Multi-system-based intelligent question answering method and apparatus, and device

Patent metadata

Field	Value
Publication number	US-12511318-B2
Application number	US-202217820285-A
Country	US
Kind code	B2
Filing date	Aug 17, 2022
Priority date	Mar 2, 2022
Publication date	Dec 30, 2025
Grant date	Dec 30, 2025

Abstract

The present disclosure provides a multi-system-based intelligent question answering method and apparatus, and a device, relating to the field of artificial intelligence, in particular to the field of knowledge graph. The specific implementation solution is: determining a question category of question information in response to a question answering instruction of a user, wherein the question answering instruction is used to indicate the question information; determining a query engine corresponding to the question category, and invoking multiple question analysis systems corresponding to the query engine according to the query engine; and feeding back answer information to the user when the answer information corresponding to the question information is determined according to a current question analysis system in a process of processing the question information by sequentially using the multiple question analysis systems according to system priorities of the question analysis systems.

First claim

What is claimed is:

1 . A multi-system-based intelligent question answering method applied in intelligent technology and natural language technology, applied to a server that interacts with a client through digital data communication, comprising:
receiving a question answering instruction sent by the client, wherein the question answering instruction is triggered by a user through the client; determining a question category of question information in response to the question answering instruction, wherein the question answering instruction is used to indicate the question information; determining a query engine corresponding to the question category, and invoking multiple question analysis systems corresponding to the query engine according to the query engine; and sending answer information to the client when the answer information corresponding to the question information is determined according to a current question analysis system in a process of processing the question information by sequentially using the multiple question analysis systems according to system priorities of the question analysis systems, when the answer information corresponding to the question information is not determined according to the current question analysis system, a next question analysis system determined according to the system priorities of the question analysis systems is used to process the question information; wherein the query engine is used for providing users with comprehensive information services based on preset algorithms; wherein the multiple question analysis systems comprise: a knowledge graph system, a reasoning question answering system, a document question answering system, and a frequently-asked questions (FAQ) question answering system; wherein the knowledge graph system is used to indicate answer information corresponding to an entity in the question information; the reasoning question answering system is used to indicate a reasoning template corresponding to the question information, wherein the reasoning template is used to determine the answer information, the reasoning template comprises multiple filling points with an order relationship, and the filling point has a word characteristic; the document question answering system is used to indicate a paragraph corresponding to the question information, wherein the paragraph is used to determine the answer information; and the FAQ question answering system is used to indicate a further question related to the question information, wherein the further question and the question information are used to determine the answer information.

2 . The method according to claim 1 , wherein when the current question analysis system is the document question answering system, determining the answer information corresponding to the question information according to the current question analysis system comprises:
performing according to a first database in the document question answering system, similarity matching on the question information to obtain an optimal paragraph corresponding to the question information, wherein the first database comprises multiple documents, and the document has multiple paragraphs; and performing according to a first enhanced representation through knowledge integration ERNIE) model, a word count reduction processing on the optimal paragraph to obtain the answer information corresponding to the question information.

3 . The method according to claim 2 , wherein performing according to the first database in the document question answering system, the similarity matching on the question information to obtain the optimal paragraph corresponding to the question information comprises:
performing the similarity matching on the question information according to a first database in the document question answering system to obtain multiple candidate paragraphs similar to the question information, and obtaining core element information corresponding to the question information; and performing similarity calculation on the core element information and the candidate paragraph, and determining a candidate paragraph with a highest similarity as the optimal paragraph corresponding to the question information.

4 . The method according to claim 3 , wherein the core element information comprises one or more of the following: user location information, user attribute information, and information of a terminal device used by the user.

5 . The method according to claim 1 , wherein when the current question analysis system is the FAQ question answering system, determining the answer information corresponding to the question information according to the current question analysis system comprises:
performing according to a second database in the FAQ question answering system, similarity matching on the question information to obtain multiple question texts similar to the question information, wherein the second database comprises multiple question texts; and determining the answer information corresponding to the question information according to the question information and the multiple question texts similar to the question information.

6 . The method according to claim 5 , wherein determining the answer information corresponding to the question information according to the question information and the multiple question texts similar to the question information comprises:
determining the question information and the multiple question texts similar to the question information as a candidate question in a question set, wherein the candidate question has feature information; inputting the feature information of the candidate question in the question set into the preset neural network model to obtain an optimal candidate question; and determining according to a correspondence between a preset candidate question and the answer information, the answer information corresponding to the optimal candidate question as the answer information corresponding to the question information.

7 . The method according to claim 5 , further comprising:
obtaining an original document, performing a segmentation processing on the original document to obtain multiple short texts, and performing an occlusion processing on the short text according to a second ERNIE model to generate the question text in the second database; or, obtaining an original document, determining a title in the original document, and determining the title as the question text in the second database; or, obtaining an original document, performing a content structure analysis processing on the original document to obtain a short segment with a questioning characteristic, and inputting the short segment and an original paragraph of the short segment in the original document into a preset model to obtain the question text in the second database, wherein the short segment is answer information corresponding to the obtained question text.

8 . The method according to claim 1 , wherein when the current question analysis system is the knowledge graph system, determining the answer information corresponding to the question information according to the current question analysis system comprises:
extracting an entity in the question information; and identifying according to a preset knowledge graph in the knowledge graph system and the third ERNIE model, the entity in the question information to obtain the answer information corresponding to the question information, wherein the preset knowledge graph comprises multiple entities, there is a connection relationship between the entities in the preset knowledge graph, and the third ERNIE model is used to process the entity in the question information.

9 . The method accordin

Assignees

• Beijing Baidu Netcom Sci & Tech Co Ltd

Inventors

• Feng Xinwei
• Tian Meng
• Li Feifei
• Shi Hongjian
• Jiang Wenbin
• Wu Xueqian
• Guo Chenyang
• Wang Yu
• Sun Yu
• Chen Shuaiyu

Classifications

• G06F40/30

  Semantic analysis · CPC title

• G06F16/2455

  Query execution · CPC title

• G06N3/02

  Neural networks · CPC title

• G06F16/9038

  Presentation of query results · CPC title

• G06F16/90332

  Natural language query formulation or dialogue systems · CPC title