Method, device, equipment, medium and product for detecting quality of network security insurance waiting report
Through the training of the quality detection model, the peer-to-peer guarantee report is automatically detected, which solves the problem of poor detection accuracy caused by manual review, and realizes efficient and accurate detection of the peer-to-peer guarantee report.
Patent Information
- Application Number
- CN202510205955.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-25
- Publication Date
- 2025-08-01
AI Technical Summary
In the prior art, the quality inspection of guaranteed reports relies on manual review, resulting in poor inspection accuracy and the problem in the report cannot be fully discovered.
By training the quality detection model, LoRA fine-tuning technology is used to automatically detect network security and other guarantee reports, calculate the matching, completeness and recognition, and improve detection accuracy.
Accurate audit of peer-to-peer guarantee reports has been achieved, improving the accuracy of detection and the accuracy of evaluation results.
Smart Images
Figure CN120408180A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of artificial intelligence technology, and particularly to a method, device, equipment, medium and product for detecting the quality of network security equal protection reports. Background Art
[0002] A network security level protection assessment report (i.e., an equal protection report) is an official document formed after a network system is subjected to a level protection assessment according to relevant network security standards. The equal protection report aims to evaluate the security protection level of the network system and propose corresponding security protection measures and suggestions to ensure that the network system can maintain a certain security level when facing various security threats.
[0003] During the level protection assessment process, the assessors detect relevant risk assessment items according to the requirements of relevant network security standards and record and organize the assessment results in the corresponding chapters of the equal protection report. However, due to the large number of assets involved in the system under test, the large number of equal protection assessment items, and the parallel execution of multiple projects, etc., there may be omissions during the recording process by the assessors, resulting in incomplete, contradictory, and incorrect records in the equal protection report.
[0004] Therefore, it is necessary to detect the quality of the equal protection report. Usually, an equal protection report has hundreds of pages of content, and the assessment agency will form a special team of senior assessors to detect the quality of the report. However, due to reasons such as time, energy, and focus of attention, the assessors still cannot completely discover all the problems existing in the report, resulting in various quality problems in the officially issued equal protection report. It can be seen that the equal protection report cannot be accurately reviewed through manual review, resulting in poor accuracy of the detection of the equal protection report. Summary of the Invention
[0005] The purpose of the present application is to provide a method, device, equipment, medium and product for detecting the quality of network security equal protection reports, which can improve the accuracy of detecting equal protection reports.
[0006] To achieve the above purpose, the present application provides the following solutions:
[0007] In a first aspect, the present application provides a method for detecting the quality of network security equal protection reports, including:
[0008] Inputting a network security equal protection report into a quality detection model to obtain a quality detection result output by the quality detection model;
[0009] The quality detection model is trained through the following steps:
[0010] Obtain a training data set; wherein, the training data set includes a plurality of training data groups, and each training data group includes an input training data and an output result data corresponding to the input training data; the input training data includes a historical network security equal protection report;
[0011] Input the input training data into the base large language model, and perform LoRA fine-tuning on the trainable weight parameters in the base large language model to obtain a quality detection model and training result data;
[0012] Based on the output result data and the training result data, calculate evaluation data. In the case where the evaluation data does not meet the preset evaluation index, repeat the following steps:
[0013] Perform LoRA fine-tuning on the trainable weight parameters in the quality detection model to obtain an adjusted quality detection model; input the input training data into the adjusted quality detection model to obtain the training result data output by the quality detection model; and calculate evaluation data based on the output result data and the training result data;
[0014] Until the evaluation data meets the preset evaluation index, it is determined that the training of the quality detection model is completed.
[0015] Optionally, the obtaining of the training data set specifically includes:
[0016] Obtain a plurality of historical network security equal protection reports;
[0017] Determine the indication information corresponding to each historical network security equal protection report;
[0018] Based on a plurality of historical network security equal protection reports and a plurality of indication information, obtain a plurality of input training data; wherein, one input training data only includes a target historical network security equal protection report and the indication information corresponding to the target historical network security equal protection report;
[0019] Analyze each of the plurality of input training data to obtain the output result data corresponding to each input training data;
[0020] Determine the respective input training data and the output result data corresponding to each input training data together as the training data set.
