Examination and approval method, device and equipment for tobacco monopoly license and medium
By processing historical application data and fine-tuning large-scale models, an intelligent model for approval of tobacco monopoly licenses is generated, which solves the problem of inefficiency of traditional manual approval and realizes an efficient and accurate intelligent approval process.
Patent Information
- Application Number
- CN202510023150.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-07
- Publication Date
- 2025-04-18
AI Technical Summary
The traditional manual approval method is inefficient in the approval process of tobacco monopoly licenses, resulting in lagging in the approval process, affecting the normal business development and social and economic efficiency.
By processing the pre-collected historical application data, the target training data is generated, and the model fine-tuning and confrontation training is performed based on the large model to be fine-tuned, the target tobacco monopoly license approval model is obtained, which is used to intelligently approve the current application data.
It realizes efficient and intelligent approval of tobacco monopoly licenses, improves the efficiency and accuracy of the approval process, and reduces labor costs and subjective errors.
Smart Images
Figure CN119850144A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method, device, equipment and medium for approving a tobacco monopoly license. Background Art
[0002] With the rapid development of information technology, traditional manual approval methods appear to be inefficient in many business scenarios, especially those involving complex document approval processes, such as the approval of tobacco monopoly licenses. Traditional manual approval not only requires a lot of time and manpower costs, but is also prone to unstable approval quality due to subjective factors or irregular operations. This inefficient approval method often causes delays in the approval process when faced with a large number of applications, seriously affecting the normal development of business and the efficiency of social and economic operations. In recent years, with the continuous advancement of artificial intelligence technology, especially the rise of large language models, the ability to process information intelligently has been greatly improved. Large models can perform well on a variety of tasks by pre-training massive data. However, due to the differences in data characteristics and task requirements of each business scenario, relying solely on pre-trained large models cannot directly meet all actual business needs. Therefore, how to extract key information from documents for tobacco monopoly licenses through large models and accurately complete the compliance approval of documents is an urgent problem to be solved. Summary of the invention
[0003] In view of this, the purpose of the present invention is to provide a tobacco monopoly license approval method, device, equipment and medium, which can extract key information from the document of the tobacco monopoly license through a large model and accurately complete the compliance approval of the document. The specific scheme is as follows:
[0004] In the first aspect, the present application provides a method for approving a tobacco monopoly license, comprising:
[0005] Processing the previously collected historical application data for tobacco monopoly licenses to obtain target training data;
[0006] Determine the learning rate corresponding to each training layer in the large model to be fine-tuned based on each training task corresponding to the large model to be fine-tuned, and generate variant samples corresponding to the target training data based on a preset data enhancement technology;
[0007] Inputting the target training data into the large model to be fine-tuned for model fine-tuning training, and using the variant samples to perform model adversarial training on the large model to be fine-tuned to obtain a target tobacco monopoly license approval model;
[0008] The current application data for the tobacco monopoly license is obtained, and the current application data is approved by using the target tobacco monopoly license approval model to obtain the target tobacco monopoly license approval result.
[0009] Optionally, the step of processing the pre-collected historical application data for tobacco monopoly licenses to obtain target training data includes:
[0010] Collect historical application data for tobacco monopoly licenses, and convert paper document data in the historical application data into electronic document data using optical character recognition technology;
[0011] Based on preset data annotation rules, a data annotation operation is performed on the electronic file data to obtain annotated data, and a data format adjustment is performed on the annotated data to obtain target training data.
[0012] Optionally, determining the learning rate corresponding to each training layer in the large model to be fine-tuned based on each training task corresponding to the large model to be fine-tuned includes:
[0013] The target base model is determined as the large model to be fine-tuned, and the contribution of each training layer to each training task is determined based on the gradient change of each training layer in the large model to be fine-tuned under each training task, and then the learning rate corresponding to each training layer is determined according to the contribution.
[0014] Optionally, after using the variant samples to perform model adversarial training on the large model to be fine-tuned to obtain a target tobacco monopoly license approval model, the method further includes:
[0015] Inputting the model verification data into the target tobacco monopoly license approval model to perform model performance evaluation to obtain a model evaluation result, and determining whether the model evaluation result meets a preset model evaluation standard;
[0016] If the model evaluation result does not meet the preset model evaluation standard, the hyperparameters of the target tobacco monopoly license approval model are trained based on a preset parameter fine-tuning method to obtain a new target tobacco monopoly license approval model.
[0017] Optionally, after obtaining the current application data for the tobacco monopoly license, the step further includes:
[0018] Determine whether the current application data meets the preset data format requirements;
[0019] If the current application data meets the preset data format requirement, a unique application number corresponding to the current application data is generated, and the current application data is backed up at the same time;
[0020] If the current application data does not meet the preset data format requirement, a reminder for supplementary application data is sent to the user terminal through a preset reminder method, so that the user terminal adjusts the data format of the current application data based on the reminder for supplementary application data.
