A method for automatically generating compliance reports based on large models

Abstract

The present invention belongs to the technical field of compliance reporting, and specifically is a method for automatically generating compliance reports based on a large model, comprising the following steps: Step 1: reading a report generation request; Step 2: setting a draft preparation phase: responding to a report generation request, retrieving and extracting a data packet from a database; Step 3: setting a draft generation phase: based on a report generation model, inputting a data packet to generate a first report draft; by pre-setting a fixed report template, and setting a blank form field that can be filled with data in the template, based on a report generation request generated by a user operation, extracting a data packet from the database, and then filling the data packet into the blank form field, thereby obtaining a first report draft, based on the first report draft, manually reviewing it again, and obtaining a second report draft after the manual review, and then deleting the blank form field that is not filled with a data packet in the second report draft by a computer to obtain a compliance report.

Description

Technical Field

[0001] The present invention belongs to the technical field of compliance reporting, and in particular to a method for automatically generating compliance reports based on a large model. Background Art

[0002] Compliance means that financial institutions' operations must be consistent with laws, regulations, and guidelines. This means that compliance reports must be generated to monitor and regulate their operations. Related technologies primarily rely on manual methods to generate compliance reports. While manual methods are feasible, they are costly and difficult to control in terms of quality.

[0003] Based on the above, methods for automatically generating compliance reports have emerged in the prior art, such as a compliance report generation method disclosed in application number 202410186630.6, which includes: collecting corresponding business data from multiple business systems and distributing the business data to a cloud database; obtaining business data from the cloud database in response to a compliance report generation request, the compliance report generation request including compliance requirements and user needs; determining a compliance report template that meets the compliance requirements and user needs from a compliance report template library; the compliance report template library is determined in the following manner: obtaining an initial compliance report template; obtaining multiple preset compliance requirements and multiple preset user needs, and combining the multiple preset compliance requirements and the multiple preset user needs to generate multiple preset compliance report requirements; adjusting the initial compliance report template according to the multiple preset compliance report requirements to obtain compliance report templates corresponding to the multiple preset compliance report requirements;

[0004] However, in the above-mentioned existing technologies, since there are many types of compliance reports at this stage, such as internal compliance reports for enterprises, such as audit reports and risk management reports, and external compliance reports for regulatory authorities, such as financial reports, environmental impact assessment reports, etc., assuming that as time goes by, compliance requirements and user needs increase, the number of template libraries will increase. Among them, when compliance requirements and user needs are not determined in advance, templates corresponding to compliance requirements and user needs cannot be obtained. That is, the above-mentioned existing technologies are limited to the predetermined template library, and when a report type different from the existing template appears, it cannot adapt.

[0005] To this end, the present invention provides a method for automatically generating compliance reports based on a large model. Summary of the Invention

[0006] In order to make up for the deficiencies of the prior art, at least one technical problem raised in the background technology is solved.

[0007] The technical solution adopted by the present invention to solve the technical problem is: a method for automatically generating a compliance report based on a large model, comprising the following steps:

[0008] Step 1: Read the report generation request;

[0009] Step 2: Set up the draft preparation phase: respond to the report generation request, retrieve and extract the data package from the database;

[0010] Step 3: Setting up the draft generation phase: Based on the report generation model, inputting the data package to generate a first report draft; the report generation model pre-sets a report template, and the data package is filled into the blank form fields in the report template;

[0011] Step 4: Set up the manual review stage: Based on the review and modification of the first draft report, generate the second draft report;

[0012] Step 5: Set up the final verification stage: Check the second draft report, remove the blank form fields that are not filled in the data package, and obtain a compliant report that meets the report generation request.

[0013] Preferably, the report generation request in step 1 is a compliance report generation operation generated by a user based on a computer operation, input information corresponding to the compliance report generation operation is obtained, and a compliance report generation request corresponding to the input information is generated; the input information includes report type information and user information;

[0014] The data packet in step 2 is composed of multiple data sets, and the data sets include any one of data information and regulatory information; the data sets are pre-distributedly stored in the database.