[0021] Optionally, the evaluation data at least includes a matching degree, a completeness degree, and a recognition degree. The calculating of the evaluation data based on the output result data and the training result data specifically includes:
[0022] Determine the initial matching degree between the output result data and the training result data;
[0023] Determine the standard result length of the output result data and determine the training result length of the training result data;
[0024] Based on the initial matching degree, the standard result length, and the training result length, calculate the matching degree;
[0025] Determine the standard reference abstract in the output result data;
[0026] Determine the abstract to be evaluated in the training result data;
[0027] Use the standard reference abstract and the abstract to be evaluated to calculate the integrity;
[0028] Use the output result data and the training result data to determine the recognition degree.
[0029] Optionally, the calculation formula for the matching degree is:
[0030]
[0031] Among them, Bleu represents the matching degree, BP is the penalty factor obtained based on the standard result length and the training result length, N = 4, W n represents the nth byte segment in the training result data, P n represents the initial matching degree.
[0032] Optionally, the calculation formula for the penalty factor BP is:
[0033]
[0034] Among them, L c represents the standard result length, L s represents the training result length.
[0035] Optionally, the calculation formula for the integrity is:
[0036]
[0037] Among them, ROUGE represents the integrity, S represents the standard reference abstract, References represents multiple standard reference abstracts in multiple output result data in the training dataset, match(gram-n) represents the number of target byte segments that appear in both the abstract to be evaluated and the standard reference abstract, and count(gram-n) represents the total number of the target byte segments in the standard reference abstract.
[0038] In a second aspect, the present application provides a device for detecting the quality of a network security equal protection report, which is used to input a network security equal protection report into a quality detection model and obtain a quality detection result output by the quality detection model;
[0039] The quality detection model is constructed by a training device for the quality detection model. The training device for the quality detection model includes:
[0040] An acquisition unit, configured to acquire a training data set; wherein, the training data set includes a plurality of training data groups, and each training data group includes an input training data and an output result data corresponding to the input training data; the input training data includes historical network security equal protection reports;
[0041] An input unit, configured to input the input training data into a base large language model and perform LoRA fine-tuning on the trainable weight parameters in the base large language model to obtain a quality detection model and training result data;
[0042] A calculation unit, configured to calculate evaluation data based on the output result data and the training result data. In the case that the evaluation data does not meet a preset evaluation index, the following steps are repeatedly executed:
[0043] A fine-tuning unit, configured to perform LoRA fine-tuning on the trainable weight parameters in the quality detection model to obtain an adjusted quality detection model; and input the input training data into the adjusted quality detection model to obtain training result data output by the quality detection model; and calculate evaluation data based on the output result data and the training result data;
[0044] A determination unit, configured to determine that the training of the quality detection model ends until the evaluation data meets the preset evaluation index.
[0045] In a third aspect, the present application provides a computer device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor. The processor executes the computer program to implement the steps of the network security equal protection report quality detection method described in any one of the above.
[0046] In a fourth aspect, the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the network security equal protection report quality detection method described in any one of the above are implemented.
[0047] In a fifth aspect, the present application provides a computer program product, including a computer program. When the computer program is executed by a processor, the steps of the network security equal protection report quality detection method described in any one of the above are implemented.
[0048] According to the specific embodiments provided in this application, the following technical effects are disclosed in this application:
[0049] This application provides a method, device, equipment, medium and product for detecting the quality of network security equal protection reports. Through the trained quality detection model, the quality of the input network security equal protection report can be detected, and the quality detection result output by the quality detection model can be obtained. It can be seen that by accurately auditing the network security equal protection report through the quality detection model, the accuracy of the equal protection report detection can be improved. Description of the Drawings
[0050] In order to more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings in the following description are only some embodiments of this application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.
[0051] Figure 1 It is a schematic flowchart of a method for training a quality detection model in an embodiment of this application;
[0052] Figure 2 It is a schematic diagram of the functional modules of a quality detection model provided in an embodiment of this application;
[0053] Figure 3 It is a schematic diagram of the functional modules of a training device for a quality detection model provided in an embodiment of this application;
[0054] Figure 4 It is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Embodiments
[0055] The following will clearly and completely describe the technical solutions in the embodiments of this application with reference to the drawings in the embodiments of this application. Obviously, the described embodiments are only some embodiments of this application, rather than all embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of this application.
