Large model examination and approval method for government affair industry and related equipment
By using large language models and machine learning models in the government approval process, the approval results are automatically analyzed and predicted, and the problem of inefficiency of the traditional approval process is solved, and efficient and accurate approval decisions are achieved.
Patent Information
- Application Number
- CN202510028829.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-08
- Publication Date
- 2025-04-15
AI Technical Summary
The traditional government approval process relies on manual operations, is inefficient and prone to errors, and is difficult to meet the modern society's demand for fast and accurate services, while extracting valuable information from massive data.
A large language model is used to extract feature data from the approval documents, combine the condition matching engine and a real-time optimized machine learning model, automatically analyze and predict approval results, reduce manual intervention and improve decision-making speed.
It realizes automated approval, reduces the probability of errors, improves the quality and speed of decisions, and can extract valuable information from massive data more efficiently.
Smart Images
Figure CN119829919A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of machine learning technology, and in particular to a large model approval method and related equipment for the government affairs industry. Background Art
[0002] With the rapid development of information technology, the government affairs industry is facing increasing data processing needs and high requirements for service efficiency.
[0003] The traditional government approval process relies on manual operations, which is not only inefficient but also prone to errors, making it difficult to adapt to the modern society's demand for fast and accurate services. In addition, with the explosive growth of data volume, how to extract valuable information from massive data and improve the quality and speed of decision-making has become an urgent problem to be solved in the government industry. Summary of the invention
[0004] In view of this, an embodiment of the present invention provides a large model approval method and related equipment for the government affairs industry to achieve automated approval to reduce the probability of errors and improve the quality and speed of decision-making.
[0005] To achieve the above objectives, the embodiments of the present invention provide the following technical solutions:
[0006] A first aspect of an embodiment of the present invention discloses a large model approval method for a government affairs industry, the method comprising:
[0007] Obtain approval documents;
[0008] Extracting feature data from the approval document using a pre-trained large language model;
[0009] Using a condition matching engine, analyzing whether the characteristic data matches a preset approval condition;
[0010] If so, determining that the approval document has passed the approval;
[0011] If not, the feature data is input into a real-time optimized machine learning model for prediction to obtain the confidence level of the approval document;
[0012] If the confidence level is greater than or equal to the confidence level threshold, determining that the approval document has passed the approval;
[0013] If the confidence level is less than the confidence level threshold, an audit notification is generated to remind the auditor, and when a pass instruction input by the auditor is received, it is determined that the approval document has passed the approval; when a reject instruction input by the auditor is received, it is determined that the approval document has failed the approval.
[0014] Preferably, the extracting feature data from the approval document using a pre-trained large language model includes:
[0015] Detecting the structural type of the approval document;
[0016] If the document to be approved is a structured document, feature data is extracted from the approval document using a gradient boosting decision tree algorithm;
[0017] If the document to be approved is an unstructured document, the approval document is input into a pre-trained large language model for semantic analysis to obtain feature data output by the large language model;
[0018] If the document to be approved is an unstructured document, the unstructured part of the approval document is input into the large language model for semantic analysis to obtain a first part of feature data output by the large language model; the second part of feature data is extracted from the structured part of the approval document using the gradient boosting decision tree algorithm; the first part of feature data and the second part of feature data are combined to obtain feature data of the approval document.
[0019] Preferably, after extracting feature data from the approval document using the pre-trained large language model, the method further comprises:
[0020] Performing a feature integrity check on the feature data;
[0021] If the characteristic data is incomplete, a supplementary information request is generated and the process returns to execute the step of obtaining the approval document; the supplementary information request is used to prompt the approval requester to supplement the information and re-enter the approval document.
[0022] Preferably, the training process of the large language model includes:
[0023] Build a large language model to be trained;
[0024] Acquire multiple historical approval documents, and perform data cleaning on each of the historical approval documents;
[0025] Dividing each of the historical approval documents after data cleaning into a training set and a validation set;
[0026] Using the training set to train the large language model to be trained to obtain a trained large language model;
[0027] Using the validation set to validate the performance of the trained large language model;
[0028] If the verification passes, the trained large language model is obtained.
