Method and system for compiling and reviewing security assessment reports based on multi-AIAgengt collaboration

Abstract

The embodiment of the present invention provides a security assessment report editing method and system based on multi-AIAgengt collaboration, which belongs to the field of artificial intelligence technology. The method includes: collecting historical security assessment reports, and performing vectorized processing on the historical security assessment reports to obtain training sample information; fine-tuning and training each pre-selected model based on the training sample information to obtain a security assessment report editing model; responding to the report processing requirement information triggered by the user, matching the corresponding execution sub-model in the security assessment report editing model; based on the instructions corresponding to the report processing requirement information triggered by the user, executing the corresponding sub-model, and outputting and pushing the report processing results to the user end. The scheme of the present invention further improves the work efficiency and quality of network security level assessment through intelligent information processing and automated report generation, and provides strong technical support for the high-quality development of the industry.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a multi-AI agent collaboration-based security assessment report editing and reviewing method and a multi-AI agent collaboration-based security assessment report editing and reviewing system. Background Art

[0002] With the acceleration of the informatization process and the increasing importance of network security work, enterprises are paying more and more attention to network security level protection, and the level protection assessment report plays a vital role in this process. These reports comprehensively evaluate the security protection status of information systems from ten dimensions, including secure physical environment, secure communication network, and secure area boundaries, and provide enterprises with gap analysis between the corresponding security level protection requirements. However, in the current editing and review process of the level protection assessment report, there are a series of technical problems that need to be solved urgently.

[0003] First of all, the preparation and review process of the security assessment report usually involves a lot of tedious document preparation, verification and review work. These tasks not only require multiple assessors to invest a lot of time and energy, but also require them to have high professional knowledge to ensure the accuracy, consistency and professionalism of the report. Since each report must follow certain chapter arrangement and content requirements, covering technical parameters and detailed text descriptions in multiple dimensions, the complexity of the workflow has increased significantly. This cumbersome process not only consumes the assessors' working time, but also compresses their effective time and energy in core technology research and innovation, restricting the efficient implementation of network security level protection work.

[0004] Secondly, in the existing editing and review scheme, due to the lack of efficient document processing and information mining methods, it is difficult for technical personnel to dig out high-value information and data from a large number of long-term accumulated technical documents. This not only affects the efficiency of report writing, but also reduces the industry competitiveness of the assessment agency and hinders the process of high-quality development of graded protection work. Therefore, how to effectively reduce the repetitive labor and work intensity faced by assessors in the report editing and review process and improve work efficiency has become an important issue that needs to be urgently solved in the current field of network security grade assessment. Summary of the invention

[0005] The purpose of the embodiments of the present invention is to provide a method and system for compiling and reviewing security assessment reports based on multi-AIAgengt collaboration, so as to at least solve the problem of low efficiency in compiling and reviewing existing security assessment reports.

[0006] In order to achieve the above-mentioned objectives, the first aspect of the present invention provides a method for editing and reviewing security assessment reports based on multi-AIAgengt collaboration, the method comprising: collecting historical security assessment reports, and performing vectorized processing on the historical security assessment reports to obtain training sample information; fine-tuning and training each pre-selected model based on the training sample information to obtain a security assessment report editing model; matching the corresponding execution sub-model in the security assessment report editing model in response to report processing requirement information triggered by the user; executing the corresponding sub-model based on the prompt instruction corresponding to the report processing requirement information triggered by the user, and outputting and pushing the report processing result to the user end.

[0007] Optionally, performing vectorization processing on the historical security assessment reports to obtain training sample information includes: classifying the historical security assessment reports based on data type, and matching the vectorization methods corresponding to each classified report set based on the classification results; wherein the vectorization methods include any one or more of the TF-IDF method, the Word2Vec method, and the BERT method; performing vectorization processing on the corresponding classified report set based on the matched vectorization method to obtain training sample information.

[0008] Optionally, before executing the quantization processing of the corresponding classification report set, the method also includes: pushing each historical security assessment report to the user end for the user to calibrate the report.

