A governance method for large model data leakage

Abstract

The present invention relates to the field of data governance, and in particular to a governance method for large model data leakage. The present invention establishes a characteristic behavior database for the model training terminal based on the regular characteristics of instruction response data screening. When the model training terminal is accessed, the behavior fitting characterization coefficient corresponding to the model training terminal is determined based on the comparison between the instruction response data collected in real time and the data in the characteristic behavior database, the characteristic fitting category of the current model training terminal is identified, and the interaction of the model training terminal is controlled in time. Under the weak characteristic fitting category, the data transmission regularity of the normal characteristic port is verified, and it is determined whether to close the corresponding normal characteristic port. The present invention considers the regularity of the instruction response data corresponding to the model training terminal, timely distinguishes the characteristic fitting category, and intervenes in the interaction of the model training terminal by characterizing the abnormality of the access behavior, thereby improving the security of the model training terminal and reducing the risk of data leakage.

Description

Technical Field

[0001] The present invention relates to the field of data governance, and in particular to a governance method for large model data leakage. Background Art

[0002] With the development of computer technology and Internet technology, data models under different architectures are widely used in various fields and are widely connected to terminals during the application process. In addition, various terminals, especially those used for model training, may contain a large amount of sensitive data. Therefore, terminal security is also crucial. Relevant terminal security and technologies to prevent data leakage are gradually being paid attention to.

[0003] For example, the patent application publication number: CN114676456A discloses a data privacy protection method, device and storage medium based on edge computing, which relates to the field of information security technology. The method includes: in a federated machine learning mechanism, obtaining multiple training sub-models, wherein each training sub-model is obtained by training the initial model according to multiple original data by different terminal users, wherein the multiple original data are different terminal user data collected; according to the distributed hash table deployed on the edge node, each training sub-model is associated, so that the corresponding parameters between each training sub-model are interactively shared, and the shared model parameters are obtained; according to the model parameters, the target training sub-model is trained in the target edge device to obtain the target training model; according to the target training model, each original data is detected whether there is a risk of data leakage. Through this application, the problem of high risk of data leakage in related technologies is solved.

[0004] However, the following problems still exist in the prior art. When different terminals are accessed by authorized visitors, due to differences in habits, the data interaction of the terminals when accessed and the instructions of the corresponding instruction input devices often have certain regularities in the data dimension. In the prior art, the above regularities are not considered to timely identify abnormal accesses, resulting in low terminal security and easy data leakage. Summary of the invention

[0005] In order to solve the problem that when different terminals are accessed by authorized visitors, the data interaction of the terminals and the instructions of the corresponding instruction input devices often have certain regularity in the data dimension due to differences in habits, and the prior art does not consider timely identification of abnormal access based on regularity, resulting in low terminal security and easy data leakage. The present invention provides a large model data leakage management method, which includes:

[0006] Obtaining command response data within a historical period of the model training terminal, including the amount of data interaction between the model training terminal and each port and the command trigger target of the corresponding command input device;

[0007] Screening regular features based on the instruction response data to establish a characteristic behavior database for the model training terminal, wherein the regular features include characteristic port combinations and characteristic function component combinations;

[0008] In response to the model training terminal being accessed, the instruction response data of the model training terminal during the acquisition period is compared with the data in the characteristic behavior database;

[0009] Determine the behavior fit representation coefficient corresponding to the model training terminal according to the comparison result to identify the feature fit category of the current model training terminal;

[0010] Based on the characteristics of the current access behavior, the category control model is trained to control the data interaction of the terminal, including:

[0011] Calling the characteristic behavior database to identify the normal characteristic port, verifying the data transmission rule of the normal characteristic port, determining whether to close the corresponding normal characteristic port, and synchronously sending verification requirement information to the model training terminal to determine whether to issue an early warning signal based on the feedback information of the model training terminal;

[0012] Alternatively, the interaction data of the model training terminal and the instruction data of the corresponding instruction input device within a continuous collection period are used to determine the feature matching category of the model training terminal.

[0013] Further, the process of screening the characteristic port combination based on the instruction response data includes:

[0014] Determine a first conditional probability that a model training terminal interacts with different port combinations during a number of accesses;

[0015] Determine a characteristic port combination based on the first conditional probability corresponding to each port combination, record the average value of the data interaction amount of each port in the characteristic port combination during data interaction, and store it in a characteristic behavior database;

[0016] If the first conditional probability corresponding to the port combination is greater than a predetermined port combination conditional probability threshold, the port combination is determined to be a characteristic port combination.