[0029] Preferably, the process of optimizing the machine learning model in real time includes:
[0030] Using data collection tools, raw data is collected in real time; the raw data includes: the target approval document that has completed the approval, the approval result of the target approval document, the analysis information obtained by analyzing the feedback information input by the approval requester, and the log data used to characterize the operation record of the approval process;
[0031] Preprocessing the original data to obtain target feature data;
[0032] The target feature data is input into the machine learning model for retraining to obtain the machine learning model after real-time optimization.
[0033] Preferably, the preprocessing of the raw data to obtain target feature data includes:
[0034] Performing data cleaning on each of the original data;
[0035] Determine whether the original data after each data cleaning is qualified;
[0036] For each qualified raw data, extract feature data for training using a feature extraction tool;
[0037] For each unqualified raw data, receiving a manually input correction instruction, correcting the unqualified raw data based on the correction instruction, and extracting feature data for training from the corrected raw data using the feature extraction tool;
[0038] The feature data used for training is subjected to data standardization processing to obtain target feature data that conforms to the input format of the machine learning model.
[0039] Preferably, the step of inputting the target feature data into the machine learning model for retraining to obtain the machine learning model after real-time optimization includes:
[0040] Inputting the target feature data into the machine learning model for retraining;
[0041] evaluating whether the performance of the machine learning model after retraining is improved;
[0042] If so, the machine learning model is optimized in real time;
[0043] If not, modify the model parameters of the machine learning model, and return to the step of inputting the target feature data into the machine learning model for retraining until the machine learning model is optimized in real time.
[0044] A second aspect of an embodiment of the present invention discloses a large model approval device for a government affairs industry, the device comprising:
[0045] An acquisition unit, used to acquire approval documents;
[0046] An extraction unit, used to extract feature data from the approval document using a pre-trained large language model;
[0047] The approval unit is used to use a condition matching engine to analyze whether the feature data matches the preset approval conditions; if so, determine that the approval document has passed the approval; if not, input the feature data into a real-time optimized machine learning model for prediction to obtain the confidence of the approval document; if the confidence is greater than or equal to a confidence threshold, determine that the approval document has passed the approval; if the confidence is less than the confidence threshold, generate an audit notification for reminding the auditor, and when receiving a pass instruction input by the auditor, determine that the approval document has passed the approval, and when receiving a reject instruction input by the auditor, determine that the approval document has not passed the approval.
[0048] A third aspect of an embodiment of the present invention discloses a storage medium for storing a computer program. When the computer program is executed, it is specifically used to implement the large model approval method for the government affairs industry described in any one of the first aspects of the embodiments of the present invention.
[0049] A fourth aspect of an embodiment of the present invention discloses an electronic device, including: a memory and a processor;
[0050] The memory is used to store computer programs;
[0051] The processor is used to execute the computer program, specifically to implement the large model approval method for the government affairs industry as described in any one of the first aspects of the embodiments of the present invention.
[0052] Based on the above-mentioned embodiment of the present invention, a large model approval method and related equipment for the government affairs industry are provided to obtain approval documents; a pre-trained large language model is used to extract feature data from the approval documents; a condition matching engine is used to analyze whether the feature data matches the preset approval conditions; if so, it is determined that the approval document has passed the approval; if not, the feature data is input into a real-time optimized machine learning model for prediction to obtain the confidence of the approval document; if the confidence is greater than or equal to the confidence threshold, it is determined that the approval document has passed the approval; if the confidence is less than the confidence threshold, an audit notification is generated to remind the auditor, and when the approval instruction input by the auditor is received, it is determined that the approval document has passed the approval, and when the rejection instruction input by the auditor is received, it is determined that the approval document has not passed the approval. In this solution, a large language model is used to accurately understand the content and context of the approval document, to achieve automated approval and reduce the probability of errors, and to use a continuously optimized machine learning model to improve the quality and speed of decision-making. BRIEF DESCRIPTION OF THE DRAWINGS
[0053] 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.
[0054] Figure 1 A flowchart of a large model approval method for the government affairs industry disclosed in an embodiment of the present invention;
[0055] Figure 2 A flowchart of an approval decision disclosed in an embodiment of the present invention;
[0056] Figure 3 A flowchart of data collection and preprocessing disclosed in an embodiment of the present invention;
[0057] Figure 4 A flowchart of a real-time optimized machine learning model disclosed in an embodiment of the present invention;
[0058] Figure 5 A structural diagram of a large model approval device for the government affairs industry disclosed in an embodiment of the present invention;
[0059] Figure 6 The present invention is a structural diagram of an electronic device disclosed in an embodiment of the present invention. DETAILED DESCRIPTION
[0060] 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.