[0009] Optionally, the fine-tuning training of each preselected model based on the training sample information to obtain the security assessment report editing model includes: performing data preprocessing on the training sample information to obtain standardized training samples; matching corresponding preselected models for various requirements based on user preset report processing requirements; based on the standardized training samples, performing fine-tuning training corresponding to each preselected model using cross-validation and / or hyperparameter tuning to obtain execution sub-models corresponding to report processing requirements; and merging the execution sub-models based on each report processing requirement to construct the security assessment report editing model.

[0010] Optionally, the report processing requirement information triggered by the user includes any one or more of: target text generation requirement, report writing requirement based on standard paradigm, report review requirement, report evaluation requirement, key information extraction requirement and key information analysis requirement of the report.

[0011] Optionally, if the report processing requirement information triggered by the user is a target text generation requirement, the corresponding instruction is the customized prompt information of the work requirement under the corresponding business scenario; if the report processing requirement information triggered by the user is a report writing requirement based on the standard paradigm, the corresponding instruction is the standard paradigm sequence number and report content information; if the report processing requirement information triggered by the user is a report review requirement or a report evaluation requirement, the corresponding instruction is the target report and the corresponding requirement instruction; if the report processing requirement information triggered by the user is a key information extraction requirement for the report or a key information analysis requirement for the report, the corresponding instruction is the target report and the corresponding requirement instruction.

[0012] Optionally, the method further includes: each time report processing is performed, corresponding report processing information is retained; at every predetermined period, new training samples are constructed based on all report processing information retained in the corresponding period, and model updates are performed on the corresponding sub-models constructed based on the new training samples.

[0013] The second aspect of the present invention provides a security assessment report editing system based on multi-AIAgengt collaboration, the system comprising: a collection unit, used to collect historical security assessment reports, and perform vectorized processing on the historical security assessment reports to obtain training sample information; a model fine-tuning unit, used to fine-tune and train each pre-selected model based on the training sample information to obtain a security assessment report editing model; a model matching unit, used to match the corresponding execution sub-model in the security assessment report editing model in response to report processing requirement information triggered by a user; an execution unit, used to execute the corresponding sub-model based on the prompt instruction corresponding to the report processing requirement information triggered by the user, and output and push the report processing result to the user end.

[0014] Optionally, the fine-tuning training of each preselected model based on the training sample information to obtain the security assessment report editing model includes: performing data preprocessing on the training sample information to obtain standardized training samples; matching corresponding preselected models for various requirements based on user preset report processing requirements; based on the standardized training samples, performing fine-tuning training corresponding to each preselected model using cross-validation and / or hyperparameter tuning to obtain execution sub-models corresponding to report processing requirements; and merging the execution sub-models based on each report processing requirement to construct the security assessment report editing model.

[0015] On the other hand, the present invention provides a computer-readable storage medium, which stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned multi-AIAgengt collaboration-based security assessment report editing and review method.

[0016] Through the above technical scheme, the scheme of the present invention successfully generates high-quality training sample information by collecting and vectorizing historical security assessment reports. This process effectively converts complex text data into a form that can be processed by machine learning, laying the foundation for the training of subsequent models. By fine-tuning and training each pre-selected model based on the training sample information, an optimized security assessment report editing model is obtained, which significantly improves the accuracy and efficiency of report editing. When the user triggers the report processing requirement, the system can quickly match and call the corresponding execution sub-model. By responding to the user's instructions and executing the corresponding sub-model, the report processing results can be efficiently output and pushed to the user end. This not only reduces the cumbersome steps in the traditional report editing process, but also greatly reduces the labor cost and work intensity, enabling professional and technical personnel to devote more energy to core technology research and innovation. In addition, the technical solution further improves the work efficiency and quality of network security level assessment through intelligent information processing and automated report generation, providing strong technical support for the high-quality development of the industry.

