CN-121998090-A - Dual-path engineering process intelligent question-answering system and method based on posterior probability dynamic routing

CN121998090ACN 121998090 ACN121998090 ACN 121998090ACN-121998090-A

Abstract

The application discloses a dual-path engineering process intelligent question-answering system and method based on posterior probability dynamic routing, which relate to the field of computers and comprise a data processing module, a data processing module and a data processing module, wherein the data processing module is used for preprocessing multi-source heterogeneous engineering data to obtain preprocessed data; the system comprises a user interaction module, an information processing module and a user interaction module, wherein the user interaction module is used for carrying out intention understanding and task decomposition on a received input problem through a large language model agent, selecting an adaptive information retrieval mode and generating a query request, the information processing module is used for carrying out knowledge retrieval and answer generation according to the query request by utilizing a preset symbolized knowledge graph and/or a semantic vector database, and the symbolized knowledge graph and the semantic vector database are constructed based on the preprocessing data. The application can ensure that the finally generated answers are not only comprehensive and consistent, but also have the accuracy of the spectrum level on the key technological parameters, effectively eliminates the numerical illusion and improves the reliability and the fidelity of the answers.

Inventors

HUANG JIN
Shen Jundao
WANG YONGJING
NIU ZHIXIANG
ZHANG HUA
ZHANG WANQIANG
HENG XIN
ZHOU ZHUOCHENG
LI KEDONG

Assignees

西南交通大学

Dates

Publication Date: 20260508
Application Date: 20260123

Claims (10)

1. A dual-path engineering process intelligent question-answering system based on posterior probability dynamic routing is characterized by comprising the following steps: The data processing module is used for preprocessing the multi-source heterogeneous engineering data to obtain preprocessed data; The user interaction module is used for carrying out intention understanding and task decomposition on the received input problem through the large language model agent, selecting an adaptive information retrieval mode and generating a query request; and the information processing module is used for executing knowledge retrieval and answer generation by utilizing a preset symbolic knowledge graph and/or a semantic vector database according to the query request, and the symbolic knowledge graph and the semantic vector database are constructed based on the preprocessing data.
2. The dual path engineering process intelligent question-answering system based on posterior probability dynamic routing of claim 1, wherein the data processing module comprises: The data cleaning unit is used for carrying out multi-mode analysis on the multi-source heterogeneous engineering data, extracting text characteristic data, removing noise and redundant information and obtaining cleaning data; And the data conversion unit is used for normalizing the cleaning data into standard data and generating the preprocessing data.
3. The dual-path engineering process intelligent question-answering system based on posterior probability dynamic routing of claim 1, further comprising a bimodal engineering knowledge base construction module, the bimodal engineering knowledge base construction module comprising: The knowledge graph construction unit is used for constructing pivot nodes and process parameter nodes based on the structured data and/or the semi-structured data in the preprocessing data, extracting causal logic relations based on unstructured data in the preprocessing data and generating a symbolized knowledge graph according to the pivot nodes, the process parameter nodes and the causal logic relations; The vector database construction unit is used for constructing a semantic vector database through a semantic blocking and vectorizing method based on unstructured text data in the preprocessing data.
4. The dual-path engineering process intelligent question-answering system based on posterior probability dynamic routing of claim 3, wherein the knowledge graph construction unit comprises: The hub node construction subunit is used for modeling a core service object in the structural data and/or the semi-structural data as a hub node and extracting attributes or related entities of the hub node in the structural data and/or the semi-structural data; A process parameter node construction subunit, configured to abstract process parameters in the structured data and/or the semi-structured data into independent process parameter nodes; The causal chain extraction subunit is used for extracting information from the unstructured data, identifying causal entities and establishing causal logic relations among the causal entities; And the map generation subunit is used for constructing a symbolized knowledge map with explicit logic association based on the pivot node, the process parameter node and the causal logic relationship.
5. A dual path engineering process intelligent question-answering system based on posterior probability dynamic routing according to claim 3, wherein the vector database construction unit comprises: the semantic segmentation subunit is used for segmenting the unstructured text data by adopting a semantic-based sliding window strategy to obtain a plurality of text blocks; gao Weixiang a quantization subunit for converting the text block into a high-dimensional dense vector; and the metadata association subunit is used for binding the high-dimensional dense vector and the associated metadata and storing the high-dimensional dense vector and the associated metadata into a semantic vector database.
6. The dual-path engineering process intelligent question-answering system based on posterior probability dynamic routing of claim 1, wherein the user interaction module comprises: The confidence coefficient calculating unit is used for calculating posterior probability confidence coefficient scores of the input problems respectively belonging to different query types; the intention understanding judging unit is used for judging the input problem as an accurate fact query, a fuzzy semantic query or a complex mixed query according to a preset threshold value and the posterior probability confidence score; And the retrieval path selection unit is used for selecting and activating a knowledge graph retrieval path, a vector retrieval path or a mixed retrieval path according to the judging result.
7. The dual path engineering process intelligent question-answering system based on posterior probability dynamic routing of claim 6, wherein the information processing module comprises a hybrid search unit comprising: The task receiving subunit is used for receiving the mixed path instruction sent by the user interaction module; the double-path searching subunit is used for executing the knowledge graph searching path and the vector searching path in parallel to obtain a structured fact set and a related text fragment set; A structured injection subunit, configured to inject the structured fact set into the related text segment set, and generate an enhancement context; and the numerical consistency checking subunit is used for executing numerical consistency checking and correction on the enhanced context.
8. The dual-path engineering process intelligent question-answering system based on posterior probability dynamic routing of claim 7, wherein the information processing module further comprises a generating answer generating unit, specifically: The answer generation and tracing subunit is used for receiving the enhanced context and generating a final natural language answer, and the reference mark which converts the data source identifier embedded in the enhanced context into a preset format in the natural language answer generation is added to the corresponding parameter; The closed-loop evolution subunit is used for acquiring feedback data of the natural language answer, triggering a differential correction strategy according to the type of the feedback data, and updating a symbolized knowledge graph, a semantic vector database or optimizing decision logic of the user interaction module.
9. The dual path engineering process intelligent question-answering system based on posterior probability dynamic routing of claim 8, wherein the closed loop evolution subunit comprises: The feedback collection component is used for acquiring explicit feedback and implicit feedback of the natural language answers; an anomaly log generation component for storing the explicit feedback, the implicit feedback, the input problem, and the retrieved data in association as an anomaly log; a knowledge base correction component for updating the signed knowledge graph and/or semantic vector database that is confirmed to be a fact error based on the anomaly log; And the agent optimization component is used for determining an optimization data set based on the abnormal log and determining logic for the user interaction module based on the optimization data set.
10. A dual-path engineering process intelligent question-answering method based on posterior probability dynamic routing is characterized by comprising the following specific steps: Preprocessing multi-source heterogeneous engineering data to obtain preprocessed data; Constructing a symbolized knowledge graph and a semantic vector database based on the preprocessing data; Performing intention understanding and task decomposition on the received input problem through a large language model agent, and selecting an adaptive information retrieval mode to generate a query request; And performing knowledge retrieval and answer generation by using the symbolized knowledge graph and/or the semantic vector database according to the query request.