[0061] In this application, the terms "comprises", "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising 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, an element defined by the sentence "comprising a ..." does not exclude the presence of other identical elements in the process, method, article or device comprising the element.
[0062] As can be seen from the background technology, the traditional government approval process relies on manual operation, which is not only inefficient but also prone to errors, and it is difficult to adapt to the modern society's demand for fast and accurate services. In addition, with the explosive growth of data volume, how to extract valuable information from massive data and improve the quality and speed of decision-making has become a problem that the government industry needs to solve urgently.
[0063] Therefore, an embodiment of the present invention discloses a large model approval method and related equipment for the government affairs industry. In this solution, a large language model is used to accurately understand the content and context of the approval document, to achieve automated approval and reduce the probability of errors, and to use a continuously optimized machine learning model to improve the quality and speed of decision-making.
[0064] like Figure 1 FIG. 1 is a flowchart of a large model approval method for a government affairs industry disclosed in an embodiment of the present invention, comprising the following steps:
[0065] Step S101: Obtain approval documents.
[0066] Among them, approval documents refer to documents that need to be reviewed and decided through the approval process.
[0067] The structural types of approval documents include: structured documents, unstructured documents and semi-structured documents. Structured documents refer to data documents with fixed formats and structures, such as spreadsheets. The data in such documents are usually organized in a table form, and each field has a clear data type and format. Unstructured documents are data documents that do not have a fixed format and structure. Semi-structured documents contain both structured and unstructured parts.
[0068] In one embodiment, approval documents of different structural types are collected through corresponding data collection tools.
[0069] Data collection tools include: Apache Kafka, RabbitMQ, and Apache NiFi.
[0070] Apache Kafka is an open source distributed stream processing platform; RabbitMQ is an open source message broker software; Apache NiFi is an open source data integration and data flow management tool designed to automate the creation, monitoring, and management of data flows.
[0071] Step S102: extract feature data from the approval document using a pre-trained large language model.
[0072] In step S102, the large language model may be BERT or RoBERTa, and feature data is extracted by using a named entity recognition technology based on BERT or RoBERTa.
[0073] Among them, BERT (Bidirectional Encoder Representations from Transformers) is a deep learning model based on the Transformer architecture.
[0074] RoBERTa (Robustly Optimized BERT Pretraining Approach) is a large language model based on BERT.
[0075] It should be noted that feature data includes key information (such as applicant name, application date), semantic features of the approval document content (i.e., features obtained through semantic analysis of a large language model), and other data that may affect the approval decision.
[0076] like Figure 2 FIG. 1 is a flowchart of an approval decision disclosed in an embodiment of the present invention. The process of extracting feature data from the approval document using a pre-trained large language model is as follows:
[0077] First, detect the structure type of the approval document.
[0078] If the document to be approved is a structured document, feature data is extracted from the approval document using a gradient boosting decision tree algorithm (i.e., XGBoost).
[0079] It should be noted that the gradient boosting decision tree algorithm evaluates the importance of features in structured documents and extracts feature data with higher importance.
[0080] If the document to be approved is an unstructured document, the approval document is input into a pre-trained large language model for semantic analysis to obtain feature data output by the large language model.
[0081] It should be noted that the large language model is used to understand and analyze the semantic content of approval documents. It can identify the meaning of keywords, phrases and entire sentences in the document, thereby extracting feature data that is useful for approval decisions.
[0082] If the document to be approved is an unstructured document, the unstructured part of the approval document is input into the large language model for semantic analysis to obtain the first part of feature data output by the large language model; the second part of feature data is extracted from the structured part of the approval document using a gradient boosting decision tree algorithm; the first part of feature data and the second part of feature data are combined to obtain the feature data of the approval document.
[0083] Among them, the training process of the large language model is as follows:
[0084] Build a large language model to be trained; obtain multiple historical approval documents, and perform data cleaning on each historical approval document; divide each historical approval document after data cleaning into a training set and a verification set; use the training set to train the large language model to be trained to obtain a trained large language model; use the verification set to verify the performance of the trained large language model; if the verification passes, obtain a trained large language model.