[0017] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following specific embodiments, they are used to explain the embodiments of the present invention, but do not constitute a limitation on the embodiments of the present invention. In the accompanying drawings:

[0019] Figure 1 It is a flowchart of the steps of a method for compiling and reviewing a security assessment report based on multi-AIAgengt collaboration provided by an embodiment of the present invention;

[0020] Figure 2 It is a system structure diagram of a security assessment report editing system based on multi-AIAgengt collaboration provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0021] The specific implementation of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the present invention, and is not used to limit the present invention.

[0022] Figure 1 This is a method flow chart of a method for compiling and reviewing a security assessment report based on multi-AIAgengt collaboration provided by an embodiment of the present invention. Figure 1 As shown, an embodiment of the present invention provides a method for compiling and reviewing a security assessment report based on multi-AIAgengt collaboration, the method comprising:

[0023] Step S10: Collect historical security assessment reports, and perform vectorization processing on the historical security assessment reports to obtain training sample information.

[0024] Specifically, the process of collecting historical security assessment reports is the basis of the entire solution. Historical reports contain rich security protection experience and technical parameters. By systematically collecting these reports, valuable data support can be provided for subsequent model training. On this basis, classifying the data types of the collected reports is the key to achieving efficient vectorization. The classification step ensures that the features of different types of reports are reasonably extracted and effectively utilized, thereby improving the accuracy and relevance of subsequent vectorization processing.

[0025] Furthermore, for each set of classified reports, we will match the appropriate vectorization method. Vectorization is the process of converting text data into a numerical format so that it can be effectively processed by machine learning models. Common vectorization methods include TF-IDF (term frequency-inverse document frequency), Word2Vec, and BERT. Each method has its own advantages and disadvantages. The TF-IDF method is simple and easy to use, suitable for quickly extracting keywords; Word2Vec builds word vectors by learning the relationship between words, which can capture deeper semantics; and BERT is a context-related word representation method based on deep learning, which is suitable for processing complex language structures. Therefore, using corresponding vectorization methods for different categories of reports can maximize the richness and accuracy of text information.

[0026] Furthermore, after completing the classification and vectorization matching, vectorization processing is performed on each classified report set based on the selected vectorization method to convert the text information into computer-readable training sample information. This process not only improves the efficiency of data processing, but also lays a solid foundation for the subsequent training of machine learning models. By converting historical security assessment reports into structured training sample information, the model can better understand and learn the patterns and features in the reports.

[0027] Based on the solution of the present invention, historical security assessment reports are collected and vectorized, so that important information in the reports can be systematized and structured, thereby improving the efficiency of data utilization. By classifying and processing different types of reports and combining appropriate vectorization methods, the diversity and accuracy of training sample information are ensured, which provides strong data support for subsequent model training.

[0028] Furthermore, by using a variety of vectorization methods, the model can flexibly cope with different types of text data, further improving the generalization and adaptability of the model. Whether it is simple keyword extraction or complex context understanding, the system can choose the most appropriate processing method, ensuring the efficiency and accuracy of the report editing process.

[0029] Furthermore, by building training sample information based on historical reports, the system can automatically generate security assessment reports, greatly reducing the workload of manual review and writing. This technology not only reduces the work pressure of professional and technical personnel, but also enables them to devote more energy to core technology research and innovative research. In addition, automated report generation also improves the consistency and professionalism of reports, reduces the risk of human error, and thus improves the overall efficiency and quality of network security protection work.

[0030] Preferably, before executing the quantization processing of the corresponding classification report set, the method also includes: pushing each historical security assessment report to the user end for the user to calibrate the report.

[0031] In an embodiment of the present invention, report calibration refers to the user reviewing and adjusting the pushed historical security assessment report to ensure the accuracy and applicability of the report content. By allowing users to participate in this process, user feedback information can be effectively collected to further improve the quality of the report. The user calibration process mainly includes the verification, modification and supplementation of key data in the report. This link allows users to conduct in-depth analysis and evaluation of key information such as security measures, vulnerabilities and historical event responses in the report based on actual conditions. The user's professional knowledge and practical experience can help identify potential deficiencies in the report, thereby ensuring that the report can better reflect the actual security status and needs in subsequent vector processing.

