Search

CN-121979899-A - Automatic template updating method and system

CN121979899ACN 121979899 ACN121979899 ACN 121979899ACN-121979899-A

Abstract

The invention discloses a template automatic updating method and a system, which relate to the technical field of audit informatization, and are characterized in that original policy information is standardized to obtain standardized corpus data by responding to an audit template updating request, a pre-training fine-tuning large language model comprising a policy audit semantic analysis mapping layer and a mapping result calibration optimization layer is called, the standardized corpus is input into the policy audit semantic analysis mapping layer to complete semantic analysis and mapping so as to output policy audit initial mapping data, the initial mapping data is checked and corrected by the mapping result calibration optimization layer to obtain structured mapping correction data, a target industry template is matched based on the structured data, if the target industry template does not meet a preset template evaluation condition, fine tuning is carried out on the fine-tuning large language model and the mapping process is circularly executed until the template meets the condition, the suitability and the accuracy of the target industry template on the policy requirement are ensured, and the problem of insufficient accuracy of the policy and the corresponding result in the prior art is solved.

Inventors

  • TAN HAI

Assignees

  • 南京审计大学

Dates

Publication Date
20260505
Application Date
20260126

Claims (10)

  1. 1. A method for automatically updating a template, comprising: responding to the audit template updating request, acquiring original policy information and carrying out standardization to obtain standardized corpus data; Acquiring a pre-trained fine-tuning large language model, wherein the fine-tuning large language model comprises a policy audit semantic analysis mapping layer and a mapping result calibration optimization layer; Inputting the standardized corpus data into the policy audit semantic analysis mapping layer to perform semantic analysis and mapping, and outputting policy audit initial mapping data; Inputting the initial mapping data of the policy audit into the mapping result calibration optimization layer to carry out verification and correction to generate structured mapping correction data; matching the target industry template based on the structured mapping correction data; When the target industry template does not meet the preset template evaluation condition, performing fine tuning on the fine-tuning large language model, and jumping to execute the step of inputting the standardized corpus data into the policy audit semantic analysis mapping layer for semantic analysis and mapping until the target industry template meets the preset template evaluation condition.
  2. 2. The automatic template updating method according to claim 1, wherein the obtaining and normalizing the original policy information in response to the audit template updating request to obtain normalized corpus data includes: Responding to the audit template update request to obtain original policy information; Data cleaning is carried out on the original policy information to obtain policy text data; extracting key corpus from the policy text data to obtain policy core element data; performing semantic noise reduction and unnecessary information screening on the policy core element data to obtain key corpus data; Non-standardized term conversion is carried out on the key corpus data, and standardized element data are obtained; and carrying out structured packaging on the standardized element data to obtain standardized corpus data.
  3. 3. The automatic template updating method according to claim 1, wherein the policy audit semantic analysis mapping layer includes a corpus preprocessing unit, a domain feature enhancement extraction unit, a policy audit semantic fusion unit and a mapping reasoning unit, the standardized corpus data is adopted to input the policy audit semantic analysis mapping layer for semantic analysis and mapping, and policy audit initial mapping data is output, and the method includes: inputting the standardized corpus data into the corpus preprocessing unit to sequentially perform format verification, noise filtering and policy audit term alignment, and outputting the standardized corpus data; inputting the normalized corpus data into the field feature enhancement extraction unit to extract the dual features of the policy semantics and the audit factors and output the dual-field feature vector; inputting the dual-domain feature vector into the policy audit semantic fusion unit to perform feature fusion and normalization processing, and outputting fusion features; And inputting the mapping reasoning unit to map by adopting the fusion characteristic, and outputting the initial mapping data of the policy audit.
  4. 4. The automatic template updating method according to claim 3, wherein the bi-domain feature vector includes a policy semantic feature vector and an audit element feature vector, the feature fusion and normalization processing is performed by the policy audit semantic fusion unit using the bi-domain feature vector, and the outputting of the fusion feature includes: Performing bidirectional semantic matching on the policy semantic feature vector and the audit element feature vector, and calculating element-level similarity matrixes among vectors to obtain associated weight data; based on the association weight data, respectively carrying out weighting processing on the policy semantic feature vector and the audit element feature vector to obtain policy association features and audit association features; performing channel dimension splicing on the policy-related features and the audit-related features, and performing feature integration on the spliced features by using 1X 1 convolution to obtain related feature data; performing element weighted fusion on the associated feature data and the dual-domain feature vector according to preset weights to obtain initial fusion features; and carrying out layer normalization processing on the initial fusion characteristics to obtain fusion characteristics.
  5. 5. The automatic template updating method according to claim 1, wherein the inputting the initial mapping data of the policy audit into the mapping result calibration optimization layer for verification and correction to generate the structured mapping correction data comprises: Compliance verification is carried out on the policy audit initial mapping data to obtain compliance mapping data and abnormal mapping data; correcting the abnormal mapping data to obtain corrected mapping data; And integrating and structuring the compliance mapping data and the correction mapping data to obtain structured mapping correction data.
  6. 6. The method of any one of claims 1-5, wherein the matching the target industry template based on the structured mapping correction data comprises: Extracting applicable industry labels and audit elements from the structured mapping correction data as template matching search keywords; Searching a preset industry template set by utilizing the search keywords, and screening a plurality of candidate industry templates consistent with the applicable industry; Carrying out semantic comparison on the audit elements and audit indexes of each candidate industry template, and calculating matching similarity; and selecting a candidate industry template with highest similarity as a target industry template based on the matching similarity sorting.
  7. 7. The automatic template updating method according to claim 1, wherein the preset template evaluation condition is that a coverage ratio of an audit index of the target industry template to the audit element in the structural mapping correction data is not lower than a preset coverage ratio threshold.
  8. 8. An automatic template updating system, comprising: the policy acquisition and analysis module is used for responding to the audit template update request, acquiring original policy information and carrying out standardization to obtain standardized corpus data; The model acquisition module is used for acquiring a pre-trained fine-tuning large language model, wherein the fine-tuning large language model comprises a policy audit semantic analysis mapping layer and a mapping result calibration optimization layer; The policy semantic understanding and influence analyzing module is used for inputting the standardized corpus data into the policy audit semantic analysis mapping layer to conduct semantic analysis and mapping and outputting policy audit initial mapping data; The verification and correction module is used for inputting the initial mapping data of the policy audit into the mapping result calibration optimization layer to carry out verification and correction, and generating structured mapping correction data; the template generation and configuration module is used for matching the target industry template based on the structural mapping correction data; And the template pushing and feedback module is used for carrying out fine adjustment on the fine adjustment large language model when the target industry template does not meet the preset template evaluation condition, and skipping to execute the step of inputting the standardized corpus data into the policy audit semantic analysis mapping layer for semantic analysis and mapping until the target industry template meets the preset template evaluation condition.
  9. 9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program that, when executed by the processor, causes the processor to perform the steps of the automatic template updating method according to any one of claims 1-7.
  10. 10. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when executed, implements the automatic template updating method according to any of claims 1-7.