[0021] Optionally, the using the target tobacco monopoly license approval model to approve the current application data to obtain the target tobacco monopoly license approval result includes:
[0022] Preprocessing the current application data to obtain processed data, and extracting corresponding text information from the processed data using optical character recognition technology;
[0023] The text information is input into the target tobacco monopoly license approval model, and the text information is subjected to compliance check through the target tobacco monopoly license approval model, and then the target tobacco monopoly license approval result corresponding to the current application data is determined based on the test result of the compliance check.
[0024] Optionally, performing compliance detection on the text information through the target tobacco monopoly license approval model, and then determining the target tobacco monopoly license approval result corresponding to the current application data based on the detection result of the compliance detection, includes:
[0025] Generate a key-value pair corresponding to the target information in the text information through the target tobacco monopoly license approval model, and compare the key-value pair with the historical data information in the internal database to determine the data accuracy comparison result;
[0026] Perform identity information judgment through the face recognition comparison tool in the target tobacco monopoly license approval model to determine the identity information comparison result;
[0027] Acquiring geographic information provided by a geographic information system through the target tobacco monopoly license approval model, and calculating the distance between the sales point address in the text information and the address of the surrounding facilities based on the geographic information;
[0028] A compliance assessment is performed on the distance to obtain a compliance assessment result, and the target tobacco monopoly license approval result corresponding to the current application data is determined based on the data accuracy comparison result, the identity information comparison result and the compliance assessment result.
[0029] In a second aspect, the present application provides a tobacco monopoly license approval device, comprising:
[0030] A data processing module, used for processing the pre-collected historical application data for tobacco monopoly licenses to obtain target training data;
[0031] An information generation module, used to determine the learning rate corresponding to each training layer in the large model to be fine-tuned based on each training task corresponding to the large model to be fine-tuned, and to generate variant samples corresponding to the target training data based on a preset data enhancement technology;
[0032] A model fine-tuning module, used for inputting the target training data into the large model to be fine-tuned to perform model fine-tuning training, and at the same time using the variant samples to perform model adversarial training on the large model to be fine-tuned to obtain a target tobacco monopoly license approval model;
[0033] The data approval module is used to obtain the current application data for the tobacco monopoly license, and use the target tobacco monopoly license approval model to approve the current application data to obtain the target tobacco monopoly license approval result.
[0034] In a third aspect, the present application provides an electronic device, including:
[0035] Memory, used to store computer programs;
[0036] The processor is used to execute the computer program to implement the aforementioned tobacco monopoly license approval method.
[0037] In a fourth aspect, the present application provides a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, the aforementioned tobacco monopoly license approval method is implemented.
[0038] In this application, firstly, the pre-collected historical application data for tobacco monopoly licenses are processed to obtain target training data; based on the training tasks corresponding to the large model to be fine-tuned, the learning rate corresponding to each training layer in the large model to be fine-tuned is determined, and the variant samples corresponding to the target training data are generated based on the preset data enhancement technology; the target training data is input into the large model to be fine-tuned for model fine-tuning training, and the large model to be fine-tuned is trained with the variant samples to obtain the target tobacco monopoly license approval model; the current application data for the tobacco monopoly license is obtained, and the current application data is approved with the target tobacco monopoly license approval model to obtain the target tobacco monopoly license approval result. As can be seen from the above, before approving the target tobacco monopoly license, this application first collects the historical application data for the tobacco monopoly license, processes the historical application data to obtain the target training data, and then determines the corresponding learning rate according to the training tasks of each layer of the large model to be fine-tuned, and uses the data enhancement technology to process the processed target training data to generate the variant samples corresponding to the target training data. Then, the target training data is used to fine-tune the large model to be fine-tuned, and the variant samples are used to perform adversarial training on the large model to obtain the target tobacco monopoly license approval model. Finally, the target tobacco monopoly license approval model is used to approve the current application data for the tobacco monopoly license sent by the user end to obtain the approval result. In this way, this application uses a dynamic learning rate adjustment strategy and a method combining data augmentation with adversarial training to fine-tune the large model, and realizes the intelligent approval of compliance for tobacco monopoly licenses using a large model. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.
[0040] Figure 1 A flowchart of a tobacco monopoly license approval method disclosed in this application;
[0041] Figure 2 A flowchart of a specific tobacco monopoly license approval method disclosed in this application;
[0042] Figure 3 A schematic diagram of the structure of a tobacco monopoly license approval device disclosed in this application;
[0043] Figure 4This is a structural diagram of an electronic device disclosed in this application. DETAILED DESCRIPTION
[0044] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0045] With the rapid development of information technology, traditional manual approval methods appear to be inefficient in many business scenarios, especially those involving complex document approval processes, such as the approval of tobacco monopoly licenses. Traditional manual approval not only requires a lot of time and manpower costs, but is also prone to unstable approval quality due to subjective factors or irregular operations. In recent years, with the continuous advancement of artificial intelligence technology, especially the rise of large language models, the ability to process information intelligently has been greatly improved. However, due to differences in data characteristics and task requirements for each business scenario, relying solely on pre-trained large models cannot directly meet all actual business needs. Therefore, this application will specifically introduce a method for approving tobacco monopoly licenses that can solve the above problems.
