CN-122019700-A - Intelligent question-answering method, device and storage medium

CN122019700ACN 122019700 ACN122019700 ACN 122019700ACN-122019700-A

Abstract

The embodiment of the application provides an intelligent question-answering method, equipment and a storage medium, wherein the method comprises the steps of judging the domain of a user input problem, wherein the domain comprises a general knowledge domain and a professional knowledge domain; determining a language model adopted for analyzing the user input problem according to the field, analyzing the user input problem according to the language model, and generating a reply to the user input problem. By judging the belonging field of the user input problem and analyzing the problems in the professional knowledge field and the general knowledge field by adopting different models respectively, the models can adopt a targeted processing strategy according to the belonging field of the user input problem, the continuity of multiple rounds of conversations is ensured, and the solving capability of the models for complex problems is improved.

Inventors

LI XINQIN
LI GUOHUA
Zhao Yinjiang
DAI MINGRUI
SHI WEIFENG
YANG TAOCUN
Du Wenran
LI WENHAO
HOU BO

Assignees

中国国家铁路集团有限公司
中国铁道科学研究院集团有限公司
中国铁道科学研究院集团有限公司电子计算技术研究所
北京经纬信息技术有限公司

Dates

Publication Date: 20260512
Application Date: 20251212

Claims (11)

1. An intelligent question-answering method is characterized by comprising the following steps: judging the belonging field of the user input problem, wherein the belonging field comprises a general knowledge field and a professional knowledge field; Determining a language model adopted for analyzing the user input problem according to the belonging field; and analyzing the user input problem according to the language model to generate a reply to the user input problem.
2. The method of claim 1, wherein determining a language model to be used for parsing the user input question based on the domain comprises: If the domain of the user input problem is a general knowledge domain, analyzing the user input problem by adopting a first language model; if the domain of the user input problem is the professional knowledge domain, performing reliability matching on knowledge subgraphs related to the user input problem by adopting a second language model to generate a trusted knowledge subgraph; analyzing the user input problem according to the trusted knowledge subgraph by adopting the first language model; Wherein the parameter of the first language model is larger than the parameter of the second language model.
3. The method as recited in claim 2, further comprising: if the domain to which the user input question belongs is a general knowledge domain, storing the user input question and a reply of the first language model to the user input question as a first history; if the domain to which the user input question belongs is a professional knowledge domain, storing the user input question and a reply of the first language model to the user input question as a second history; The first history record is used as a context adopted when the first language model is used for analyzing the user input problems in the general knowledge field, and the second history record is used as a context adopted when the first language model is used for analyzing the user input problems in the professional knowledge field.
4. The method of claim 2, wherein employing a second language model to perform reliability matching on knowledge subgraphs related to the user input problem, generating a trusted knowledge subgraph comprises: determining the comprehensive evaluation result of the knowledge subgraph according to the topic correlation, the context coherence and/or the semantic coherence of the knowledge subgraph; Determining the trusted knowledge subgraph according to a comparison result of the comprehensive evaluation result of the knowledge subgraph and a first preset evaluation result; The topic correlation of the knowledge subgraph is the topic correlation of the knowledge subgraph and the user input problem, the context coherence of the knowledge subgraph is the context coherence of the knowledge subgraph and the history record, and the semantic coherence of the knowledge subgraph is the semantic coherence of the first language model and the generation trend of the user input problem.
5. The method of claim 4, wherein after determining the trusted knowledge sub-graph based on a comparison of the comprehensive evaluation result of the knowledge sub-graph and a first preset evaluation result, the method further comprises: if fact conflict exists between the trusted knowledge subgraphs, determining a comprehensive evaluation result of the trusted knowledge subgraphs according to the authority of the data sources of the trusted knowledge subgraphs and/or update timeliness; And determining a preferred trusted knowledge subgraph according to a comparison result of the comprehensive evaluation result of the trusted knowledge subgraph and a second preset evaluation result.
6. The method of claim 4, wherein the determining the comprehensive evaluation result of the knowledge sub-graph based on the topic relevance, the context consistency, and/or the semantic consistency of the knowledge sub-graph comprises: Determining a first evaluation result of the knowledge subgraph according to the topic relevance of the knowledge subgraph, wherein the topic relevance is positively correlated with the first evaluation result; Determining a second evaluation result of the knowledge subgraph according to the context coherence of the knowledge subgraph, wherein the context coherence is positively correlated with the second evaluation result; determining a third evaluation result of the knowledge subgraph according to the semantic consistency of the knowledge subgraph, wherein the semantic consistency is positively correlated with the third evaluation result; and determining the comprehensive evaluation result of the knowledge subgraph according to the weighted values of the first evaluation result, the second evaluation result and the third evaluation result of the knowledge subgraph.
7. The method of claim 5, wherein the determining the comprehensive evaluation result of the trusted knowledge base according to the authority of the data sources and/or the update timeliness of the trusted knowledge base comprises: Determining a first evaluation result of the trusted knowledge subgraph according to the authority of the data source of the trusted knowledge subgraph, wherein the first evaluation result of the trusted knowledge subgraph is positively correlated with the authority of the data source; Determining a second evaluation result of the trusted knowledge subgraph according to the update timeliness of the trusted knowledge subgraph, wherein the second evaluation result of the trusted knowledge subgraph is positively correlated with the update timeliness of the trusted knowledge subgraph; And determining the comprehensive evaluation result of the trusted knowledge subgraph according to the weighted values of the first evaluation result and the second evaluation result of the trusted knowledge subgraph.
8. The method of claim 4, wherein the determining the trusted knowledge sub-graph based on a comparison of the comprehensive evaluation result of the knowledge sub-graph and a first preset evaluation result comprises: And determining the knowledge subgraph with the comprehensive evaluation result of the knowledge subgraph being better than the first preset evaluation result as the trusted knowledge subgraph.
9. The method of claim 5, wherein determining a preferred trusted knowledge base based on a comparison of the integrated evaluation result of the trusted knowledge base with a second preset evaluation result comprises: And determining the trusted knowledge subgraph with the comprehensive evaluation result of the trusted knowledge subgraph being better than the second preset evaluation result as the preferable trusted knowledge subgraph.
10. An electronic device, comprising: A processor; a memory; and a computer program, wherein the computer program is stored in the memory, which when executed by the processor, causes the electronic device to perform the method of any one of claims 1 to 9.
11. A computer readable storage medium, characterized in that the computer readable storage medium stores a computer program which, when executed by a processor, implements the method of any one of claims 1 to 9.