[0015] Preferably, the training method of the report generation model in step 3 is:

[0016] Get a sample of data packets;

[0017] Input the data package sample into the initial report generation model, and generate a compliance report sample corresponding to the data package based on the report template;

[0018] Based on the differences between the compliance report sample and the standard compliance report, the initial report generation model is trained to obtain an optimized report generation model;

[0019] The report template includes a plurality of blank form fields, and a label is set in each blank form field.

[0020] Preferably, the method for generating the first draft report in step 3 is:

[0021] Identify multiple data sets based on the data recognition model and extract feature parameters;

[0022] Match and map the extracted feature parameters with the labels corresponding to each blank form field in the report template;

[0023] When the feature parameters and labels are matched successfully, the data set corresponding to the feature parameters is filled into the blank form field corresponding to the label;

[0024] Perform a completeness check on the data set. If there are no omissions, generate a first draft report. If there are omissions, generate an error message along with the first draft report.

[0025] Preferably, the method for reviewing and modifying the first draft report and generating the second draft report in step 4 is:

[0026] Determine whether there is an error message:

[0027] When there is an error message, manually fill in the missing data set into the corresponding blank form field;

[0028] If there is no error message, check whether the data set matches the blank form field:

[0029] If the data set matches the blank form field, a second draft report is generated;

[0030] If the data set does not match the blank form field, manually modify it and generate a second draft report.

[0031] Preferably, the method for deleting the unfilled blank form field in step 5 is:

[0032] The computer steps through each blank form field in the report template when:

[0033] If any blank form field is filled with data, it is considered a valid form field.

[0034] On the contrary, an unfilled blank form field is determined to be an invalid form field;

[0035] Remove the blank form fields that are determined to be invalid form fields.

[0036] Preferably, the training method of the data recognition model comprises the following steps:

[0037] Collect data set samples and prepare report templates;

[0038] Based on the initial data recognition model, a deep learning algorithm is used to perform semantic and keyword recognition on the dataset samples;

[0039] Based on the recognition results, generate feature parameters of the data set samples;

[0040] Match feature parameters with preset labels;

[0041] Repeatedly generate feature parameters and optimize the matching relationship between feature parameters and labels;

[0042] Set up a test set to verify the effectiveness of the data recognition model;

[0043] Achieve the preset matching accuracy and obtain the data recognition model.

[0044] Preferably, the method further includes step 6: creating a log record for each report and storing the log record in a database; the log record includes the time when the report was generated, the source of the data package used, the editing history, and the review and modification information; the report includes a first draft report, a second draft report, and a compliance report;

[0045] A report recording module is established in the database for storing and managing reports; the report recording module creates a unique identifier for each report, and each version of the report shares the unique identifier; the various versions of the report are the first draft report, the second draft report and the compliance report.

[0046] Preferably, the step 4 further comprises the following steps:

[0047] When there is an error message, the missing data set is displayed;

[0048] Manually match the missing datasets with preset labels, and

[0049] Manually create feature parameters for the missing dataset;

[0050] Based on the created feature parameters and corresponding labels, the matching relationship between feature parameters and labels in the data recognition model is updated.

[0051] Preferably, the data packet in step 2 is stored in a database after preprocessing, and the preprocessing method further includes the following steps:

[0052] Obtaining the pre-processed original data packet, encrypting the original data packet, and generating an encrypted data packet;

[0053] Using a signature algorithm, multiple parameters that make up the data packet are sorted, concatenated, and calculated to generate a signature;

[0054] The generated signature is stored together with the encrypted data packet in the database;

[0055] In the step 2, extracting the data packet further comprises the following steps:

[0056] Use the same signature algorithm to verify the validity of the signature:

[0057] When the verification result is valid, the encrypted data packet is decrypted to restore the original data packet;

[0058] Otherwise, a warning message will be issued.

