Answer information generation method

Patent metadata

Field	Value
Publication number	US-12411876-B2
Application number	US-202418747415-A
Country	US
Kind code	B2
Filing date	Jun 18, 2024
Priority date	Nov 27, 2023
Publication date	Sep 9, 2025
Grant date	Sep 9, 2025

Abstract

A method is provided. The method includes: obtaining, in response to receiving a question text from a user, a semantic vector of the question text and event information related to a specific field; obtaining a plurality of candidate documents from a document library of the specific field based on at least two of the semantic vector of the question text, the at least one piece of argument information and the event category; determining quality evaluation information for each candidate document in the plurality of candidate documents based on the event category; and determining at least one target document from the plurality of candidate documents based on the quality evaluation information of each candidate document and a correlation between each candidate document and the question text, to obtain, based on the at least one target document, answer information used to answer the question text.

First claim

What is claimed is:

1. An answer information generation method based on a large language model, comprising:
obtaining, in response to receiving a question text from a user, a semantic vector of the question text and event information related to a specific field, wherein the event information includes an event category concerning the question text and at least one piece of argument information in the question text;
obtaining a plurality of candidate documents from a document library of the specific field based on at least two of the semantic vector of the question text, the at least one piece of argument information and the event category;
determining, based on the event category, at least one document evaluation dimension corresponding to the event category, wherein:
the at least one document evaluation dimension corresponds to at least one aspect of document quality,
each document evaluation dimension of the at least one document evaluation dimension corresponds to a plurality of document categories, and
the plurality of document categories are determined based on an aspect of document quality that a corresponding document evaluation dimension focuses on;

determining, based on the event category, a quality score corresponding to each document category of the plurality of document categories of each document evaluation dimension;
determining quality evaluation information for each candidate document in the plurality of candidate documents based on a document category corresponding to each of the at least one document evaluation dimension of the candidate document and a quality score corresponding to the event category for the document category;
and
determining at least one target document from the plurality of candidate documents based on the quality evaluation information of each candidate document and a correlation between each candidate document and the question text, to obtain, based on the at least one target document, answer information used to answer the question text.

2. The method according to claim 1 , wherein the determining the at least one target document from the plurality of candidate documents based on the quality evaluation information of each candidate document and a correlation between each candidate document and the question text comprises:
determining a comprehensive score for each candidate document in the plurality of candidate documents based on the quality evaluation information of each candidate document and the correlation between each candidate document and the question text; and
determining, as the at least one target document, at least one candidate document whose comprehensive score meets a preset condition in the plurality of candidate documents.

3. The method according to claim 2 , wherein the quality evaluation information includes a quality score corresponding to each of at least one document evaluation dimension of a corresponding candidate document, and wherein the determining the comprehensive score for each candidate document in the plurality of candidate documents based on the quality evaluation information of each candidate document and the correlation between each candidate document and the question text comprises:
performing, for each candidate document in the plurality of candidate documents, following operations:
inputting at least the question text, the candidate document, and the quality evaluation information for the candidate document into a trained ranking model;
determining the correlation between the candidate document and the question text by using the ranking model based on at least the question text and the candidate document; and

determining the comprehensive score for the candidate document by using the ranking model based on the quality evaluation information for the candidate document and the correlation between the candidate document and the question text.

4. The method according to claim 3 , wherein the document library includes a plurality of preset documents, wherein each preset document of the plurality of preset documents includes a corresponding document semantic vector, at least one document event category and at least one piece of document argument information, and wherein the obtaining the plurality of candidate documents from the document library of the specific field based on at least two of the semantic vector of the question text, the at least one piece of argument information, and the event category comprises:
retrieving, based on the semantic vector of the question text and the corresponding document semantic vector of each preset document in the document library a plurality of first candidate documents with highest semantic similarities from the document library;
obtaining a plurality of second candidate documents from the document library, wherein each second candidate document of the plurality of second candidate documents meets at least one of following conditions: a document event category of the second candidate document matches the event category; and the second candidate document includes at least one piece of document argument information that matches one or more pieces of argument information in the at least one piece of argument information; and
obtaining the plurality of candidate documents based on the plurality of first candidate documents and the plurality of second candidate documents.

5. The method according to claim 4 , wherein obtaining the document semantic vector of each preset document in the document library comprises:
performing, for each preset document in the document library, following operations:
paragraphing the preset document to obtain at least one document paragraph;
obtaining at least one paragraph semantic vector corresponding to the at least one document paragraph; and
obtaining the document semantic vector of the preset document based on the at least one paragraph semantic vector.

6. The method according to claim 5 , wherein obtaining, based on the at least one target document, the answer information used to answer the question text comprises:
organizing the question text and the at least one target document into an instruction text based on a preset instruction template; and
inputting the instruction text into an answer information generation model to obtain the answer information by the answer information generation model.

7. An electronic device, comprising:
one or more processors; and
a memory storing one or more programs configured to be executed by the one or more processors, the one or more programs comprising instructions for:
obtaining, in response to receiving a question text from a user, a semantic vector of the question text and event information related to a specific field, wherein the event information includes an event category concerning the question text and at least one piece of argument information in the question text;
obtaining a plurality of candidate documents from a document library of the specific field based on at least two of the semantic vector of the question text, the at least one piece of argument information and the event category;
determining, based on the event category, at least one document evaluation dimension corresponding to the event category, wherein:
the at least one document evaluation dimension corresponds to at least one aspect of document quality,
each document evaluation dimension of the at least one document evaluation dimension corresponds to a plurality of document categories, and
the plurality of document categories are determined based on an aspect of document quality that a corresponding document evaluation dimension focuses on;

determining, based on the event category, a quality score corresponding to each document category of the plurality of document categories of each doc

Assignees

• Beijing Baidu Netcom Sci & Tech Co Ltd

Inventors

• Cui Zhuangzhuang
• Fu Bo

Classifications

• G06F40/30

  Semantic analysis · CPC title

• G06F16/3347

  using vector based model · CPC title

• G06F16/353

  into predefined classes · CPC title

• G06F40/186 Primary

  Templates · CPC title

• G06F16/3329 Primary

  Natural language query formulation · CPC title