[0056] To make the above objects, features, and advantages of this application more obvious and understandable, the following will further describe this application in detail with reference to the drawings and specific embodiments.
[0057] In the embodiments of this application, the method for detecting the quality of network security equal protection reports can be implemented through a pre-trained quality detection model, specifically as follows:
[0058] Input the network security equal protection report into the quality inspection model to obtain the quality inspection result output by the quality inspection model.
[0059] The quality inspection model is trained through the following steps. In an exemplary embodiment, as Figure 1 shown, a training method for the quality inspection model is provided. This method is executed by a computer device, which can be specifically executed by a computer device such as a terminal or a server alone, or jointly executed by a terminal and a server. In the embodiments of the present application, the following steps 101 to 105 are included. Among them:
[0060] Step 101, obtain a training data set.
[0061] In the embodiments of the present application, the training data set includes multiple training data groups, and each training data group includes an input training data and an output result data corresponding to the input training data; the input training data includes historical network security equal protection reports.
[0062] In the embodiments of the present application, the network security equal protection report usually needs to record the evaluation results of risk evaluation items. Normally, high-risk evaluation items should be recorded in Chapter 4 of the report; non-high-risk evaluation items should be recorded in Chapter 5 of the report. Whether it belongs to a high-risk evaluation item needs to be determined according to the "Guidelines for Determining High Risks in Network Security Level Protection Evaluation" (hereinafter referred to as the "Determination Guidelines").
[0063] For example, when detecting whether there is a problem with the recording of a risk evaluation item, triple judgments will be made:
[0064] 1. Determine the chapter to which it should belong according to the "Determination Guidelines";
[0065] 2. Determine whether there is a record in the corresponding chapter;
[0066] 3. Determine whether the record is complete.
[0067] After the triple judgments are completed, the category corresponding to the risk evaluation item will be output, that is, the detection conclusion.
[0068] There are six categories of detection conclusions, as follows:
[0069] The first category: It should be recorded in Chapter 5, but there is no record;
[0070] The second category: It should be recorded in Chapter 5, the risk scenario has been analyzed and recorded, but the record is incomplete;
[0071] The third category: It should be recorded in Chapter 5, the risk scenario has been analyzed and recorded, and the record is complete;
[0072] The fourth category: It should be recorded in Chapter 4, but there is no record;
[0073] Category 5: It should be recorded in Chapter 4. The risk scenarios have been analyzed and recorded, but the records are incomplete;
[0074] Category 6: It should be recorded in Chapter 4. The risk scenarios have been analyzed and recorded, and the records are complete.
[0075] As an alternative implementation, the method of obtaining the training dataset in step 101 may include:
[0076] Obtain multiple historical network security equal protection reports;
[0077] Determine the indication information corresponding to each historical network security equal protection report;
[0078] Based on the multiple historical network security equal protection reports and the multiple indication information, obtain multiple input training data; wherein, one input training data only includes one target historical network security equal protection report and the indication information corresponding to the target historical network security equal protection report;
[0079] Analyze each of the multiple input training data to obtain the output result data corresponding to each input training data;
[0080] Determine the input training data and the output result data corresponding to each input training data together as the training dataset.
[0081] Among them, by implementing this implementation, both the historical network security equal protection reports and the indication information corresponding to the historical network security equal protection reports can be used as training data. At the same time, the output result data obtained by analyzing the historical network security equal protection reports can also be used as training data, improving the comprehensiveness of the training data in the training dataset.
[0082] In the embodiments of the present application, the indication information corresponding to the historical network security equal protection report is used to represent the operations that need to be performed on the historical network security equal protection report.
[0083] Step 102: Input the input training data into the base large language model, and perform LoRA fine-tuning on the trainable weight parameters in the base large language model to obtain a quality detection model and training result data.
[0084] Step 103: Calculate evaluation data based on the output result data and the training result data.
[0085] In the embodiments of the present application, the evaluation data at least includes matching degree, integrity, and recognition degree.