[0032] Step S20: fine-tune and train each pre-selected model based on the training sample information to obtain a security assessment report editing model.

[0033] Specifically, data preprocessing of training sample information is the basis for ensuring the quality of model training. The steps of data preprocessing include data cleaning, format standardization, feature selection, and data enhancement. Through cleaning, redundant and noisy data are removed; format standardization makes data from different sources have a consistent structure, so that the model can efficiently learn data features during training. Feature selection ensures that only data related to the target task is used, reducing the complexity of model training. Data enhancement can expand the sample size by synthesizing data to improve the generalization ability of the model.

[0034] Furthermore, model matching for user-preset report processing requirements is the key to improving model adaptability. Report processing requirements may include target text generation requirements, report writing requirements based on standard paradigms, report review requirements, report evaluation requirements, key information extraction requirements, and key information analysis requirements. Different pre-selected models may need to be selected for different requirements. For example, for target text generation requirements, a large language model may be preferred due to its powerful text generation capabilities; while for key information extraction requirements, a rule-based model or a structured data processing model may be more suitable. Through reasonable matching, it can ensure that each task has the most suitable model to perform, thereby improving the efficiency and accuracy of the task.

[0035] Furthermore, in the fine-tuning training phase, the pre-selected model is fine-tuned based on the standardized training samples using cross-validation and / or hyperparameter tuning. Cross-validation is a commonly used model evaluation method that ensures the stability and generalization ability of the model by dividing the data into multiple parts for training and testing. Hyperparameter tuning is to find the best hyperparameter settings through search algorithms (such as grid search, random search, etc.) to improve the performance of the model. After this phase, the generated execution sub-model will be better adapted to the specific report processing needs.

[0036] Furthermore, each execution sub-model is combined to construct a graded protection assessment report editing model. This comprehensive model is flexible and adaptable, and can generate corresponding reports based on different inputs and requirements to meet the various needs of users in network security graded protection assessment. Since the model is carefully tuned and trained, it can provide high-quality output in actual applications, ensuring the professionalism and accuracy of the report.

[0037] Step S30: In response to the report processing requirement information triggered by the user, the corresponding execution sub-model in the security assessment report editing model is matched.

[0038] Specifically, the report processing requirement information triggered by the user includes any one or more of: target text generation requirement, report writing requirement based on standard paradigm, report review requirement, report evaluation requirement, key information extraction requirement and key information analysis requirement of the report.

[0039] In an embodiment of the present invention, the report processing requirement information triggered by the user may cover various types, including target text generation requirements, report writing requirements based on standard paradigms, report review requirements, report evaluation requirements, key information extraction requirements, and key information analysis requirements. For each requirement, the security assessment report editing model pre-built within the system contains multiple execution sub-models, each of which is optimized for a specific task. This structured design enables the system to quickly locate the most suitable execution sub-model upon receiving a user request and perform corresponding processing.

[0040] For example, when a user needs to generate a target text, the system will match the execution sub-model specifically for text generation. This model has been fine-tuned and has powerful natural language processing capabilities, and can automatically generate expected text content based on given input. If the user needs to review the report, the system will call the review-related sub-model, combine the review standards and processes, and complete the review task quickly and accurately. Similarly, for key information extraction and analysis needs, the system can extract the most important information in the report based on built-in algorithms and rules, and conduct in-depth analysis.

[0041] Furthermore, based on the instructions corresponding to the report processing requirement information triggered by the user, the corresponding sub-model is executed, and the report processing results are output and pushed to the user end.

[0042] Specifically, if the report processing requirement information triggered by the user is a target text generation requirement, the corresponding instruction is the customized prompt instruction information of the work requirement under the corresponding business scenario; if the report processing requirement information triggered by the user is a report writing requirement based on the standard paradigm, the corresponding instruction is the standard paradigm sequence number and report content information; if the report processing requirement information triggered by the user is a report review requirement or a report evaluation requirement, the corresponding instruction is the target report and the corresponding requirement instruction; if the report processing requirement information triggered by the user is a key information extraction requirement for the report or a key information analysis requirement for the report, the corresponding instruction is the target report and the corresponding requirement instruction.