[0017] Further, the process of screening the characteristic function component combination based on the instruction response data includes:

[0018] Obtaining a second conditional probability that each combination of functional components is triggered during a number of accesses to the model training terminal;

[0019] Determine a characteristic function component combination based on a second conditional probability corresponding to the function component combination, and record a trigger frequency of each function component in the characteristic function component combination;

[0020] If the second conditional probability corresponding to the functional component combination is greater than a predetermined functional component combination probability threshold, the current functional component combination is determined to be a characteristic functional component combination.

[0021] Furthermore, the process of comparing the command response data of the model training terminal with the data in the characteristic behavior database during the collection period includes:

[0022] Record the ports that currently generate data interaction. If the current port combination exists in the characteristic behavior database, determine that the port combination matches, and record the average value of the current data interaction volume corresponding to each port;

[0023] The difference between the current data interaction volume mean value corresponding to each port and the data interaction volume mean value stored in the characteristic behavior database is calculated respectively, and the data interaction volume matching feature is obtained by averaging the obtained differences;

[0024] Record the currently triggered functional components, if the current functional component combination exists in the characteristic behavior database, determine that the functional component combination matches, and record the current triggering frequency of each functional component;

[0025] The difference between the current trigger frequency of each functional component and the trigger frequency stored in the characteristic behavior database is calculated, and the trigger frequency matching feature is obtained by averaging the obtained differences.

[0026] Furthermore, the process of determining the behavior fitting representation coefficient corresponding to the model training terminal according to the comparison result includes:

[0027] Calculating a ratio of a predetermined benchmark data interaction amount matching threshold to a data interaction amount matching feature to obtain a first fitting characterization factor;

[0028] Calculating a ratio of a predetermined reference trigger frequency matching threshold to a trigger frequency matching characteristic to obtain a second fit characterization factor;

[0029] The sum of the first fitting characterization factor and the second fitting characterization factor is determined as the fitting characterization coefficient.

[0030] Furthermore, the process of identifying the feature matching category of the current model training terminal includes:

[0031] If the fitting characterization coefficient is greater than or equal to a predetermined fitting characterization coefficient threshold, the current model training terminal is determined to be a strong feature fitting category;

[0032] If the fitting characterization coefficient is less than a predetermined fitting characterization coefficient threshold, the current model training terminal is determined to be a weak feature fitting category.

[0033] Furthermore, the data interaction of the terminal based on the feature fitting category control model training of the current access behavior includes:

[0034] If the current model training terminal is a weak feature fitting category, the feature behavior database is called to identify the normal feature port, the data transmission rule of the normal feature port is verified, and it is determined whether to close the corresponding normal feature port, and verification requirement information is synchronously sent to the model training terminal to determine whether to issue an early warning signal based on the feedback information of the model training terminal;

[0035] If the current model training terminal is a strong feature fitting category, the interaction data of the model training terminal during the continuous collection period and the instruction data of the corresponding instruction input device are used to determine the feature fitting category of the model training terminal.

[0036] Furthermore, the process of calling the characteristic behavior database to identify the normal characteristic port includes:

[0037] Arrange each characteristic port combination in descending order according to its corresponding first conditional probability to obtain a characteristic port combination sequence;

[0038] A predetermined proportion of characteristic port combination numbers are selected from the head end of the characteristic port combination sequence;

[0039] Mark each characteristic port combination based on the sequence number of each characteristic port combination;

[0040] Identify the ports included in the marked characteristic port combination as normal characteristic ports;

[0041] Wherein, each characteristic port combination corresponds to a unique serial number, and the characteristic port combination sequence is composed of the unique serial numbers.

[0042] Further, the process of verifying the data transmission rule of the normal characteristic port and determining whether to close the corresponding normal characteristic port includes:

[0043] Obtain the current data interaction volume of each normal characteristic port, construct the data interaction volume time domain curve respectively, and determine the normal data interaction characteristics;

[0044] Call the characteristic behavior database to obtain the data interaction volume corresponding to each normal characteristic port, construct a sample time domain curve of the data interaction volume, and determine the sample data interaction characteristics;

[0045] Compare the normal data interaction features corresponding to the normal feature port with the sample data interaction features to determine the feature difference amount;

[0046] If the feature difference is greater than or equal to a predetermined feature difference threshold, it is determined to close the normal feature port;

[0047] The regular data interaction characteristics and the sample data interaction characteristics both include the average amplitude of the data interaction volume, the average slope of the rising segment, and the average slope of the falling segment.

[0048] Furthermore, the feedback information includes a key, and if the key does not match a predetermined key, it is determined to generate a warning signal.