[0086] As an alternative implementation, the method of calculating the evaluation data based on the output result data and the training result data in step 103 may include:
[0087] Determine the initial matching degree between the output result data and the training result data;
[0088] Determine the standard result length of the output result data and the training result length of the training result data;
[0089] Calculate the matching degree based on the initial matching degree, the standard result length, and the training result length;
[0090] Determine the standard reference abstract in the output result data;
[0091] Determine the abstract to be evaluated in the training result data;
[0092] Calculate the integrity using the standard reference abstract and the abstract to be evaluated;
[0093] Determine the recognition degree using the output result data and the training result data.
[0094] Among them, by implementing this implementation method, the output result data and the training result data can be evaluated multiple times to calculate the matching degree, integrity, and recognition degree, thereby improving the accuracy of the evaluation result.
[0095] Optionally, the calculation formula for the matching degree is:
[0096]
[0097] Among them, Bleu represents the matching degree, BP is the penalty factor obtained based on the standard result length and the training result length, N = 4, W n represents the nth byte segment in the training result data, P n represents the initial matching degree.
[0098] In the embodiments of the present application, the matching degree can be the Bleu-4 index, and the Bleu-4 index calculates the performance of the model according to the matching of words, phrases, and n-grams (sequences of byte segments with length n, where n is at most 4) between the training result data and the output result data.
[0099] Optionally, the calculation formula for the penalty factor BP is:
[0100]
[0101] Among them, L c represents the standard result length, L s represents the training result length.
[0102] The Bleu-4 metric typically ranges from 0 to 1, where 1 represents a perfect match.
[0103] Optionally, the formula for calculating the completeness is:
[0104]
[0105] Where ROUGE represents the completeness, S represents the standard reference summary, References represents multiple standard reference summaries in the training dataset output results data, match(gram-n) represents the number of target byte fragments that appear in both the summary to be evaluated and the standard reference summary, and count(gram-n) represents the total number of the target byte fragments in the standard reference summary.
[0106] In the embodiments of the present application, the completeness can be Rouge-1. Rouge-1 mainly focuses on whether the summary to be evaluated captures the information in the standard reference summary, and emphasizes the integrity of covering the content and information of the standard reference summary. The Rouge-1 score also ranges from 0 to 1, and the closer it is to 1, the higher the quality.
[0107] In the embodiments of the present application, the recognition rate can include accuracy, precision, recall, and balanced F-score (F1 value). Accuracy, precision, recall, and F1 value are used to evaluate the recognition performance of the quality detection model for different output categories in the quality detection task.
[0108] The calculation methods of accuracy, precision, recall, and F1 value are shown in formulas (7)-(10).
[0109]
[0110] Where TP: the output result data is determined to be a positive sample, and the training result data is also a positive sample; FP: the output result data is determined to be a positive sample, but the training result data is a negative sample; TN: the output result data is determined to be a negative sample, and the training result data is also a negative sample; FN: the output result data is determined to be a negative sample, but the training result data is a positive sample.
[0111] Step 104, in the case that the evaluation data does not meet the preset evaluation metrics, perform LoRA fine-tuning on the trainable weight parameters in the quality detection model to obtain an adjusted quality detection model; and execute steps 102 to 103.
[0112] In the embodiments of the present application, the preset evaluation metrics are determined according to the values of the matching degree, completeness, and recognition rate.
[0113] Please refer to Figure 2 as well, Figure 2 which is a schematic diagram of the functional modules of a quality detection model provided by an embodiment of this application. Among them, the quality detection model includes a pre-trained model, a low-rank decomposition matrix A, and a low-rank decomposition matrix B. The ranks of the low-rank decomposition matrix A and the low-rank decomposition matrix B are r, and the dimension of the input x input into the quality detection model is d-dimensional. It can be input into the pre-trained model and the low-rank decomposition matrix A simultaneously. The pre-trained model processes the input x and outputs the first training result data, and the dimension of this first training result data is also d-dimensional; the input x enters the low-rank decomposition matrix A, and the dimension of the input x is reduced to r-dimensional through the low-rank decomposition matrix A, and then the dimension of the input x that has been reduced to r-dimensional is raised back to d-dimensional through the low-rank decomposition matrix B, obtaining the second training result data output by the low-rank decomposition matrix B; finally, the first training result data and the second training result data can be combined to obtain the final training result data (i.e., the output h).