[0046] See also Figure 1 As shown, the embodiment of the present invention discloses a method for approving a tobacco monopoly license, which may include:
[0047] Step S11: Process the pre-collected historical application data for tobacco monopoly licenses to obtain target training data.
[0048] In this embodiment, historical application data for tobacco monopoly licenses are collected, and optical character recognition technology is used to convert paper document data in the historical application data into electronic document data. Specifically, optical character recognition technology is used to scan the paper document data in the historical application data, and then the scanned text content is converted into electronic document data. It should be pointed out that the open source PaddleOCR (Paddle Optical Character Recognition, i.e., optical character recognition technology based on Baidu Feijiang) can be used to convert the scanned image data into electronic document data. In order to ensure the integrity and accuracy of the electronic file data, this embodiment will mark unrecognizable special characters and data with abnormal formats, where the marking method can be manual marking.
[0049] In this embodiment, the electronic file data is annotated based on preset data annotation rules to obtain annotated data, and the data format of the annotated data is adjusted to obtain target training data. In a specific implementation, the annotation information annotated by the user end for the historical application data is obtained through a preset information port. The annotation information may include the applicant's name, ID number, address, sales point information, etc. The sales point information includes but is not limited to sales point location information and sales point surrounding environment information. The sales point information is evaluated for compliance and each application document data in the historical application data is scored, and the deduction points and reasons for the deductions are marked.
[0050] In this embodiment, the annotated historical application data is sorted and converted into an input format suitable for the large model to be fine-tuned. Specifically, the manually marked historical application data is first formatted into json. The three keys are Instruct, Input, and Output. Instruct represents the task instruction of the system, Input is the input information, and Output is the output of the expected large model. For example:
[0051] "Instruct":"Perform key information extraction tasks",
[0052] "Input":"Applicant Wang Xiaoming was born in June 1997...",
[0053] "Output":{"Name":"Wang Xiaoming","Birthdate":"1997.06"}
[0054] Conduct a compliance assessment of the point of sale information based on the point of sale location information and the surrounding environment information of the point of sale, input about 5 pieces of relevant information, and output the assessment results.
[0055] "Instruct":"Perform point of sale compliance assessment tasks",
[0056] "Input":"1. Location: xx community, Jinan City 2. Surrounding schools: xxxx 3. Surrounding license procedures: xxxx",
[0057] "Output":{"result":"Compliance"}
[0058] Compare the extracted key information, point of sale information, etc., and score the application document data in the historical application data.
[0059] "Instruct":"Execute the audit scoring task",
[0060] "Input":"1. Face matching: consistent 2. Business license information: inconsistent 3. Whether the point of sale is compliant: not compliant",
[0061] "Output":{"score":"60","reason":"1. Business license information is inconsistent 2. Sales location is not compliant"}
[0062] In this embodiment, the historical application data after sorting and transformation is divided into target training data and target verification data, and the division ratio can be set to 9:1. The target training data is used for subsequent training of the large model to be fine-tuned, and the target verification data is used for performance verification of the trained large model to be fine-tuned. It should be noted that in the process of dividing the historical application data after sorting and transformation, it should be ensured that the target training data and the target verification data are consistent in data distribution to avoid overfitting. At the same time, it is necessary to ensure that the historical application data after sorting and transformation is sufficient so that the large model to be fine-tuned can learn enough information during the training process, so as to ensure that accurate verification results are obtained in the subsequent verification process of the large model to be fine-tuned.
[0063] Step S12: determining the learning rate corresponding to each training layer in the large model to be fine-tuned based on each training task corresponding to the large model to be fine-tuned, and generating variant samples corresponding to the target training data based on a preset data enhancement technology.
[0064] In this embodiment, the target base model is determined as the large model to be fine-tuned, and the contribution of each training layer to each training task is determined based on the gradient change of each training layer in the large model to be fine-tuned under each training task, and then the learning rate corresponding to each training layer is determined according to the contribution. It should be noted that in this embodiment, the chat model (i.e., a large language model) of the open source Qwen or Baichuan series with outstanding Chinese capabilities is selected as the base model for fine-tuning. This embodiment provides a method based on dynamic weight adjustment, which can automatically adjust the learning rate of each layer according to the degree of dependence of different tasks such as key information extraction, compliance assessment, and audit scoring on the large model to be fine-tuned. Specifically, the shallow part of the large model to be fine-tuned focuses more on general feature extraction, while the deep layer is used for optimization of specific tasks. When fine-tuning the large model to be fine-tuned, the performance of the large model to be fine-tuned under different tasks can be improved by setting the dynamic learning rate of each layer.
