the change query corresponding to the input query and the reference result.

First claim

We claim:

1 . A result generation method, comprising:
acquiring a change query corresponding to an input query; obtaining a reference result by searching according to the input query and the change query corresponding to the input query; and generating an output result corresponding to the input query according to the input query, the change query corresponding to the input query and the reference result.

2 . The method of claim 1 , wherein the acquiring of the change query corresponding to the input query, comprises:
searching a change query dictionary for the change query corresponding to the input query.

3 . The method of claim 1 , wherein the obtaining of the reference result by searching according to the input query and the change query corresponding to the input query comprises:
obtaining a multi-intent query according to the input query and the change query corresponding to the input query; and inputting the multi-intent query into a search engine to obtain the reference result.

4 . The method of claim 3 , wherein the obtaining of the multi-intent query according to the input query and the change query corresponding to the input query comprises:
obtaining the multi-intent query by using a large language model to induce the input query and the change query corresponding to the input query.

5 . The method of claim 3 , wherein the generating of the output result corresponding to the input query according to the input query, the change query corresponding to the input query and the reference result comprises:
inputting the multi-intent query and the reference result into a generation model to obtain the output result corresponding to the input query.

6 . The method of claim 1 , wherein a training sample of a generation model comprises a prompt and an answer, and the prompt comprises an original query, a change query, a search result, and a target instruction.

7 . The method of claim 1 , wherein an original query and the change query are obtained by sampling a change query dictionary.

8 . The method of claim 1 , further comprising:
cleaning a plurality of change queries associated with an original query in a session according to a search intention of the original query.

9 . The method of claim 8 , wherein cleaning the plurality of change queries associated with the original query in the session according to the search intention of the original query comprises at least one of:
obtaining, based on keyword matching, a first similarity between the original query and the change query in the session, determining whether the search intention of the original query is similar to a search intention of the change query according to the first similarity, and retaining the change query has the search intention similar to the original query; or obtaining, based on semantic understanding for intention discrimination, a second similarity between the original query and the change query in the session, determining whether the search intention of the original query is similar to the search intention of the change query according to the second similarity, and retaining the change query has the search intention similar to the original query.

10 . The method of claim 8 , further comprising:
ranking the plurality of change queries associated with the original query based on a key feature, wherein the key feature includes at least one of: a change query rate, a change query source or feedback information after change query; and selecting a retained change query from the plurality of change queries associated with the original query according to a ranking result.

11 . The method of claim 10 , further comprising:
aggregating, based on a dynamic time window, the original query and the retained change query according to the search intention.

12 . A generation model training method, comprising:
inputting a prompt of a training sample into a generation model to be adjusted to obtain a predicted answer; and adjusting the generation model according to an expected answer of the training sample and the predicted answer, wherein the prompt of the training sample includes an original query, a change query, a search result and a target instruction.

13 . The method of claim 12 , wherein the prompt is assembled by:
obtaining the original query and the change query corresponding to the original query by sampling a change query dictionary; obtaining a multi-intent query according to the original query and the change query; inputting the multi-intent query into a search engine to obtain a reference result; and assembling a task description area, an interactive information area, a search result area and an instruction area of the prompt according to the original query, the change query, the reference result and the target instruction.

14 . An electronic device, comprising:
at least one processor; and a memory connected in communication with the at least one processor; wherein the memory stores an instruction executable by the at least one processor, and the instruction, when executed by the at least one processor, enables the at least one processor to execute the method of claim 1 .

15 . The electronic device of claim 14 , wherein the acquiring of the change query corresponding to the input query, comprises:
searching a change query dictionary for the change query corresponding to the input query.

16 . An electronic device, comprising:
at least one processor; and a memory connected in communication with the at least one processor; wherein the memory stores an instruction executable by the at least one processor, and the instruction, when executed by the at least one processor, enables the at least one processor to execute the method of claim 12 .

17 . The electronic device of claim 16 , wherein the prompt is assembled by:
obtaining the original query and the change query corresponding to the original query by sampling a change query dictionary; obtaining a multi-intent query according to the original query and the change query; inputting the multi-intent query into a search engine to obtain a reference result; and assembling a task description area, an interactive information area, a search result area and an instruction area of the prompt according to the original query, the change query, the reference result and the target instruction.

18 . A non-transitory computer readable storage medium storing a computer instruction wherein the computer instruction causes a computer to perform the method of claim 1 .

19 . The non-transitory computer readable storage medium of claim 18 , wherein the acquiring of the change query corresponding to the input query, comprises:
searching a change query dictionary for the change query corresponding to the input query.

20 . A non-transitory computer readable storage medium storing a computer instruction wherein the computer instruction causes a computer to perform the method of claim 12 .

Assignees

• Baidu Com Times Tech Beijing Co Ltd

Inventors

• Lu Yujing
• Xia Qiaolin
• Wang Kaixiang
• Wang Hao
• Shi Haibo
• Gu Simiu

Classifications

• G06N20/00

  Machine learning · CPC title

• G06F40/30

  Semantic analysis · CPC title

• G06F16/3344

  using natural language analysis · CPC title

• G06F16/3349 Primary

  Reuse of stored results of previous queries · CPC title

• G06F16/90324 Primary

  using system suggestions · CPC title