[0114] Among them, the pre-trained weight parameter matrix in the pre-trained model is W0 ∈ R d×d , and the pre-trained weight parameters in the pre-trained model are not fine-tuned; the trainable weight parameters of the low-rank decomposition matrix A and the low-rank decomposition matrix B are fine-tuned. The parameter matrix to be updated is ΔW = BA, where B ∈ R d×r , A ∈ R r×d . Then the final output data of the quality detection model is the sum of two parts of output data:
[0115] h = W0x + ΔWx = W0x + BAx
[0116] The objective function after fine-tuning is shown in the following formula:
[0117]
[0118] Among them, Φ is the optimization parameter of the model, and P Φ (y|x) is a general multi-task learning machine, where x and y correspond to the input and output of the large model. is the learning machine for fine-tuning, and the parameter Θ is the main part optimized by the large model, thereby generating the fine-tuned model; is the data composed of sequences; t is the length of the model output. LLaMA-Factory can be selected as the fine-tuning tool. The selected base model is glm-4-9b-1m-chat. The hyperparameter configuration for fine-tuning is: number of training steps: 40; learning rate: 5e-5; maximum gradient norm: 1; rank of LoRA: 32; truncation length: 1024; batch size: 2; optimizer: Adam.
[0119] Step 105. Until the evaluation data meets the preset evaluation criteria, it is determined that the training of the quality detection model is completed.
[0120] By implementing the above steps 101 to 105, the quality detection model can accurately review the network security equal protection report, which can improve the accuracy of the equal protection report detection. In addition, the present application can also improve the comprehensiveness of the training data in the training dataset. In addition, the present application can also improve the accuracy of the evaluation results.
[0121] The present application also provides an application scenario that applies the above network security equal protection report quality detection method. Specifically: The network security equal protection report quality detection method provided in this embodiment can be applied to the quality detection scenario of network security equal protection reports. The network security equal protection report is input into the quality detection model, and the quality detection model performs quality detection on the network security equal protection report, and the detection result is output through the quality detection model. The network security equal protection report quality detection method provided in this embodiment belongs to the quality detection link of the network security equal protection report.
[0122] Based on the same inventive concept, an embodiment of the present application also provides a network security equal protection report quality detection device for implementing the above-mentioned network security equal protection report quality detection method. The implementation solutions provided by this device to solve problems are similar to those recorded in the above method. Therefore, the specific limitations in one or more embodiments of the network security equal protection report quality detection device provided below can refer to the limitations on the network security equal protection report quality detection method in the above text, and will not be repeated here.
[0123] The network security equal protection report quality detection device is used to input the network security equal protection report into the quality detection model to obtain the quality detection result output by the quality detection model.
[0124] The quality detection model is constructed by the training device of the quality detection model. In an exemplary embodiment, as Figure 3 shown, a training device for the quality detection model is provided, including:
[0125] An acquisition unit 301, configured to acquire a training dataset; wherein, the training dataset includes a plurality of training data groups, and each training data group includes an input training data and an output result data corresponding to the input training data; the input training data includes historical network security equal protection reports;
[0126] An input unit 302, configured to input the input training data into the base large language model and perform LoRA fine-tuning on the trainable weight parameters in the base large language model to obtain a quality detection model and training result data;
[0127] The calculation unit 303 is configured to calculate evaluation data based on the output result data and the training result data. In the case where the evaluation data does not meet the preset evaluation criteria, the following steps are repeatedly executed:
[0128] The fine-tuning unit 304 is configured to perform LoRA fine-tuning on the trainable weight parameters in the quality detection model to obtain an adjusted quality detection model; input the input training data into the adjusted quality detection model to obtain the training result data output by the quality detection model; and calculate evaluation data based on the output result data and the training result data;
[0129] The determination unit 305 is configured to determine that the training of the quality detection model is completed until the evaluation data meets the preset evaluation criteria.