[0065] Specifically, we first analyze the gradient changes of each training layer in the large model to be fine-tuned under each training task, determine the contribution of each training layer to each training task, and then dynamically adjust the learning rate according to the contribution. For example, the learning rate of the shallow layer is , the depth is , the middle layer can be realized by linear interpolation, wherein the linear interpolation algorithm is as follows:
[0066] ;
[0067] Among them, l is the index of the layer, L is the total number of layers, is the learning rate of the lth layer, is the shallow learning rate, is the deep learning rate, is the learning rate of the current layer. The method based on dynamic weight adjustment can refine the parameter update process of the large model to be fine-tuned without significantly increasing the computational burden, so that the large model to be fine-tuned can perform better on different training tasks and avoid overfitting or underfitting. At the same time, it can also speed up the convergence speed and reduce the training time.
[0068] In this embodiment, in order to deal with the data imbalance problem in the tobacco monopoly license application data, a fine-tuning strategy combining data enhancement and adversarial training is introduced. By generating adversarial samples during the fine-tuning process, the large model to be fine-tuned can still maintain high stability and accuracy when facing abnormal data. Specifically, the data enhancement part of the data can generate variant samples based on semantic replacement of text, synonym replacement, or calling the commercial large model API (Application Programming Interface).
[0069] Step S13: input the target training data into the large model to be fine-tuned for model fine-tuning training, and simultaneously use the variant samples to perform model adversarial training on the large model to be fine-tuned to obtain a target tobacco monopoly license approval model.
[0070] In this embodiment, the variant samples are used to perform model adversarial training on the large model to be fine-tuned. It can be understood that adversarial training generates slightly disturbed data to ensure that the model exhibits stable prediction results for the disturbed data. The specific algorithm of adversarial training is as follows:
[0071] ;
[0072] in, Perturbation data, x is the variant sample, is the perturbation intensity, sign(⋅) is the sign function, which returns the sign of the input. It means to find the gradient of x. is the loss function, is the output of the large model to be fine-tuned.
[0073] In this embodiment, in order to ensure that the target tobacco monopoly license approval model can achieve the expected accuracy in each task, after the variant sample is used to perform model adversarial training on the large model to be fine-tuned to obtain the target tobacco monopoly license approval model, it also includes: inputting the model verification data into the target tobacco monopoly license approval model to perform model performance evaluation to obtain a model evaluation result, and judging whether the model evaluation result meets the preset model evaluation standard; if the model evaluation result does not meet the preset model evaluation standard, the hyperparameters of the target tobacco monopoly license approval model are trained based on the preset parameter fine-tuning method to obtain a new target tobacco monopoly license approval model. It can be understood that this embodiment uses the target verification data to perform performance verification on the target tobacco monopoly license approval model. The target verification data includes three tasks: key information extraction, point of sale information compliance evaluation, and information comparison. The indicators for evaluating the target tobacco monopoly license approval model include accuracy, recall, and F1Score (F1 score, i.e., harmonic mean). If the evaluation result of the target tobacco monopoly license approval model does not meet the preset model evaluation criteria, the target tobacco monopoly license approval model can be retrained by improving the quality of target training data, increasing the number of samples, or using efficient parameter fine-tuning methods such as LoRA (Low-Rank Adaptation), and adjusting hyperparameters such as learning rate and batch size until the performance of the target tobacco monopoly license approval model meets the evaluation criteria.
[0074] Step S14: obtaining current application data for a tobacco monopoly license, and using the target tobacco monopoly license approval model to approve the current application data to obtain an approval result of the target tobacco monopoly license.
[0075] In this embodiment, the current application data for the tobacco monopoly license submitted by the user is first obtained, and the current application data includes the user's personal ID card, photo, tobacco monopoly license application form, etc., and then the current application data is preprocessed to obtain processed data, and the key information of the preprocessed data is input into the target tobacco monopoly license approval model, and the target tobacco monopoly license approval model is used for subsequent approval to obtain the target tobacco monopoly license approval result.
[0076] In this embodiment, in order to ensure the smooth progress of the approval process, after obtaining the current application data for the tobacco monopoly license, it also includes: determining whether the current application data meets the preset data format requirements; if the current application data meets the preset data format requirements, then generating a unique application number corresponding to the current application data, and backing up the current application data at the same time; if the current application data does not meet the preset data format requirements, then sending an application data supplementary submission prompt to the user terminal through a preset prompt method, so that the user terminal adjusts the data format of the current application data based on the application data supplementary submission prompt. Specifically, this embodiment performs a format check on the current application data to ensure that the type and size of the current application data meet the preset data format requirements. In specific application scenarios, regular expressions can be used to perform format checks on the current application data, or third-party APIs can be used to perform format checks on the current application data. Next, a unique application number is assigned to the current application data that meets the preset data format requirements, so as to track the processing progress and status of each current application data. At the same time, the current application data is backed up to obtain backup data. When the current application data is lost or damaged, the backup data can restore the normal operation of the system and ensure data integrity. It should be pointed out that the reasons for the loss or damage of the current application data include but are not limited to hardware failure, software vulnerabilities, human errors, etc. For the current application data that does not meet the preset data format requirements, a reminder for supplementary application data is sent to the user terminal through a preset prompt method. For example, a dialog box pops up on the user terminal interface to remind the user to supplement the application data. The dialog box can clearly inform the user of the data with format problems in the current application data and remind the user to adjust the current application data.