Description

Automatic template updating method and system Technical Field The invention relates to the technical field of audit informatization, in particular to a template automatic updating method and system. Background Along with the rapid development of digital economy, the informatization and intelligent transformation of audit work become industry core trends, an audit template is used as a key carrier for the development of audit business, the adaptation efficiency and the accuracy of the audit template to the latest industry rules are directly related to the compliance and the professionality of the audit work, and under the background, the automatic update of the audit template is realized by depending on a pre-training language model, so that the method becomes an important direction of technical research, development and industry application. At present, related technologies try to apply a large language model to an audit template updating scene, and assist corresponding association of policy related information and audit related content by extracting key information of the two types of information, but when the two types of information are processed by the technologies, a simple integration or unidirectional association mode is mostly adopted, so that deep mutual adaptation is not carried out on the characteristics of the two types of information, targeted processing is not carried out on the basis of the association degree between the information, and a processing mechanism for adapting the characteristics of the two types of information cannot be constructed. The limitation of the processing mode can lead to the fact that the integrated information cannot fully bear the internal relation between the policies and the audit related content, the effective association degree of the information is insufficient, and further the generated policies and the audit corresponding results are insufficient in accuracy, the updating accuracy and efficiency of the audit templates are finally affected, and the actual requirements of the fields of finance, enterprises and the like for the rapid policy adjustment adaptation of the audit templates cannot be met. Disclosure of Invention The invention provides a template automatic updating method and system, which solve the technical problems that in the prior art, the integrated information cannot fully bear the inherent connection of the policy related information and the audit related content due to lack of a deep bidirectional adaptation and association degree pertinence processing mechanism, so that the generated corresponding result of the policy and the audit is insufficient in accuracy. The automatic template updating method provided by the first aspect of the invention comprises the following steps: responding to the audit template updating request, acquiring original policy information and carrying out standardization to obtain standardized corpus data; Acquiring a pre-trained fine-tuning large language model, wherein the fine-tuning large language model comprises a policy audit semantic analysis mapping layer and a mapping result calibration optimization layer; Inputting the standardized corpus data into the policy audit semantic analysis mapping layer to perform semantic analysis and mapping, and outputting policy audit initial mapping data; Inputting the initial mapping data of the policy audit into the mapping result calibration optimization layer to carry out verification and correction to generate structured mapping correction data; matching the target industry template based on the structured mapping correction data; When the target industry template does not meet the preset template evaluation condition, performing fine tuning on the fine-tuning large language model, and jumping to execute the step of inputting the standardized corpus data into the policy audit semantic analysis mapping layer for semantic analysis and mapping until the target industry template meets the preset template evaluation condition. Optionally, the responding to the audit template update request obtains the original policy information and performs standardization to obtain standardized corpus data, including: Responding to the audit template update request to obtain original policy information; Data cleaning is carried out on the original policy information to obtain policy text data; extracting key corpus from the policy text data to obtain policy core element data; performing semantic noise reduction and unnecessary information screening on the policy core element data to obtain key corpus data; Non-standardized term conversion is carried out on the key corpus data, and standardized element data are obtained; and carrying out structured packaging on the standardized element data to obtain standardized corpus data. Optionally, the policy audit semantic analysis mapping layer includes a corpus preprocessing unit, a domain feature enhancement extraction unit, a policy audit semantic fu