[0130] As an alternative implementation, the manner in which the acquisition unit 301 acquires the training data set may specifically be:
[0131] Acquire multiple historical network security equal protection reports;
[0132] Determine the indication information corresponding to each historical network security equal protection report;
[0133] Based on multiple historical network security equal protection reports and multiple indication information, obtain multiple input training data; wherein, one input training data only includes one target historical network security equal protection report and the indication information corresponding to the target historical network security equal protection report;
[0134] Analyze multiple input training data respectively to obtain the output result data corresponding to each input training data;
[0135] Jointly determine each input training data and the output result data corresponding to each input training data as the training data set.
[0136] Among them, by implementing this implementation, both the historical network security equal protection report and the indication information corresponding to the historical network security equal protection report can be used as training data. At the same time, the output result data obtained by analyzing the historical network security equal protection report can also be used as training data, improving the comprehensiveness of the training data in the training data set.
[0137] As an alternative implementation, the evaluation data at least includes a matching degree, a completeness degree, and an identification degree. The manner in which the calculation unit 303 calculates the evaluation data based on the output result data and the training result data may specifically be:
[0138] Determine the initial matching degree between the output result data and the training result data;
[0139] Determine the standard result length of the output result data and determine the training result length of the training result data;
[0140] Based on the initial matching degree, the standard result length, and the training result length, calculate the matching degree;
[0141] Determine the standard reference abstract in the output result data;
[0142] Determine the abstract to be evaluated in the training result data;
[0143] Use the standard reference abstract and the abstract to be evaluated to calculate the integrity;
[0144] Use the output result data and the training result data to determine the recognition degree.
[0145] Among them, by implementing this implementation manner, the output result data and the training result data can be evaluated multiple times, and the matching degree, integrity, and recognition degree can be calculated, thereby improving the accuracy of the evaluation result.
[0146] Optionally, the calculation formula for the matching degree is:
[0147]
[0148] Among them, Bleu represents the matching degree, BP is the penalty factor obtained based on the standard result length and the training result length, N = 4, W n represents the nth byte segment in the training result data, P n represents the initial matching degree.
[0149] Optionally, the calculation formula for the penalty factor BP is:
[0150]
[0151] Among them, L c represents the standard result length, L s represents the training result length.
[0152] Optionally, the calculation formula for the integrity is:
[0153]
[0154] Among them, ROUGE represents the integrity, S represents the standard reference summary, References represents multiple standard reference summaries in the output result data in the training dataset, match(gram-n) represents the number of target byte segments that appear in both the summary to be evaluated and the standard reference summary, and count(gram-n) represents the total number of the target byte segments in the standard reference summary.
[0155] Implementing the above embodiments, the accurate review of the network security equal protection report by the quality detection model can improve the accuracy of the detection of the equal protection report. In addition, the present application can also improve the comprehensiveness of the training data in the training dataset. In addition, the present application can also improve the accuracy of the evaluation results.
[0156] In an exemplary embodiment, a computer device is provided. The computer device can be a server or a terminal, and its internal structure diagram can be as Figure 4 shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store network security equal protection report quality detection data. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, it implements a method for detecting the quality of a network security equal protection report.
[0157] Those skilled in the art can understand that Figure 4 the structure shown in
[0158] is only a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.
[0159] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, which, when executed by a processor, implements the steps in the above method embodiments.
[0160] In an exemplary embodiment, a computer program product is provided, including a computer program, which, when executed by a processor, implements the steps in the above method embodiments.
[0161] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.
[0162] Those of ordinary skill in the art can understand that all or part of the processes in the above method embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above method embodiments. Among them, any reference to a memory, database, or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memories. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.
[0163] In each of the embodiments provided in this application, the database involved may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., without limitation. In each of the embodiments provided in this application, the processor may be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., without limitation.
[0164] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.
[0165] In this article, specific examples are used to elaborate on the principle and implementation manner of this application. The description of the above embodiments is only used to help understand the method and its core idea of this application; at the same time, for those of ordinary skill in the art, according to the idea of this application, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to this application.