[0043] In an embodiment of the present invention, when a user triggers different types of report processing requirements, the system generates corresponding instructions based on these requirements. For example, if the user's requirement is to generate a target text, the system generates customized prompt instruction information. This instruction is designed according to the work requirements of a specific business scenario to ensure that the generated text content can accurately reflect the subject and information required by the user. This flexible instruction generation mechanism allows the model to quickly adapt to different text generation tasks, improving user satisfaction and work efficiency.

[0044] Furthermore, if the user's needs are to write a report based on a standard format, the system will generate corresponding instructions based on the standard format serial number and report content information. Such instructions can ensure that the structure and content of the report meet industry standards, making the report not only professional, but also recognized in audits and compliance checks. By following a standardized process, the system effectively reduces errors and inconsistencies that may occur during manual writing.

[0045] Furthermore, when it comes to the need for report review and evaluation, the system will execute the processing of the target report and the corresponding requirement instructions. During this process, the system will automatically review the target report to ensure the integrity and consistency of the content. This not only improves the review efficiency, but also reduces the delays caused by manual review. At the same time, the processing instructions for report evaluation requirements can help users evaluate the quality of the report, provide optimization suggestions, and ensure that the final output report meets high standards.

[0046] Furthermore, when the requirements involve extracting or analyzing key information, the system will generate corresponding processing instructions based on the target report and the required instructions. These instructions enable the system to quickly identify and extract important data and information in the report, and conduct in-depth analysis and mining. This capability not only improves the efficiency of information processing, but also provides decision support for users, allowing them to better understand and respond to potential security risks.

[0047] Based on the solution of the present invention, by accurately responding to the report processing needs triggered by users, the system can achieve fast and efficient report generation and processing. Compared with the traditional manual generation process, this automated mechanism greatly shortens the report processing time, allowing users to obtain the required security reports faster and improving the user experience. The customized instruction generation mechanism improves the quality and accuracy of the report. Each type of demand can get a specially designed processing instruction to ensure that the generated report meets the actual business needs. This flexibility enables the system to adapt to the specific circumstances of different users and provide personalized services. The credibility and reliability of the report are improved by automating the review and evaluation process. Users no longer need to invest a lot of manpower for review, and the system can complete a comprehensive check of the report in a short time. This technical advantage not only improves work efficiency, but also reduces the risks caused by human errors, thereby providing more reliable support for network security management.

[0048] Preferably, the method also includes: each time report processing is performed, corresponding report processing information is retained; every predetermined period, new training samples are constructed based on all report processing information retained in the corresponding period, and model updates are performed on the corresponding sub-models constructed based on the new training samples.

[0049] In the embodiment of the present invention, after each report is processed, the system automatically stores relevant information, including the time when the report was generated, the content of the processing, the model and instructions used, and the user's feedback. The retention of this information not only helps to track the processing history of the report, but also provides data support for subsequent analysis. Over time, the accumulated processing information will form rich historical data, which is of great value in training and optimizing models.

[0050] Furthermore, at every predetermined period, the system will construct new training samples based on all report processing information retained in the corresponding period. These new samples will contain diverse data from different processing scenarios, including user feedback, report modification records, and model execution results. This process helps to improve the generalization ability of the model, so that it can better adapt to the ever-changing network security environment and user needs. Based on the new training samples, the constructed sub-models are updated. The model update process involves applying new data to the existing model and adjusting the model parameters and structure to optimize its performance. This dynamic update mechanism ensures that the model can continuously learn new information, adapt to the changing needs of users, and improve its performance in specific report processing tasks.

[0051] Embodiment 1: Execution process under the requirement of target text generation.

[0052] Based on the AI ​​model, we can intelligently analyze professional document knowledge such as project feasibility reports and security assessment reports, build vector data, perform pre-training and optimization, and then customize prompts according to the user's work needs in specific business scenarios, and finally automatically generate professional knowledge that meets the business content. Table 1 shows some application examples.