[0077] As can be seen from the above, in this application, the historical application data for tobacco monopoly licenses collected in advance are first processed to obtain target training data; the learning rate corresponding to each training layer in the large model to be fine-tuned is determined based on each training task corresponding to the large model to be fine-tuned, and variant samples corresponding to the target training data are generated based on the preset data enhancement technology; the target training data is input into the large model to be fine-tuned for model fine-tuning training, and the large model to be fine-tuned is trained with the variant samples to obtain the target tobacco monopoly license approval model; the current application data for the tobacco monopoly license is obtained, and the current application data is approved by the target tobacco monopoly license approval model to obtain the target tobacco monopoly license approval result. As can be seen from the above, before approving the target tobacco monopoly license, the application first collects the historical application data for the tobacco monopoly license, processes the historical application data to obtain the target training data, and then determines the corresponding learning rate according to the training tasks of each layer of the large model to be fine-tuned, and uses the data enhancement technology to process the processed target training data to generate variant samples corresponding to the target training data. Then, the target training data is used to fine-tune the large model to be fine-tuned, and the variant samples are used to perform adversarial training on the large model to obtain the target tobacco monopoly license approval model. Finally, the target tobacco monopoly license approval model is used to approve the current application data for the tobacco monopoly license sent by the user end to obtain the approval result. In this way, this application uses a dynamic learning rate adjustment strategy and a method combining data augmentation with adversarial training to fine-tune the large model, and realizes the intelligent approval of compliance for tobacco monopoly licenses using a large model.
[0078] See also Figure 2 As shown, in order to ensure the accuracy and completeness of the approval process, the present application embodiment further provides a specific tobacco monopoly license approval method, which may include:
[0079] Step S21 : pre-processing the current application data to obtain processed data, and extracting corresponding text information from the processed data using optical character recognition technology.
[0080] In this embodiment, the current application data obtained for the tobacco monopoly license is preprocessed, and the preprocessing process includes but is not limited to image denoising, edge detection and other operations. Then, optical character recognition technology is used to extract text information in the current application data after preprocessing as input to the target tobacco monopoly license approval model.
[0081] Step S22: input the text information into the target tobacco monopoly license approval model, and generate key-value pairs corresponding to the target information in the text information through the target tobacco monopoly license approval model, and compare the key-value pairs with the historical data information in the internal database to determine the data accuracy comparison result.
[0082] In this embodiment, the text information in the current application data after preprocessing is input into the target tobacco monopoly license approval model, and the target tobacco monopoly license approval model will automatically identify the key text information in the current application data after preprocessing, such as name, ID number, sales point address, etc. The target tobacco monopoly license approval model generates key-value pairs of key information from the text information in the current application data after preprocessing according to the pattern learned during the fine-tuning of the above model, and then compares the key-value pairs with the historical data information in the internal database to determine the data accuracy. The comparison task detects the matching degree of multiple fields and returns the data accuracy comparison result.
[0083] Step S23: performing identity information judgment through the face recognition comparison tool in the target tobacco monopoly license approval model to determine the identity information comparison result.
[0084] In this embodiment, the identity information of the applicant is confirmed through the high-precision face recognition comparison tool integrated in the target tobacco monopoly license approval model, and the identity information comparison result is returned.
[0085] Step S24: acquiring geographic information provided by a geographic information system through the target tobacco monopoly license approval model, calculating the distance between the sales point address and the surrounding facility address in the text information based on the geographic information, and performing compliance evaluation on the distance to obtain a compliance evaluation result.
[0086] In this embodiment, the target tobacco monopoly license approval model analyzes the distance between the sales point and surrounding schools and businesses based on the geographic information provided by the geographic information system. It should be noted that the compliance assessment rules can be based on regulations. For example, the distance between the sales point and the school is not less than 200 meters. The target tobacco monopoly license approval model calculates the actual distance and outputs an assessment result of whether it is compliant. It can be understood that if the actual distance between the sales point and the school is less than 200 meters, an assessment result of non-compliance is output and the detailed reasons for non-compliance are sent to the user end.
[0087] Step S25: determining the approval result of the target tobacco monopoly license corresponding to the current application data based on the data accuracy comparison result, the identity information comparison result and the compliance assessment result.