[0059] The beneficial effects of the present invention are as follows:

[0060] 1. The present invention discloses a method for automatically generating a compliance report based on a large model. A fixed report template is pre-set, and blank form fields that can be filled with data are provided in the template. Based on a report generation request generated by a user operation, a data packet is extracted from a database and then filled into the blank form fields, thereby generating a first draft report. The first draft report is then manually reviewed. After the manual review, a second draft report is generated. A computer then deletes the blank form fields in the second draft report that are not filled with data packets, thereby generating a compliance report.

[0061] 2. The present invention describes a method for automatically generating compliance reports based on a large model, which completes the association between the data packet and the blank form field by setting a label in the blank form field and using the mapping matching between the label and the characteristic parameters representing the data packet. Therefore, a deep learning algorithm is used to perform semantic and keyword recognition on the data sample, and then matching training is performed with the label based on the extracted characteristic parameters, thereby optimizing the extraction of the data packet characteristic parameters by the data recognition model, which is conducive to matching with the label in the blank form field. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] The present invention will be further described below with reference to the accompanying drawings.

[0063] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0064] In order to make the technical means, creative features, objectives and effects achieved by the present invention easier to understand, the present invention is further described below in conjunction with specific implementation methods.

[0065] like Figure 1 As shown, a method for automatically generating a compliance report based on a large model according to an embodiment of the present invention includes the following steps:

[0066] Step 1: Read the report generation request;

[0067] Step 2: Set up the draft preparation phase: respond to the report generation request, retrieve and extract the data package from the database;

[0068] Step 3: Setting up the draft generation phase: Based on the report generation model, inputting the data package to generate a first report draft; the report generation model pre-sets a report template, and the data package is filled into the blank form fields in the report template;

[0069] Step 4: Set up the manual review stage: Based on the review and modification of the first draft report, generate the second draft report;

[0070] Step 5: Set up the final verification stage: Check the second draft report, remove the blank form fields that are not filled in the data package, and obtain a compliant report that meets the report generation request.

[0071] In the above-mentioned existing technologies, since there are many types of compliance reports at this stage, such as internal compliance reports for enterprises, such as audit reports and risk management reports, and external compliance reports for regulatory authorities, such as financial reports and environmental impact assessment reports, etc., assuming that compliance requirements and user needs increase over time, the number of template libraries will increase. Among them, when compliance requirements and user needs are not determined in advance, templates corresponding to the compliance requirements and user needs cannot be obtained. That is, the above-mentioned existing technologies are limited to the predetermined template library, and when a report type different from the existing template appears, it cannot adapt.

[0072] In one embodiment of the present invention, a fixed report template is pre-set, and a blank form field that can be filled with data is set in the template. Based on the report generation request generated by the user operation, a data packet is extracted from the database, and then the data packet is filled into the blank form field, thereby obtaining a first draft report. Based on the first draft report, it is manually reviewed again. After the manual review, a second draft report is obtained, and then the computer deletes the blank form fields that are not filled with data packets in the second draft report to obtain a compliance report. It is worth noting that multiple chapter contents are pre-set in the report template, and blank form fields are set in each chapter content. However, due to different compliance report types, not every compliance report needs to use all chapters. Therefore, due to different compliance report types, there must be unfilled blank form fields in the report template. Therefore, after manual review, it is necessary to remove the blank form fields that are not filled with data packets based on the computer, and update the page number of the compliance report.

[0073] In one embodiment, the report generation request in step 1 is a compliance report generation operation generated by a user based on a computer operation, input information corresponding to the compliance report generation operation is obtained, and a compliance report generation request corresponding to the input information is generated; the input information includes report type information and user information;

[0074] The data packet in step 2 is composed of multiple data sets, and the data sets include any one of data information and regulatory information; the data sets are pre-distributedly stored in the database.

[0075] In one embodiment of the present invention, a report generation request includes report type information and user information. The report type information is used to extract a data package from a database. For example, a financial report requires extracting annual business data and corresponding account information from the database. The user information indicates the reporting unit of the user compliance report and is generally used on the cover of the compliance report. A data package is composed of multiple data sets. For example, a financial report may require providing business data, account information, and other related materials. These materials may be distributed and stored in different units in the database. Therefore, a data package generally consists of multiple data sets.