Description

Dual-path engineering process intelligent question-answering system and method based on posterior probability dynamic routing Technical Field The application relates to the field of computers, in particular to a dual-path engineering process intelligent question-answering system and method based on posterior probability dynamic routing. Background At present, knowledge intelligent management in the engineering technology field becomes a core link of industrial digital transformation, but the prior art is difficult to effectively search multi-source heterogeneous data in high-precision process scenes such as welding, casting and the like and meet the requirements of high-fidelity question-answering. The existing retrieval enhancement generation (RAG) technology has the obvious bottlenecks that firstly, an automatic standardized processing mechanism for multi-source heterogeneous engineering data is lacking, semi-structure and unstructured data such as a process card, a design standard, a test report and the like cannot be subjected to term normalization and unit unified conversion, so that semantic ambiguity and noise at the bottom layer of a knowledge base are serious, secondly, retrieval path allocation depends on a static classification model, self-adaptive routing decision cannot be performed based on posterior probability confidence of query intention, path misselection and key information omission are likely to occur when the mixed engineering problem of fuzzy description and accurate parameters is faced, thirdly, a cross-mode retrieval result fusion mode is shallow, structured facts and unstructured contexts are input into a large model only through text splicing, and an active anchoring and calibration mechanism for accurate values is lacking, so that the model is easy to be subjected to semantic interference, a numerical illusion is generated, and the accuracy requirement of 'zero tolerance' of technological parameters is difficult to meet. Therefore, how to overcome the above-mentioned drawbacks is a problem to be solved by those skilled in the art. Disclosure of Invention In view of the above, the present invention provides a dual-path engineering process intelligent question-answering system and method based on posterior probability dynamic routing, which overcomes the above-mentioned drawbacks. In order to achieve the above object, the present application provides the following solutions: In a first aspect, the present application provides a dual-path engineering process intelligent question-answering system based on posterior probability dynamic routing, including: The data processing module is used for preprocessing the multi-source heterogeneous engineering data to obtain preprocessed data; The user interaction module is used for carrying out intention understanding and task decomposition on the received input problem through the large language model agent, selecting an adaptive information retrieval mode and generating a query request; And the information processing module is used for executing knowledge retrieval and answer generation by utilizing the preset symbolic knowledge graph and/or semantic vector database according to the query request, and the symbolic knowledge graph and the semantic vector database are constructed based on the preprocessing data. Optionally, the data processing module includes: The data cleaning unit is used for carrying out multi-mode analysis on the multi-source heterogeneous engineering data, extracting text characteristic data, removing noise and redundant information and obtaining cleaning data; And the data conversion unit is used for normalizing the cleaning data into standard data and generating the preprocessing data. Optionally, the system further includes a bimodal engineering knowledge base construction module, the bimodal engineering knowledge base construction module including: The knowledge graph construction unit is used for constructing pivot nodes and process parameter nodes based on the structured data and/or the semi-structured data in the preprocessing data, extracting causal logic relations based on unstructured data in the preprocessing data and generating a symbolized knowledge graph according to the pivot nodes, the process parameter nodes and the causal logic relations; The vector database construction unit is used for constructing a semantic vector database through a semantic blocking and vectorizing method based on unstructured text data in the preprocessing data. Optionally, the knowledge graph construction unit includes: The hub node construction subunit is used for modeling a core service object in the structural data and/or the semi-structural data as a hub node and extracting attributes or related entities of the hub node in the structural data and/or the semi-structural data; A process parameter node construction subunit, configured to abstract process parameters in the structured data and/or the semi-structured data into indepe