[0088] In this embodiment, the target tobacco monopoly license approval model generates a final score for each current application data file based on the aforementioned comparison results and compliance assessment information. The scoring criteria can comprehensively consider multiple factors such as the above-mentioned data accuracy, identity information accuracy, compliance, etc., and then output a complete evaluation report. The evaluation report includes a score and detailed approval reasons in order to give a clear target tobacco monopoly license approval result.
[0089] As can be seen from the above, in this embodiment, the current application data is first preprocessed to obtain processed data, and the corresponding text information is extracted from the processed data using optical character recognition technology; the text information is input into the target tobacco monopoly license approval model, and the key-value pairs corresponding to the target information in the text information are generated through the target tobacco monopoly license approval model, and the key-value pairs are compared with the historical data information in the internal database to determine the data accuracy comparison result; the identity information is judged by the face recognition comparison tool in the target tobacco monopoly license approval model to determine the identity information comparison result; the geographic information provided by the geographic information system is obtained through the target tobacco monopoly license approval model, and the distance between the sales point address and the surrounding facility address in the text information is calculated based on the geographic information, and the compliance assessment is performed on the distance to obtain a compliance assessment result; the target tobacco monopoly license approval result corresponding to the current application data is determined based on the data accuracy comparison result, the identity information comparison result and the compliance assessment result. As can be seen from the above, this embodiment first pre-processes the acquired current application data, and extracts text from the pre-processed current application data, and then inputs the extracted text information into the target tobacco monopoly license approval model, and uses the target tobacco monopoly license approval model to compare the text information with the historical data information to determine the data accuracy comparison result, and then uses the face recognition comparison tool in the target tobacco monopoly license approval model to judge the identity information of the applicant, and uses the target tobacco monopoly license approval model to obtain the geographic information provided by the geographic information system to calculate the distance between the sales point address and the surrounding facility address to obtain the compliance assessment result. In this way, the target tobacco monopoly license approval result corresponding to the current application data is determined through the data accuracy comparison result, the identity information comparison result and the compliance assessment result, which greatly improves the accuracy and completeness of the approval process.
[0090] Accordingly, see Figure 3 As shown, the embodiment of the present application also provides a tobacco monopoly license approval device, which may include:
[0091] The data processing module 11 is used to process the pre-collected historical application data for tobacco monopoly licenses to obtain target training data;
[0092] An information generation module 12 is used to determine the learning rate corresponding to each training layer in the large model to be fine-tuned based on each training task corresponding to the large model to be fine-tuned, and to generate variant samples corresponding to the target training data based on a preset data enhancement technology;
[0093] A model fine-tuning module 13 is used to input the target training data into the large model to be fine-tuned for model fine-tuning training, and simultaneously use the variant samples to perform model adversarial training on the large model to be fine-tuned to obtain a target tobacco monopoly license approval model;
[0094] The data approval module 14 is used to obtain current application data for a tobacco monopoly license, and to approve the current application data using the target tobacco monopoly license approval model to obtain a target tobacco monopoly license approval result.
[0095] As can be seen from the above, in this application, the historical application data for tobacco monopoly licenses collected in advance are first processed to obtain target training data; the learning rate corresponding to each training layer in the large model to be fine-tuned is determined based on each training task corresponding to the large model to be fine-tuned, and variant samples corresponding to the target training data are generated based on the preset data enhancement technology; the target training data is input into the large model to be fine-tuned for model fine-tuning training, and the large model to be fine-tuned is trained with the variant samples to obtain the target tobacco monopoly license approval model; the current application data for the tobacco monopoly license is obtained, and the current application data is approved by the target tobacco monopoly license approval model to obtain the target tobacco monopoly license approval result. As can be seen from the above, before approving the target tobacco monopoly license, the application first collects the historical application data for the tobacco monopoly license, processes the historical application data to obtain the target training data, and then determines the corresponding learning rate according to the training tasks of each layer of the large model to be fine-tuned, and uses the data enhancement technology to process the processed target training data to generate variant samples corresponding to the target training data. Then, the target training data is used to fine-tune the large model to be fine-tuned, and the variant samples are used to perform adversarial training on the large model to obtain the target tobacco monopoly license approval model. Finally, the target tobacco monopoly license approval model is used to approve the current application data for the tobacco monopoly license sent by the user end to obtain the approval result. In this way, this application uses a dynamic learning rate adjustment strategy and a method combining data augmentation with adversarial training to fine-tune the large model, and realizes the intelligent approval of compliance for tobacco monopoly licenses using a large model.
[0096] In some specific implementations, the data processing module 11 includes:
[0097] A data collection unit, used to collect historical application data for tobacco monopoly licenses, and convert paper document data in the historical application data into electronic document data using optical character recognition technology;
[0098] The data annotation unit is used to perform a data annotation operation on the electronic file data based on a preset data annotation rule to obtain annotated data, and to adjust the data format of the annotated data to obtain target training data.