[0076] It is worth noting that when the computer retrieves an update to a pre-stored data set, it can actively download it to the database. Secondly, in order to distinguish the historical version from the current version of the data set, a timestamp is created for each type of data set to distinguish the version of the data set. When generating a compliance report, the corresponding data set is extracted according to the compliance report request to prevent errors. In addition, the data stored or downloaded to the database needs to be preprocessed, including data cleaning, data organization and data analysis, among which data analysis includes analyzing the compliance and legality of the data source.

[0077] In one embodiment, the training method of the report generation model in step 3 is:

[0078] Get a sample of data packets;

[0079] Input the data package sample into the initial report generation model, and generate a compliance report sample corresponding to the data package based on the report template;

[0080] Based on the differences between the compliance report sample and the standard compliance report, the initial report generation model is trained to obtain an optimized report generation model;

[0081] The report template includes a plurality of blank form fields, and a label is set in each blank form field.

[0082] In one embodiment of the present invention, the initial report generation model refers to an untrained model for generating compliance reports, which can be a deep learning model such as a recursive neural network (RNN), a convolutional neural network (CNN), etc. The compliance report sample corresponding to the data packet sample can be determined through the initial report generation model. The initial report generation model is trained based on multiple data packet samples to obtain an optimized report generation model. It is worth noting that the report template is set in the report generation model. Unlike the prior art, in this application, there is only one report template, and multiple chapters are set to meet the needs of known types of compliance reports. At the same time, a blank form field is set in each chapter for filling in the data packet. Specifically, in order to facilitate the matching of the data packet with the blank form field, a label is also set in the blank form field, and the label is used to strengthen the association between the data packet and the blank form field, so that the data packet can be accurately filled in the blank form field.

[0083] In one embodiment, the method for generating the first draft report in step 3 is:

[0084] Identify multiple data sets based on the data recognition model and extract feature parameters;

[0085] Match and map the extracted feature parameters with the labels corresponding to each blank form field in the report template;

[0086] When the feature parameters and labels are matched successfully, the data set corresponding to the feature parameters is filled into the blank form field corresponding to the label;

[0087] Perform a completeness check on the data set. If there are no omissions, generate a first draft report. If there are omissions, generate an error message along with the first draft report.

[0088] In one embodiment of the present invention, a data set is identified through a data recognition model, and characteristic parameters representing the data set are extracted. The extracted characteristic parameters are used to map and match with the labels corresponding to each blank form field in the report template, so that after the characteristic parameters and the labels are successfully matched, the data set corresponding to the characteristic parameters can be filled into the corresponding blank form field. It is worth noting that in the process of filling the data set into the blank form field, there may be cases where the data set is missing, that is, there is a data set that is not filled into any blank form field. Therefore, it is also necessary to review whether there is any omission in the data set. If there is any omission, it is necessary to generate the first draft report and generate an error prompt message at the same time to facilitate quick resolution during manual review.

[0089] In one embodiment, the method for reviewing and modifying the first draft report to generate the second draft report in step 4 is:

[0090] Determine whether there is an error message:

[0091] When there is an error message, manually fill in the missing data set into the corresponding blank form field;

[0092] If there is no error message, check whether the data set matches the blank form field:

[0093] If the data set matches the blank form field, a second draft report is generated;

[0094] If the data set does not match the blank form field, manually modify it and generate a second draft report.

[0095] Even though the existing technology can realize the automatic generation of compliance reports based on data packages and report templates, manual review of the generated compliance reports is still required to prevent defects. In one embodiment of the present invention, after the first draft report is generated, manual review is required, which specifically includes checking whether the data set matches the blank form field. If so, a second draft report is generated. If not, manual modification is required to generate the second draft report. In addition, when the first draft report is generated, after an error prompt message appears, it is necessary to check which blank form field the missing data set matches, and after manual filling, generate the second draft report.