[0049] Compared with the prior art, the present invention establishes a characteristic behavior database for the model training terminal based on the regular characteristics of instruction response data screening. When the model training terminal is accessed, the behavior fitting characterization coefficient corresponding to the model training terminal is determined based on the comparison between the instruction response data collected in real time and the data in the characteristic behavior database, the characteristic fitting category of the current model training terminal is identified, and the interaction of the model training terminal is controlled in time. Under the weak characteristic fitting category, the data transmission regularity of the normal characteristic port is verified, and it is determined whether to close the corresponding normal characteristic port. The present invention takes into account the regularity of the instruction response data corresponding to the model training terminal, timely distinguishes the characteristic fitting category, and intervenes in the interaction of the model training terminal by characterizing the abnormality of the access behavior, thereby improving the security of the model training terminal and reducing the risk of data leakage.

[0050] In particular, the present invention establishes a characteristic behavior database for the model training terminal based on the regular characteristics of instruction response data screening. In actual situations, due to the operating habits of authorized operators, when the model training terminal is accessed for a long time, the instructions of the corresponding instruction input device have certain regularity. For example, some functional components in some model training terminals are triggered intensively, while some functional components are rarely triggered. As a result, the model training terminal also shows regularity in the data interaction with each port. For example, data interaction is generated with a specific port, and the amount of data interaction is within a certain range. Therefore, it is considered to take the characteristic port combination and the characteristic functional component combination as regular features, construct an exclusive characteristic behavior database, provide data support for the subsequent identification of feature fitting categories, adaptively control the data interaction of the model training terminal, ensure security, and reduce the risk of data leakage.

[0051] In particular, the present invention calculates the behavior fitting characterization coefficient and constructs the feature fitting category of the model training terminal. The fitting characterization coefficient characterizes the difference between the regularity reflected by the corresponding instruction response data when the model training terminal is accessed and the regularity reflected by the corresponding instruction response data in normal state, so as to timely classify whether the current model terminal is a weak feature fitting category, identify the possible existence of disguised authorized visitors accessing the model training terminal, make subsequent interventions in time, and then adaptively control the data interaction of the model training terminal to ensure security.

[0052] In particular, the present invention identifies normal characteristic ports. In actual situations, the normal characteristic ports are determined based on the characteristic behavior database, which characterizes the port combination that generates a high frequency of data interaction in normal conditions, and has a high risk of being invaded. Furthermore, the present invention determines whether to close the normal characteristic ports in time based on the transmission rules of the normal characteristic ports. The data interaction volume and the change in the data interaction volume of the normal characteristic ports are considered when verifying the transmission rules. In actual situations, if a disguised operator accesses the model training terminal to steal data, interfere with data interaction, or causes pollution by calling the data input model, it will be reflected in the transmission rules. Therefore, the present invention identifies the above situation, closes the transmission port in time, ensures the security of the model training terminal, and reduces the risk of data leakage. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 A schematic diagram of the steps of a method for managing large model data leakage according to an embodiment of the invention;

[0054] Figure 2 A logical decision diagram for identifying the feature matching category of the current model training terminal according to an embodiment of the invention;

[0055] Figure 3 A logic decision diagram for determining whether to close a corresponding normal characteristic port according to an embodiment of the invention;

[0056] Figure 4 This is a logic decision diagram for determining whether to issue a warning signal according to an embodiment of the present invention. DETAILED DESCRIPTION

[0057] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0058] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0059] See also Figures 1 to 4 As shown, Figure 1 This is a schematic diagram of the steps of the method for managing large model data leakage according to an embodiment of the invention. Figure 2 This is a logic decision diagram for identifying the feature matching category of the current model training terminal according to an embodiment of the invention. Figure 3 This is a logic decision diagram for determining whether to close a corresponding normal characteristic port according to an embodiment of the invention. Figure 4 This is a logical decision diagram for determining whether to issue an early warning signal according to an embodiment of the invention. The method for managing large model data leakage in this embodiment includes:

[0060] S1, obtaining command response data of the model training terminal in the historical period, including the data interaction volume between the model training terminal and each port and the command trigger target of the corresponding command input device;

[0061] S2, screening regular features based on the command response data to establish a characteristic behavior database for the model training terminal, wherein the regular features include characteristic port combinations and characteristic function component combinations;

[0062] S3, in response to the model training terminal being accessed, comparing the instruction response data of the model training terminal during the acquisition period with the data in the characteristic behavior database;

[0063] S4, determining the behavior fitting representation coefficient corresponding to the model training terminal according to the comparison result, so as to identify the feature fitting category of the current model training terminal;

[0064] S5, based on the feature matching category of the current access behavior, controls the data interaction of the model training terminal, including:

[0065] Calling the characteristic behavior database to identify the normal characteristic port, verifying the data transmission rule of the normal characteristic port, determining whether to close the corresponding normal characteristic port, and synchronously sending verification requirement information to the model training terminal to determine whether to issue an early warning signal based on the feedback information of the model training terminal;

[0066] Alternatively, the interaction data of the model training terminal and the instruction data of the corresponding instruction input device within a continuous collection period are used to determine the feature matching category of the model training terminal.