[0099] In some specific implementations, the information generation module 12 includes:
[0100] A learning rate determination unit is used to determine the target base model as the large model to be fine-tuned, and determine the contribution of each training layer to each training task based on the gradient change of each training layer in the large model to be fine-tuned under each training task, and then determine the learning rate corresponding to each training layer according to the contribution.
[0101] In some specific implementations, the model fine-tuning module 13 further includes:
[0102] A model evaluation unit, used to input the model verification data into the target tobacco monopoly license approval model to perform model performance evaluation to obtain a model evaluation result, and determine whether the model evaluation result meets a preset model evaluation standard;
[0103] A model training unit is used to train the hyperparameters of the target tobacco monopoly license approval model based on a preset parameter fine-tuning method to obtain a new target tobacco monopoly license approval model if the model evaluation result does not meet the preset model evaluation standard.
[0104] In some specific implementations, the data approval module 14 further includes:
[0105] A format determination unit, used to determine whether the current application data meets the preset data format requirements;
[0106] A data backup unit, configured to generate a unique application number corresponding to the current application data and to back up the current application data if the current application data meets the preset data format requirement;
[0107] The prompt sending unit is used to send an application data supplementary submission prompt to the user terminal through a preset prompt method if the current application data does not meet the preset data format requirement, so that the user terminal adjusts the data format of the current application data based on the application data supplementary submission prompt.
[0108] In some specific implementations, the data approval module 14 includes:
[0109] An information extraction unit, used for preprocessing the current application data to obtain processed data, and extracting corresponding text information from the processed data using optical character recognition technology;
[0110] A data approval unit is used to input the text information into the target tobacco monopoly license approval model, and perform compliance check on the text information through the target tobacco monopoly license approval model, and then determine the target tobacco monopoly license approval result corresponding to the current application data based on the test result of the compliance check.
[0111] In some specific implementations, the data approval module 14 includes:
[0112] An accuracy comparison unit, used to generate a key-value pair corresponding to the target information in the text information through the target tobacco monopoly license approval model, and compare the key-value pair with the historical data information in the internal database to determine the data accuracy comparison result;
[0113] An identity information comparison unit, used to determine the identity information comparison result by performing identity information judgment through the face recognition comparison tool in the target tobacco monopoly license approval model;
[0114] A distance calculation unit, used to obtain geographic information provided by a geographic information system through the target tobacco monopoly license approval model, and calculate the distance between the sales point address in the text information and the address of the surrounding facilities based on the geographic information;
[0115] An approval result determination unit is used to perform a compliance assessment on the distance to obtain a compliance assessment result, and determine the approval result of the target tobacco monopoly license corresponding to the current application data based on the data accuracy comparison result, the identity information comparison result and the compliance assessment result.
[0116] Furthermore, the present application also discloses an electronic device. Figure 4 This is a structural diagram of an electronic device 20 according to an exemplary embodiment, and the content in the figure cannot be regarded as any limitation on the scope of use of this application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 is used to store a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the approval method of the tobacco monopoly license disclosed in any of the aforementioned embodiments. In addition, the electronic device 20 in this embodiment may specifically be an electronic computer.
[0117] In this embodiment, the power supply 23 is used to provide working voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and the external device, and the communication protocol it follows is any communication protocol that can be applied to the technical solution of the present application, and is not specifically limited here; the input and output interface 25 is used to obtain external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs and is not specifically limited here.
[0118] In addition, the memory 22, as a carrier for storing resources, can be a read-only memory, a random access memory, a disk or an optical disk, etc. The resources stored thereon can include an operating system 221, a computer program 222, etc., and the storage method can be temporary storage or permanent storage.
[0119] The operating system 221 is used to manage and control the hardware devices and computer program 222 on the electronic device 20, which can be Windows Server, Netware, Unix, Linux, etc. In addition to the computer program that can be used to complete the tobacco monopoly license approval method performed by the electronic device 20 disclosed in any of the aforementioned embodiments, the computer program 222 can further include computer programs that can be used to complete other specific tasks.
[0120] Furthermore, the present application also discloses a computer-readable storage medium for storing a computer program; wherein the computer program, when executed by a processor, implements the aforementioned disclosed method for approving a tobacco monopoly license. The specific steps of the method can be referred to the corresponding contents disclosed in the aforementioned embodiments, and will not be repeated here.
[0121] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.
[0122] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.
[0123] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.
[0124] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.
[0125] The technical solution provided by the present application is introduced in detail above. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea. At the same time, for general technicians in this field, according to the idea of the present application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.
Claims
1. A method for approving a tobacco monopoly license, characterized in that: include: Processing the previously collected historical application data for tobacco monopoly licenses to obtain target training data; Determine the learning rate corresponding to each training layer in the large model to be fine-tuned based on each training task corresponding to the large model to be fine-tuned, and generate variant samples corresponding to the target training data based on a preset data enhancement technology; Inputting the target training data into the large model to be fine-tuned for model fine-tuning training, and using the variant samples to perform model adversarial training on the large model to be fine-tuned to obtain a target tobacco monopoly license approval model; The current application data for the tobacco monopoly license is obtained, and the current application data is approved by using the target tobacco monopoly license approval model to obtain the target tobacco monopoly license approval result.