[0096] In one embodiment, the method for deleting the unfilled blank form field in step 5 is:

[0097] The computer steps through each blank form field in the report template when:

[0098] If any blank form field is filled with data, it is considered a valid form field.

[0099] On the contrary, an unfilled blank form field is determined to be an invalid form field;

[0100] Remove the blank form fields that are determined to be invalid form fields.

[0101] Since the report template contains multiple blank form fields, and due to different compliance report types, some blank form fields may not be filled with data packets. In one embodiment of the present invention, after the second report draft is generated, a computer checks each blank form field in the report template. When a blank form field is filled with a data packet, it indicates that the blank form field is a valid form field, otherwise it is an invalid form field. The invalid form field is then automatically removed based on the computer to obtain a compliance report.

[0102] In one embodiment, the data recognition model training method includes the following steps:

[0103] Collect data set samples and prepare report templates;

[0104] Based on the initial data recognition model, a deep learning algorithm is used to perform semantic and keyword recognition on the dataset samples;

[0105] Based on the recognition results, generate feature parameters of the data set samples;

[0106] Match feature parameters with preset labels;

[0107] Repeatedly generate feature parameters and optimize the matching relationship between feature parameters and labels;

[0108] Set up a test set to verify the effectiveness of the data recognition model;

[0109] Achieve the preset matching accuracy and obtain the data recognition model.

[0110] As mentioned above, this application completes the association between the data packet and the blank form field by setting a label in the blank form field and matching the label with the characteristic parameters representing the data packet. Therefore, a deep learning algorithm is used to perform semantic and keyword recognition on the data sample, and then matching training is performed with the label based on the extracted characteristic parameters, thereby optimizing the data recognition model for extracting the characteristic parameters of the data packet, which is conducive to matching with the label in the blank form field;

[0111] It is worth noting that the repeated generation of feature parameters and the optimization of the matching relationship between feature parameters and labels also include the adaptive updating of the feature parameter matching mechanism corresponding to the label; specifically, after obtaining the data recognition model, semantic and keyword recognition is performed based on the extracted data packet to obtain the feature parameters of the data packet. For continuously updated data packets, such as laws and regulations, which may change over time, when there is no report generation request, when the computer retrieves relevant legal and regulatory updates, it downloads the corresponding laws and regulations, and performs semantic and keyword recognition on the updated laws and regulations based on the deep learning algorithm to generate new feature parameters, and establish an association relationship with the label, so that when a report generation request is subsequently generated, the feature parameters can be accurately generated and matched with the label.

[0112] In one embodiment, the method further includes step 6: creating a log record for each report and storing the log record in a database; the log record includes the time the report was generated, the source of the data package used, the editing history, and review and modification information; the report includes a first draft report, a second draft report, and a compliance report;

[0113] A report recording module is established in the database for storing and managing reports; the report recording module creates a unique identifier for each report, and each version of the report shares the unique identifier; the various versions of the report are the first draft report, the second draft report and the compliance report.

[0114] In one embodiment of the present invention, when a user operation generates a report generation request, a first report draft and a second report draft are necessary in the process of extracting the data package based on the computer and generating a compliance report with the report template. In order to facilitate the establishment of a complete audit trail of the report generation process, ensure data integrity and comply with relevant regulations, a detailed log record is created for each report, so that the report generation process can be tracked and traced. In addition, as the number of compliance reports generated increases, omissions may occur when retrieving compliance reports at a later time. Therefore, in one embodiment of the present invention, a unique identifier is created for each report. The unique identifier allows users to quickly retrieve and compare different versions of reports, thereby achieving effective document management and control.

[0115] In one embodiment, the step 4 further includes the following steps:

[0116] When there is an error message, the missing data set is displayed;

[0117] Manually match the missing datasets with preset labels, and

[0118] Manually create feature parameters for the missing dataset;

[0119] Based on the created feature parameters and corresponding labels, the matching relationship between feature parameters and labels in the data recognition model is updated.