[0053] Table 1 Example of customized prompt for target text generation requirements

[0054]

[0055]

[0056] Embodiment 2: Execution process of report writing requirement based on standard paradigm.

[0057] Based on the AI ​​model, the structured analysis of existing feasibility reports, project bids, safety assessment reports, etc. is performed to output the corresponding text required for the corresponding chapters of the report. The text content is automatically organized, summarized, and typeset according to the paradigm structure specified by the technical report, and the technical report required by the user is output, thereby improving the efficiency of report preparation.

[0058] Example 3: Execution process of report review requirements and report evaluation requirements.

[0059] By utilizing natural language processing, AI Agent and other technologies, integrating the existing expert review knowledge base, and building AI automatic review logic and mechanisms based on the audit work logic, we can achieve automatic review of the core points, main content, format specifications and other contents of the report, and conduct a comprehensive score assessment of the report quality based on the problems found in the audit.

[0060] Embodiment 4: Execution process of the key information extraction requirement and the key information analysis requirement of the report.

[0061] Based on the standard paradigm requirements of reports, it automatically identifies key information in reports, extracts key contents of reports in batches, deeply explores the potential data value of reports of various categories, automatically summarizes key points and core contents of reports, realizes functions such as content summary, situation analysis, and situation forecasting, and realizes data visualization through charts and other means, so as to provide a basis for business optimization or accurate prevention and control of security risks.

[0062] Figure 2 1 is a system structure diagram of a security assessment report editing system based on multi-AIAgengt collaboration provided by an embodiment of the present invention. Figure 2 As shown, an embodiment of the present invention provides a security assessment report editing system based on multi-AIAgengt collaboration, and the system includes: a collection unit, used to collect historical security assessment reports, and perform vectorization processing on the historical security assessment reports to obtain training sample information; a model fine-tuning unit, used to fine-tune and train each pre-selected model based on the training sample information to obtain a security assessment report editing model; a model matching unit, used to match the corresponding execution sub-model in the security assessment report editing model in response to report processing requirement information triggered by a user; an execution unit, used to execute the corresponding sub-model based on the prompt instruction corresponding to the report processing requirement information triggered by the user, and output and push the report processing result to the user end.

[0063] Preferably, the fine-tuning training of each preselected model based on the training sample information to obtain the security assessment report editing model includes: performing data preprocessing on the training sample information to obtain standardized training samples; matching corresponding preselected models for various requirements based on user preset report processing requirements; based on the standardized training samples, performing fine-tuning training corresponding to each preselected model using cross-validation and / or hyperparameter tuning to obtain execution sub-models corresponding to report processing requirements; and merging the execution sub-models based on each report processing requirement to construct the security assessment report editing model.

[0064] An embodiment of the present invention also provides a computer-readable storage medium, which stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned multi-AIAgengt collaboration-based security assessment report editing and review method.

[0065] Those skilled in the art will understand that all or part of the steps in the method for implementing the above-mentioned embodiments can be completed by instructing the relevant hardware through a program, and the program is stored in a storage medium, including several instructions for making a single-chip microcomputer, a chip or a processor (processor) perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0066] The optional embodiments of the present invention are described in detail above in conjunction with the accompanying drawings. However, the embodiments of the present invention are not limited to the specific details in the above embodiments. Within the technical concept of the embodiments of the present invention, the technical scheme of the embodiments of the present invention can be subjected to a variety of simple modifications, and these simple modifications all belong to the protection scope of the embodiments of the present invention. It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the embodiments of the present invention will not further describe various possible combinations.

[0067] In addition, various embodiments of the present invention may be arbitrarily combined, and as long as they do not violate the concept of the embodiments of the present invention, they should also be regarded as the contents disclosed in the embodiments of the present invention.