2. The tobacco monopoly license approval method according to claim 1, characterized in that: The data processing of the pre-collected historical application data for tobacco monopoly licenses to obtain target training data includes: Collect historical application data for tobacco monopoly licenses, and convert paper document data in the historical application data into electronic document data using optical character recognition technology; Based on preset data annotation rules, a data annotation operation is performed on the electronic file data to obtain annotated data, and a data format adjustment is performed on the annotated data to obtain target training data.
3. The tobacco monopoly license approval method according to claim 1, characterized in that: The determining, based on each training task corresponding to the large model to be fine-tuned, the learning rate corresponding to each training layer in the large model to be fine-tuned includes: The target base model is determined as the large model to be fine-tuned, and the contribution of each training layer to each training task is determined based on the gradient change of each training layer in the large model to be fine-tuned under each training task, and then the learning rate corresponding to each training layer is determined according to the contribution.
4. The method for approving a tobacco monopoly license according to claim 1, characterized in that: After the variant samples are used to perform model adversarial training on the large model to be fine-tuned to obtain the target tobacco monopoly license approval model, the method further includes: Inputting the model verification data into the target tobacco monopoly license approval model to perform model performance evaluation to obtain a model evaluation result, and determining whether the model evaluation result meets a preset model evaluation standard; If the model evaluation result does not meet the preset model evaluation standard, the hyperparameters of the target tobacco monopoly license approval model are trained based on a preset parameter fine-tuning method to obtain a new target tobacco monopoly license approval model.
5. The tobacco monopoly license approval method according to claim 1, characterized in that: After obtaining the current application data for the tobacco monopoly license, the method further includes: Determine whether the current application data meets the preset data format requirements; If the current application data meets the preset data format requirement, a unique application number corresponding to the current application data is generated, and the current application data is backed up at the same time; If the current application data does not meet the preset data format requirement, a reminder for supplementary application data is sent to the user terminal through a preset reminder method, so that the user terminal adjusts the data format of the current application data based on the reminder for supplementary application data.
6. The method for approving a tobacco monopoly license according to any one of claims 1 to 5, characterized in that: The step of using the target tobacco monopoly license approval model to approve the current application data to obtain the target tobacco monopoly license approval result includes: Preprocessing the current application data to obtain processed data, and extracting corresponding text information from the processed data using optical character recognition technology; The text information is input into the target tobacco monopoly license approval model, and the text information is subjected to compliance check through the target tobacco monopoly license approval model, and then the target tobacco monopoly license approval result corresponding to the current application data is determined based on the test result of the compliance check.
7. The tobacco monopoly license approval method according to claim 6, characterized in that: The step of performing compliance detection on the text information through the target tobacco monopoly license approval model, and then determining the target tobacco monopoly license approval result corresponding to the current application data based on the detection result of the compliance detection, includes: Generate a key-value pair corresponding to the target information in the text information through the target tobacco monopoly license approval model, and compare the key-value pair with the historical data information in the internal database to determine the data accuracy comparison result; Perform identity information judgment through the face recognition comparison tool in the target tobacco monopoly license approval model to determine the identity information comparison result; Acquiring geographic information provided by a geographic information system through the target tobacco monopoly license approval model, and calculating the distance between the sales point address in the text information and the address of the surrounding facilities based on the geographic information; A compliance assessment is performed on the distance to obtain a compliance assessment result, and the target tobacco monopoly license approval result corresponding to the current application data is determined based on the data accuracy comparison result, the identity information comparison result and the compliance assessment result.
8. A tobacco monopoly license approval device, characterized in that: include: A data processing module, used for processing the pre-collected historical application data for tobacco monopoly licenses to obtain target training data; An information generation module, used to determine the learning rate corresponding to each training layer in the large model to be fine-tuned based on each training task corresponding to the large model to be fine-tuned, and to generate variant samples corresponding to the target training data based on a preset data enhancement technology; A model fine-tuning module, used for inputting the target training data into the large model to be fine-tuned to perform model fine-tuning training, and at the same time using the variant samples to perform model adversarial training on the large model to be fine-tuned to obtain a target tobacco monopoly license approval model; The data approval module is used to obtain the current application data for the tobacco monopoly license, and use the target tobacco monopoly license approval model to approve the current application data to obtain the target tobacco monopoly license approval result.
9. An electronic device, characterized in that: include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the tobacco monopoly license approval method as described in any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that: Used to store a computer program; wherein, when the computer program is executed by a processor, the method for approving a tobacco monopoly license as described in any one of claims 1 to 7 is implemented.