[0120] Based on the above, after obtaining the data recognition model, the matching relationship between the preset feature parameters and labels can be obtained based on the data samples used for training. However, in actual application, the data recognition model may not be able to recognize updated laws and regulations, resulting in the inability to generate feature parameters that can match the labels, resulting in individual data sets being missed. In the manual review stage, it is necessary to manually establish a matching association between the missed data sets and the preset labels, and then create feature parameters based on the missed data sets, and update the matching relationship between the created feature parameters and the corresponding labels, so as to achieve real-time optimization of the data recognition model. It is worth noting that in the data recognition model, the preset label may correspond to multiple feature parameters, that is, multiple feature parameters all point to the same label.

[0121] In one embodiment, the data packets in step 2 are stored in a database after being pre-processed, and the pre-processing method further includes the following steps:

[0122] Obtaining the pre-processed original data packet, encrypting the original data packet, and generating an encrypted data packet;

[0123] Using a signature algorithm, multiple parameters that make up the data packet are sorted, concatenated, and calculated to generate a signature;

[0124] The generated signature is stored together with the encrypted data packet in the database;

[0125] In the step 2, extracting the data packet further comprises the following steps:

[0126] Use the same signature algorithm to verify the validity of the signature:

[0127] When the verification result is valid, the encrypted data packet is decrypted to restore the original data packet;

[0128] Otherwise, a warning message will be issued.

[0129] With the development of network technology, data faces security threats such as interception and tampering during transmission; for compliance reports generated by enterprises, if the original data packet is tampered with or deleted, the data packet will be incomplete or missing data, which will lead to the generation of non-compliant reports, affecting the enterprise; based on the above, in one embodiment of the present invention, the original data packet is encrypted and a signature is generated at the same time, and the signature and the encrypted data packet are stored together in a database. When a compliance report generation request occurs, the validity of the encrypted data packet is verified, that is, the accuracy of the signature is verified by the same signature algorithm. If the signature changes, it means that the encrypted data packet has been tampered with, and the integrity of the data packet needs to be checked, otherwise it will affect the enterprise, that is, a warning prompt message is generated to inform the user. On the contrary, if the signature has not changed, that is, the signature is valid, the encrypted data packet is decrypted, and the original data packet is restored for generating a compliance report.

[0130] Working principle: A fixed report template is set in advance, and blank form fields that can be filled with data are set in the template. Based on the report generation request generated by the user operation, the data package is extracted from the database, and then the data package is filled into the blank form field, thereby obtaining the first draft report. Based on the first draft report, it is manually reviewed again. After the manual review, the second draft report is obtained, and then the computer deletes the blank form fields that are not filled with data packages in the second draft report to obtain a compliance report; it is worth noting that multiple chapter contents are pre-set in the report template, and blank form fields are set in each chapter content, but due to different compliance report types, not every compliance report needs to use all chapters. Therefore, due to different compliance report types, there must be unfilled blank form fields in the report template. Therefore, after manual review, it is necessary to remove the blank form fields that are not filled with data packages based on the computer, and update the page number of the compliance report.

[0131] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the foregoing embodiments. The foregoing embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for automatically generating compliance reports based on a large model, characterized in that: The following steps are involved: Step 1: Read the report generation request; Step 2: Set up the draft preparation phase: respond to the report generation request, retrieve and extract the data package from the database; Step 3: Setting up the draft generation phase: Based on the report generation model, inputting the data package to generate a first report draft; the report generation model pre-sets a report template, and the data package is filled into the blank form fields of the report template; The preset report template is a fixed report template that is pre-set, and a blank form field that can be filled with data is set in the template; Step 4: Set up the manual review stage: Based on the review and modification of the first draft report, generate the second draft report; Step 5: Set up the final verification stage: Check the second draft report, remove the blank form fields that are not filled in the data package, and obtain a compliant report that meets the report generation request; The method for generating the first draft report in step 3 is: Identify multiple data sets based on the data recognition model and extract feature parameters; Match and map the extracted feature parameters with the labels corresponding to each blank form field in the report template; When the feature parameters and labels are matched successfully, the data set corresponding to the feature parameters is filled into the blank form field corresponding to the label; Perform a completeness check on the data set. If there are no omissions, generate a first draft report. If there are omissions, generate an error message along with the first draft report. The training method of the data recognition model comprises the following steps: Collect data set samples and prepare report templates; Based on the initial data recognition model, a deep learning algorithm is used to perform semantic and keyword recognition on the dataset samples; Based on the recognition results, generate feature parameters of the data set samples; Match feature parameters with preset labels; Repeatedly generate feature parameters and optimize the matching relationship between feature parameters and labels; Set up a test set to verify the effectiveness of the data recognition model; Achieve the preset matching accuracy and obtain the data recognition model.