[0085] Among them, data cleaning refers to segmenting historical approval documents and removing irrelevant information.
[0086] The specific process of dividing each historical approval document after data cleaning into a training set and a validation set is as follows: extract features (such as context information) used to train a large language model from each historical approval document, perform part-of-speech tagging, and divide the tagged features into a training set and a validation set.
[0087] It should be noted that data cleaning is performed using Apache Spark and feature extraction is performed using TensorFlow.
[0088] Among them, Apache Spark is an open source distributed computing framework, mainly used for big data processing and analysis. TensorFlow is an open source machine learning framework, which is used to build and train deep learning models and perform various machine learning tasks.
[0089] like Figure 2As shown, after extracting feature data from the approval document using a pre-trained large language model, a feature integrity check is performed on the feature data; if the feature data is incomplete, a supplementary information request is generated, and the process returns to execute the step of obtaining the approval document.
[0090] The supplementary information request is used to prompt the approval requester to supplement the information and re-enter the approval document.
[0091] Step S103: Utilize the condition matching engine to analyze whether the characteristic data matches the preset approval condition; if so, execute step S104; if not, execute step S105.
[0092] In step S103, the condition matching engine refers to a pre-developed rule-based engine that automatically matches approval conditions and policies in combination with the output of the large language model.
[0093] Step S104: Determine whether the approval document has passed the approval.
[0094] Step S105: Input the feature data into the real-time optimized machine learning model for prediction to obtain the confidence level of the approval document.
[0095] In step S105, the prediction target of the machine learning model is usually the approval result, that is, whether the application is approved. The machine learning model predicts the probability of the application being approved or rejected based on the input feature data, and the probability is positively correlated with the confidence level.
[0096] like Figure 3 FIG. 1 is a flowchart of data collection and preprocessing disclosed in an embodiment of the present invention.
[0097] The process of real-time optimization of machine learning models includes:
[0098] Using data collection tools, raw data is collected in real time, the raw data is preprocessed to obtain target feature data, and the target feature data is input into the machine learning model for retraining to obtain a real-time optimized machine learning model.
[0099] Among them, data collection tools include: Apache Kafka, RabbitMQ and Apache NiFi.
[0100] Apache Kafka is an open source distributed stream processing platform; RabbitMQ is an open source message broker software; Apache NiFi is an open source data integration and data flow management tool designed to automate the creation, monitoring, and management of data flows.
[0101] According to the data source type corresponding to the original data, the corresponding data collection tool is used. Specifically, structured data is collected using Apache Kafka, unstructured data is collected using RabbitMQ, and semi-structured data is collected using Apache NiFi.
[0102] The original data includes: the target approval document that has completed the approval, the approval result of the target approval document, the analysis information obtained by analyzing the feedback information input by the approval requester, and the log data used to characterize the operation record of the approval process.
[0103] The target approval document is the main data that needs to be processed in the approval process.
[0104] The analysis information may include satisfaction surveys, complaints, suggestions, etc. The analysis information uses sentiment analysis technology VADER (Valence Aware Dictionary and sEntiment Reasoner) to analyze the feedback information input by the approval requester.
[0105] The approval result of the target approval document is the result data after the approval process is completed, which may include approval, rejection or other approval decisions.
[0106] Log data is a log record of system operations, which may include user operations, system errors, abnormal events, etc.
[0107] like Figure 3 As shown, the process of preprocessing the original data to obtain the target feature data is as follows:
[0108] Clean each original data; determine whether each original data after data cleaning is qualified;
[0109] For each qualified original data, feature data for training is extracted using a feature extraction tool; for each unqualified original data, manually input correction instructions are received, the unqualified original data is corrected based on the correction instructions, and feature data for training is extracted from the corrected original data using a feature extraction tool; data standardization is performed on the feature data used for training to obtain target feature data that conforms to the input format of the machine learning model.
[0110] More specifically, Apache Spark is used to clean the data and remove noise and outliers.
[0111] The Scikit-learn library is used to standardize data to ensure the consistency of data in model training.