Claims

1. A method for compiling and reviewing information security assessment reports based on multi-AIAgengt collaboration, characterized in that: The method comprises: Collect historical security assessment reports, and perform vectorization processing on the historical security assessment reports to obtain training sample information; Fine-tune and train each pre-selected model based on the training sample information to obtain a security assessment report editing model; In response to the report processing requirement information triggered by the user, the corresponding execution sub-model in the security assessment report editing model is matched; Based on the instructions corresponding to the report processing requirement information triggered by the user, the corresponding sub-model is executed, and the report processing results are output and pushed to the user end.

2. The method according to claim 1, characterized in that The performing vectorization processing on the historical security assessment report to obtain training sample information includes: The historical security assessment reports are classified based on data types, and the vectorization methods corresponding to each classified report set are matched based on the classification results; The vectorization method includes any one or more of the TF-IDF method, the Word2Vec method, and the BERT method; The matching-based vectorization method performs vectorization processing on the corresponding classification report set to obtain training sample information.

3. The method according to claim 2, characterized in that Before performing the vectorization processing of the corresponding classification report set, the method further includes: The historical security assessment reports are pushed to the user end for the user to calibrate the report.

4. The method according to claim 1, characterized in that: The fine-tuning training of each pre-selected model based on the training sample information to obtain the security assessment report editing model includes: Performing data preprocessing on the training sample information to obtain standardized training samples; Based on the report processing requirements preset by the user, the corresponding pre-selected models are matched to various requirements; Based on the standardized training samples, fine-tune the models corresponding to the pre-selected models by cross-validation and / or hyperparameter tuning to obtain the execution sub-models corresponding to the report processing requirements; The execution sub-models based on the processing requirements of each report are merged to build the security assessment report editing and review model.

5. The method according to claim 1, characterized in that User-triggered report processing requirements include: Any one or more of the requirements for target text generation, report writing based on standard paradigms, report review requirements, report evaluation requirements, key information extraction requirements for reports, and key information analysis requirements for reports.

6. The method according to claim 5, characterized in that If the report processing requirement information triggered by the user is a target text generation requirement, the corresponding instruction is a customized prompt instruction information corresponding to the work requirement in the business scenario; If the report processing requirement information triggered by the user is a report writing requirement based on a standard format, the corresponding instruction is the standard format number and report content information; If the report processing requirement information triggered by the user is a report review requirement or a report evaluation requirement, the corresponding instructions are the target report and the corresponding requirement instructions; If the report processing requirement information triggered by the user is a key information extraction requirement for the report or a key information analysis requirement for the report, the corresponding instructions are the target report and the corresponding requirement instructions.

7. The method according to claim 1, characterized in that The method further comprises: Each time a report is processed, the corresponding report processing information is retained; At every predetermined period, a new training sample is constructed based on all report processing information retained in the corresponding period, and a model update is performed on each corresponding sub-model constructed based on the new training sample.

8. A security assessment report editing and review system based on multi-AIAgengt collaboration, characterized in that: The system comprises: A collection unit, used to collect historical security assessment reports, and perform vectorization processing on the historical security assessment reports to obtain training sample information; A model fine-tuning unit, used to perform fine-tuning training on each pre-selected model based on the training sample information to obtain a security assessment report editing model; A model matching unit, configured to match a corresponding execution sub-model in the security assessment report editing model in response to report processing requirement information triggered by a user; The execution unit is used to execute the corresponding sub-model based on the prompt instruction corresponding to the report processing requirement information triggered by the user, and output and push the report processing result to the user end.

9. The system according to claim 8, characterized in that The fine-tuning training of each pre-selected model based on the training sample information to obtain the security assessment report editing model includes: Performing data preprocessing on the training sample information to obtain standardized training samples; Based on the report processing requirements preset by the user, the corresponding pre-selected models are matched to various requirements; Based on the standardized training samples, fine-tune the models corresponding to the pre-selected models by cross-validation and / or hyperparameter tuning to obtain the execution sub-models corresponding to the report processing requirements; The execution sub-models based on the processing requirements of each report are merged to build the security assessment report editing and review model.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the method for compiling and reviewing information security assessment reports based on multi-AIAgengt collaboration as described in any one of claims 1 to 7.

Images