Claims
1. A method for detecting the quality of network security equal protection reports, characterized in that, The method for quality detection of network security equal protection reports includes: Inputting a network security equal protection report into a quality detection model to obtain a quality detection result output by the quality detection model; The quality detection model is trained through the following steps: Obtaining a training data set; wherein, the training data set includes multiple training data groups, and each training data group includes an input training data and an output result data corresponding to the input training data; the input training data includes historical network security equal protection reports; Inputting the input training data into a base large language model, and performing LoRA fine-tuning on the trainable weight parameters in the base large language model to obtain a quality detection model and training result data; Based on the output result data and the training result data, calculating evaluation data, and in the case that the evaluation data does not meet the preset evaluation index, repeating the following steps: Performing LoRA fine-tuning on the trainable weight parameters in the quality detection model to obtain an adjusted quality detection model; inputting the input training data into the adjusted quality detection model to obtain training result data output by the quality detection model; and calculating evaluation data based on the output result data and the training result data; Until the evaluation data meets the preset evaluation index, it is determined that the training of the quality detection model is completed.
2. The quality detection method for network security equal protection reports according to claim 1, characterized in that, The obtaining of the training data set specifically includes: Obtaining multiple historical network security equal protection reports; Determining the indication information corresponding to each historical network security equal protection report; Based on multiple historical network security equal protection reports and multiple indication information, obtaining multiple input training data; wherein, one input training data only includes one target historical network security equal protection report and the indication information corresponding to the target historical network security equal protection report; Analyzing multiple input training data respectively to obtain output result data corresponding to each input training data; Jointly determining each input training data and the output result data corresponding to each input training data as the training data set.
3. The network security equal-level protection report quality detection method according to claim 1, characterized in that The evaluation data at least includes matching degree, integrity and recognition degree. The calculating of the evaluation data based on the output result data and the training result data specifically includes: Determining the initial matching degree between the output result data and the training result data; Determining the standard result length of the output result data and determining the training result length of the training result data; Calculating the matching degree based on the initial matching degree, the standard result length and the training result length; Determining the standard reference summary in the output result data; Determining the summary to be evaluated in the training result data; Calculating the integrity using the standard reference summary and the summary to be evaluated; Determining the recognition degree using the output result data and the training result data.
4. The method for detecting the quality of the network security equal protection report according to claim 3, characterized in that, The calculation formula of the matching degree is: Among them, Bleu represents the matching degree, BP is a penalty factor obtained based on the standard result length and the training result length, N = 4, and W n represents the nth byte segment in the training result data, and P n represents the initial matching degree.
5. The network security equal-level protection report quality detection method according to claim 4, characterized in that The calculation formula of the penalty factor BP is: Among them, L c represents the length of the standard result, and L s represents the length of the training result.
6. The quality detection method for network security equal protection reports according to claim 3, characterized in that The calculation formula of the integrity is: Among them, ROUGE represents the integrity, S represents the standard reference summary, References represents multiple standard reference summaries in the output result data of the training dataset, match(gram-n) represents the number of target byte segments that appear in both the summary to be evaluated and the standard reference summary, and count(gram-n) represents the total number of the target byte segments in the standard reference summary.
7. A device for detecting the quality of a network security equal protection report, characterized in that, The network security equal protection report quality detection device is used to input a network security equal protection report into a quality detection model and obtain a quality detection result output by the quality detection model. The quality detection model is constructed by a training device of the quality detection model. The training device of the quality detection model includes: An acquisition unit, configured to acquire a training dataset; wherein, the training dataset includes multiple training data groups, and each training data group includes an input training data and output result data corresponding to the input training data; the input training data includes historical network security equal protection reports. An input unit, configured to input the input training data into a base large language model and perform LoRA fine-tuning on the trainable weight parameters in the base large language model to obtain a quality detection model and training result data. A calculation unit, configured to calculate evaluation data based on the output result data and the training result data. In the case where the evaluation data does not meet a preset evaluation index, the following steps are repeatedly executed: A fine-tuning unit, configured to perform LoRA fine-tuning on the trainable weight parameters in the quality detection model to obtain an adjusted quality detection model; input the input training data into the adjusted quality detection model to obtain training result data output by the quality detection model; and calculate evaluation data based on the output result data and the training result data. A determination unit, configured to determine that the training of the quality detection model ends until the evaluation data meets the preset evaluation index.
8. A computer device, comprising: A memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the network security equal protection report quality detection method according to any one of claims 1-6.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the network security equal protection report quality detection method according to any one of claims 1-6.
10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the network security equal protection report quality detection method according to any one of claims 1-6.