Description

Intelligent question-answering method, device and storage medium Technical Field The application belongs to the technical field of artificial intelligence, and particularly relates to an intelligent question-answering method, equipment and a storage medium. Background Along with the rapid development of artificial intelligence and natural language processing technology, the intelligent question-answering system has been widely applied in various industries, especially in public service fields such as railways, aviation and the like, and the intelligent question-answering system can continuously provide services such as information consultation, business handling guidance and the like for users for 24 hours, so that the service efficiency is greatly improved, and the pressure of artificial customer service is reduced. In the railway industry, intelligent question-answering systems have become a key infrastructure for optimizing service experience and ensuring operation efficiency. In principle, intelligent question-answering systems generally rely on a large language model (Large Language Model, LLM), i.e. by which questions entered by a user are parsed and corresponding replies are generated. In practical applications, the problem of user input presents a significant binarization feature. The general knowledge questions are mainly used for surrounding daily travel information such as train time, ticket service and the like, and the questioners are mostly ordinary passengers, the other general knowledge questions are special knowledge questions which are characterized by complex logic and need knowledge base support, and relate to the core knowledge fields such as scheduling rules, signal technology, safety standards and the like, and the questioners are usually professional technicians in railways. In the related art, for the two types of questions with distinct properties, the intelligent question-answering system usually adopts a unified processing mode. Specifically, two types of questions with distinct properties are input into a large language model without distinction, so that the model automatically analyzes and generates replies. However, because a differential processing strategy is not adopted for the professional knowledge problem and the general knowledge problem, the two problems are mutually interfered, so that misjudgment of a model on the real intention of a user (the professional knowledge problem or the general knowledge problem) is caused, a series of linkage problems such as multi-round dialogue interruption, complex problem solving failure and the like are caused, and the user experience is obviously influenced. It should be noted that the information disclosed in the background section of the present application is only for enhancement of understanding of the general background of the present application and should not be taken as an admission or any form of suggestion that this information forms the prior art that is well known to a person skilled in the art. Disclosure of Invention The embodiment of the application provides an intelligent question-answering method, equipment and a storage medium, which are beneficial to solving the problems of multiple rounds of dialogue interruption, complex problem solving failure and the like caused by misjudgment of a model on the real intention of a user due to the fact that a processing strategy for differentiating a general knowledge problem and a professional knowledge problem is not provided in the related technology. In a first aspect, an embodiment of the present application provides an intelligent question-answering method, including: judging the belonging field of the user input problem, wherein the belonging field comprises a general knowledge field and a professional knowledge field; Determining a language model adopted for analyzing the user input problem according to the belonging field; and analyzing the user input problem according to the language model to generate a reply to the user input problem. In one possible implementation manner, the determining, according to the domain, a language model used for parsing the user input problem includes: If the domain of the user input problem is a general knowledge domain, analyzing the user input problem by adopting a first language model; if the domain of the user input problem is the professional knowledge domain, performing reliability matching on knowledge subgraphs related to the user input problem by adopting a second language model to generate a trusted knowledge subgraph; analyzing the user input problem according to the trusted knowledge subgraph by adopting the first language model; Wherein the parameter of the first language model is larger than the parameter of the second language model. In one possible implementation, the method further includes: if the domain to which the user input question belongs is a general knowledge domain, storing the user input question and a reply of