[0112] Preferably, before data standardization, an automated machine learning tool such as Auto-sklearn is also used for feature selection and transformation to enhance the predictive ability of the model.
[0113] like Figure 4 As shown, it is a flowchart of a real-time optimized machine learning model disclosed in an embodiment of the present invention.
[0114] Specifically, the target feature data is input into the machine learning model for retraining to obtain a real-time optimized machine learning model, including:
[0115] Input the target feature data into the machine learning model for retraining; evaluate whether the performance of the retrained machine learning model is improved; if so, obtain the real-time optimized machine learning model; if not, modify the model parameters of the machine learning model, and return to execute the step of inputting the target feature data into the machine learning model for retraining until the real-time optimized machine learning model is obtained.
[0116] Among them, cross validation and A / B testing are used to evaluate model performance.
[0117] Step S106: If the confidence level is greater than or equal to the confidence level threshold, it is determined that the approval document has passed the approval.
[0118] Step S107: If the confidence level is less than the confidence level threshold, an audit notification is generated to remind the auditor, and when a pass instruction is received from the auditor, the approval document is determined to be approved; when a reject instruction is received from the auditor, the approval document is determined to be not approved.
[0119] In one embodiment, if the approval document fails to pass the approval, a cause analysis is performed, and the result of the analysis determines whether to update the rules set in the condition matching engine.
[0120] Based on the above-mentioned embodiment of the present invention, a large model approval method for the government affairs industry is disclosed, obtaining approval documents; using a pre-trained large language model to extract feature data from the approval documents; using a condition matching engine to analyze whether the feature data matches the preset approval conditions; if so, determining that the approval document has passed the approval; if not, inputting the feature data into a real-time optimized machine learning model for prediction to obtain the confidence of the approval document; if the confidence is greater than or equal to the confidence threshold, determining that the approval document has passed the approval; if the confidence is less than the confidence threshold, generating an audit notification for reminding the auditor, and when receiving the passing instruction input by the auditor, determining that the approval document has passed the approval, and when receiving the rejecting instruction input by the auditor, determining that the approval document has not passed the approval. In this solution, a large language model is used to accurately understand the content and context of the approval document, to achieve automated approval and reduce the probability of errors, and to use a continuously optimized machine learning model to improve the quality and speed of decision-making.
[0121] Corresponding to the large model approval method for the government affairs industry disclosed in the above embodiment of the present invention, Figure 5 As shown, it is a structural diagram of a large model approval device for the government affairs industry disclosed in an embodiment of the present invention, and the device includes:
[0122] The acquisition unit 501 is used to acquire the approval document.
[0123] The extraction unit 502 is used to extract feature data from the approval document using a pre-trained large language model.
[0124] The approval unit 503 is used to use the condition matching engine to analyze whether the feature data matches the preset approval conditions; if so, it is determined that the approval document has passed the approval; if not, the feature data is input into the real-time optimized machine learning model for prediction to obtain the confidence of the approval document; if the confidence is greater than or equal to the confidence threshold, it is determined that the approval document has passed the approval; if the confidence is less than the confidence threshold, an audit notification is generated to remind the auditor, and when a pass instruction input by the auditor is received, it is determined that the approval document has passed the approval, and when a rejection instruction input by the auditor is received, it is determined that the approval document has not passed the approval.
[0125] In one embodiment, the extraction unit 502 is specifically configured to:
[0126] Check the structure type of approval documents;
[0127] If the document to be approved is a structured document, feature data is extracted from the approval document using a gradient boosting decision tree algorithm;
[0128] If the document to be approved is an unstructured document, the approval document is input into a pre-trained large language model for semantic analysis to obtain feature data output by the large language model;
[0129] If the document to be approved is an unstructured document, the unstructured part of the approval document is input into the large language model for semantic analysis to obtain the first part of feature data output by the large language model; the second part of feature data is extracted from the structured part of the approval document using a gradient boosting decision tree algorithm; the first part of feature data and the second part of feature data are combined to obtain the feature data of the approval document.
[0130] In one embodiment, the device further comprises:
[0131] An integrity check unit, used for performing a feature integrity check on the feature data after extracting the feature data from the approval document using a pre-trained large language model;
[0132] If the characteristic data is incomplete, a supplementary information request is generated and the process returns to the step of obtaining the approval document; the supplementary information request is used to prompt the approval requester to supplement the information and re-enter the approval document.