2. The method for automatically generating compliance reports based on a large model according to claim 1, characterized in that: The report generation request in step 1 is in response to a compliance report generation operation generated by a user based on a computer operation, obtaining input information corresponding to the compliance report generation operation, and generating a compliance report generation request corresponding to the input information; the input information includes report type information and user information; The data packet in step 2 is composed of multiple data sets, and the data sets include any one of data information and regulatory information; the data sets are pre-distributedly stored in the database.

3. The method for automatically generating compliance reports based on a large model according to claim 1, characterized in that: The training method of the report generation model in step 3 is: Get a sample of data packets; Input the data package sample into the initial report generation model, and generate a compliance report sample corresponding to the data package based on the report template; Based on the differences between the compliance report sample and the standard compliance report, the initial report generation model is trained to obtain an optimized report generation model; The report template includes a plurality of blank form fields, and a label is set in each blank form field.

4. The method for automatically generating compliance reports based on a large model according to claim 1, characterized in that: The method for reviewing and modifying the first draft report and generating the second draft report in step 4 is: Determine whether there is an error message: When there is an error message, manually fill in the missing data set into the corresponding blank form field; If there is no error message, check whether the data set matches the blank form field: If the data set matches the blank form field, a second draft report is generated; If the data set does not match the blank form field, manually modify it and generate a second draft report.

5. The method for automatically generating compliance reports based on a large model according to claim 1, characterized in that: The method for deleting the unfilled blank form fields in step 5 is as follows: The computer steps through each blank form field in the report template when: If any blank form field is filled with data, it is considered a valid form field. On the contrary, an unfilled blank form field is determined to be an invalid form field; Remove the blank form fields that are determined to be invalid form fields.

6. The method for automatically generating compliance reports based on a large model according to claim 1, characterized in that: The method further includes step 6: creating a log record for each report and storing the log record in a database; the log record includes the time when the report was generated, the source of the data package used, the editing history, and the review and modification information; the report includes a first draft report, a second draft report, and a compliance report; A report recording module is established in the database for storing and managing reports; the report recording module creates a unique identifier for each report, and each version of the report shares the unique identifier; the various versions of the report are the first draft report, the second draft report and the compliance report.

7. The method for automatically generating compliance reports based on a large model according to claim 4, characterized in that: The step 4 further comprises the following steps: When there is an error message, the missing data set is displayed; Manually match the missing datasets with preset labels, and Manually create feature parameters for the missing dataset; Based on the created feature parameters and corresponding labels, the matching relationship between feature parameters and labels in the data recognition model is updated.

8. The method for automatically generating compliance reports based on a large model according to claim 2, characterized in that: The data packets in step 2 are stored in the database after preprocessing, and the preprocessing method further includes the following steps: Obtaining the pre-processed original data packet, encrypting the original data packet, and generating an encrypted data packet; Using a signature algorithm, multiple parameters that make up the data packet are sorted, concatenated, and calculated to generate a signature; The generated signature is stored together with the encrypted data packet in the database; In the step 2, extracting the data packet further comprises the following steps: Use the same signature algorithm to verify the validity of the signature: When the verification result is valid, the encrypted data packet is decrypted to restore the original data packet; Otherwise, a warning message will be issued.

Images