[0133] In one embodiment, the device further comprises:
[0134] A large language model training unit, used to build a large language model to be trained;
[0135] Obtain multiple historical approval documents and perform data cleansing on each historical approval document;
[0136] Divide each historical approval document after data cleaning into a training set and a validation set;
[0137] The training set is used to train the large language model to be trained, thereby obtaining a trained large language model;
[0138] Use the validation set to verify the performance of the trained large language model;
[0139] If the verification passes, the trained large language model is obtained.
[0140] In one embodiment, the device further comprises:
[0141] The real-time optimization unit is used to collect raw data in real time using data collection tools; the raw data includes: the target approval document that has completed the approval, the approval result of the target approval document, the analysis information obtained by analyzing the feedback information input by the approval requester, and the log data used to characterize the operation record of the approval process;
[0142] Preprocess the original data to obtain target feature data;
[0143] The target feature data is input into the machine learning model for retraining to obtain a real-time optimized machine learning model.
[0144] Among them, the real-time optimization unit used to pre-process the original data to obtain the target feature data is specifically used for:
[0145] Clean each original data;
[0146] Determine whether the original data after each data cleaning is qualified;
[0147] For each qualified raw data, feature extraction tools are used to extract feature data for training;
[0148] For each unqualified original data, receiving a manually input correction instruction, correcting the unqualified original data based on the correction instruction, and using a feature extraction tool to extract feature data for training from the corrected original data;
[0149] The feature data used for training is standardized to obtain target feature data that conforms to the input format of the machine learning model.
[0150] Among them, the real-time optimization unit for inputting the target feature data into the machine learning model for retraining to obtain the real-time optimized machine learning model is specifically used for:
[0151] Input the target feature data into the machine learning model for retraining;
[0152] Evaluate whether the performance of the retrained machine learning model improves;
[0153] If so, a real-time optimized machine learning model is obtained;
[0154] If not, modify the model parameters of the machine learning model and return to the step of inputting the target feature data into the machine learning model for retraining until a real-time optimized machine learning model is obtained.
[0155] Based on the above-mentioned embodiment of the present invention, a large model approval device for the government affairs industry disclosed obtains approval documents; uses a pre-trained large language model to extract feature data from the approval documents; uses a condition matching engine to analyze whether the feature data matches the preset approval conditions; if so, it is determined that the approval document has passed the approval; if not, the feature data is input into a real-time optimized machine learning model for prediction to obtain the confidence of the approval document; if the confidence is greater than or equal to the confidence threshold, it is determined that the approval document has passed the approval; if the confidence is less than the confidence threshold, an audit notification is generated to remind the auditor, and when the approval instruction input by the auditor is received, it is determined that the approval document has passed the approval, and when the rejection instruction input by the auditor is received, it is determined that the approval document has not passed the approval. In this solution, a large language model is used to accurately understand the content and context of the approval document, to achieve automated approval and reduce the probability of errors, and to use a continuously optimized machine learning model to improve the quality and speed of decision-making.
[0156] The present invention also provides an electronic device, please refer to Figure 6 , the electronic device includes a memory 601 and a processor 602.
[0157] Wherein, the memory 601 is used to store computer programs;
[0158] Processor 602 is used to execute computer programs, specifically to implement the large model approval method for the government affairs industry provided in any embodiment of the present application.
[0159] The present application also provides a computer storage medium for storing a computer program. When the computer program is executed, it is specifically used to implement the large model approval method for the government affairs industry provided in any embodiment of the present application.
[0160] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can refer to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiment. The system and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without creative work.
[0161] 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 the present invention.
[0162] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A large model approval method for the government affairs industry, characterized in that: The method comprises: Obtain approval documents; Extracting feature data from the approval document using a pre-trained large language model; Using a condition matching engine, analyzing whether the characteristic data matches a preset approval condition; If so, determining that the approval document has passed the approval; If not, the feature data is input into a real-time optimized machine learning model for prediction to obtain the confidence level of the approval document; If the confidence level is greater than or equal to the confidence level threshold, determining that the approval document has passed the approval; If the confidence level is less than the confidence level threshold, an audit notification is generated to remind the auditor, and when a pass instruction input by the auditor is received, it is determined that the approval document has passed the approval; when a reject instruction input by the auditor is received, it is determined that the approval document has failed the approval.
2. The method according to claim 1, characterized in that The method of extracting feature data from the approval document using a pre-trained large language model includes: Detecting the structural type of the approval document; If the document to be approved is a structured document, feature data is extracted from the approval document using a gradient boosting decision tree algorithm; If the document to be approved is an unstructured document, the approval document is input into a pre-trained large language model for semantic analysis to obtain feature data output by the large language model; If the document to be approved is an unstructured document, the unstructured part of the approval document is input into the large language model for semantic analysis to obtain a first part of feature data output by the large language model; the second part of feature data is extracted from the structured part of the approval document using the gradient boosting decision tree algorithm; the first part of feature data and the second part of feature data are combined to obtain feature data of the approval document.
3. The method according to claim 1, characterized in that: After extracting feature data from the approval document using the pre-trained large language model, the method further includes: Performing a feature integrity check on the feature data; If the characteristic data is incomplete, a supplementary information request is generated and the process returns to execute the step of obtaining the approval document; the supplementary information request is used to prompt the approval requester to supplement the information and re-enter the approval document.
4. The method according to claim 1, characterized in that: The training process of the large language model includes: Build a large language model to be trained; Acquire multiple historical approval documents, and perform data cleaning on each of the historical approval documents; Dividing each of the historical approval documents after data cleaning into a training set and a validation set; Using the training set to train the large language model to be trained to obtain a trained large language model; Using the validation set to validate the performance of the trained large language model; If the verification passes, the trained large language model is obtained.
5. The method according to claim 1, characterized in that: The process of optimizing the machine learning model in real time includes: Using data collection tools, raw data is collected in real time; the raw data includes: the target approval document that has completed the approval, the approval result of the target approval document, the analysis information obtained by analyzing the feedback information input by the approval requester, and the log data used to characterize the operation record of the approval process; Preprocessing the original data to obtain target feature data; The target feature data is input into the machine learning model for retraining to obtain the machine learning model after real-time optimization.
6. The method according to claim 5, characterized in that The preprocessing of the raw data to obtain target feature data includes: Performing data cleaning on each of the original data; Determine whether the original data after each data cleaning is qualified; For each qualified raw data, extract feature data for training using a feature extraction tool; For each unqualified raw data, receiving a manually input correction instruction, correcting the unqualified raw data based on the correction instruction, and extracting feature data for training from the corrected raw data using the feature extraction tool; The feature data used for training is subjected to data standardization processing to obtain target feature data that conforms to the input format of the machine learning model.
7. The method according to claim 5, characterized in that The step of inputting the target feature data into the machine learning model for retraining to obtain the machine learning model after real-time optimization includes: Inputting the target feature data into the machine learning model for retraining; evaluating whether the performance of the machine learning model after retraining is improved; If so, the machine learning model is optimized in real time; If not, modify the model parameters of the machine learning model, and return to the step of inputting the target feature data into the machine learning model for retraining until the machine learning model is optimized in real time.
8. A large model approval device for the government affairs industry, characterized in that: The device comprises: An acquisition unit, used to acquire approval documents; An extraction unit, used to extract feature data from the approval document using a pre-trained large language model; The approval unit is used to use a condition matching engine to analyze whether the feature data matches the preset approval conditions; if so, it is determined that the approval document has passed the approval; if not, the feature data is input into a real-time optimized machine learning model for prediction to obtain the confidence of the approval document; if the confidence is greater than or equal to a confidence threshold, it is determined that the approval document has passed the approval; if the confidence is less than the confidence threshold, an audit notification is generated to remind the auditor, and when a pass instruction input by the auditor is received, it is determined that the approval document has passed the approval, and when a reject instruction input by the auditor is received, it is determined that the approval document has not passed the approval.
9. A storage medium, characterized in that: Used to store computer programs, which, when executed, are specifically used to implement the large model approval method for the government affairs industry as described in any one of claims 1 to 7.
10. An electronic device, characterized in that: include: Memory and processor; The memory is used to store computer programs; The processor is used to execute the computer program, specifically to implement the large model approval method for the government affairs industry as described in any one of claims 1 to 7.