[0067] Specifically, there is no limitation on the specific structure of the model training terminal. The model training terminal is a logical component that can perform logical operations. It only needs to be able to load or store the model to train the model, such as a computer, and this will not be repeated.

[0068] Specifically, during implementation, the instruction input device is connected to the model training terminal to input instructions to the model training terminal to implement corresponding functions, such as a keyboard, mouse, etc.

[0069] Specifically, the functional component is configured in the model training terminal to realize the corresponding function, such as a software in the model training terminal. Usually, the functional component can be triggered by the command input device through the control command to realize the corresponding function, which will not be repeated here.

[0070] Specifically, the model training terminal being accessed includes a visitor accessing the model training terminal and controlling the model training terminal through a command input device.

[0071] Specifically, in computer networks, a port refers to a logical communication endpoint, which is used to distinguish different network services or processes running on the terminal. Each port is identified by a unique port number, which is a 16-bit number ranging from 0 to 65535. This will not be repeated here.

[0072] Specifically, in step S2, the process of screening the characteristic port combination based on the instruction response data includes:

[0073] Determine a first conditional probability that the model training terminal interacts with different port combinations during a number of accesses. It is understandable that a single access includes the time from when the visitor starts accessing the model training terminal to when the visitor stops accessing the model training terminal.

[0074] It is understandable that the model training terminal may interact with different ports during a single access due to different control instructions.

[0075] For example, there are 5 ports in total, and the port numbers are 1, 2, 3, 4, and 5 respectively;

[0076] If the current port combination is 1, 2, 3, and the model training terminal is accessed 10 times, among which the model training terminal is accessed 3 times under the current port combination, then the first conditional probability corresponding to the port combination is 3 / 10, that is, the ratio of the number of times the model training terminal is accessed under the current port combination to the total number of times the model training terminal is accessed;

[0077] Determine a characteristic port combination based on the first conditional probability corresponding to each port combination, record the average value of the data interaction amount of each port in the characteristic port combination during data interaction, and store it in a characteristic behavior database;

[0078] If the first conditional probability corresponding to the port combination is greater than a predetermined port combination conditional probability threshold, the port combination is determined to be a characteristic port combination.

[0079] Specifically, the port combination conditional probability threshold is determined based on the mean of the first conditional probabilities of each port combination, and is set to be the product of the mean of the conditional probability and the precision coefficient, and the precision coefficient is selected within the interval [1.05, 1.15].

[0080] Specifically, in step S2, the process of screening the characteristic function component combination based on the instruction response data includes:

[0081] Obtaining a second conditional probability that each combination of functional components is triggered during a number of accesses to the model training terminal;

[0082] It can be understood that the second conditional probability is the ratio of the number of times the model training terminal is accessed to the total number of times the model training terminal is accessed under the functional component combination;

[0083] Determine the characteristic functional component combination based on the second conditional probability corresponding to the functional component combination, and record the triggering frequency of each functional component in the characteristic functional component combination. It can be understood that the triggering frequency is the number of times triggered within the reference time. In order to make the acquired instruction response data representative, the reference time can be selected within 3 minutes to 10 minutes.

[0084] If the second conditional probability corresponding to the functional component combination is greater than a predetermined functional component combination probability threshold, the current functional component combination is determined to be a characteristic functional component combination.

[0085] Specifically, in implementation, the functional component combination probability threshold is determined based on the mean second conditional probability of each functional component combination being triggered, and is set to the product of the mean second conditional probability and the precision coefficient.

[0086] The present invention is based on the regular characteristics of instruction response data screening and establishes a characteristic behavior database for the model training terminal. In actual situations, due to the operating habits of authorized operators, when the model training terminal is accessed for a long time, the instructions of the corresponding instruction input device have certain regularity. For example, some functional components in some model training terminals are triggered intensively, while some functional components are rarely triggered. As a result, the model training terminal also shows regularity in the data interaction with each port. For example, data interaction is generated with a specific port, and the amount of data interaction is within a certain range. Therefore, it is considered to take the characteristic port combination and the characteristic functional component combination as regular characteristics, construct an exclusive characteristic behavior database, provide data support for the subsequent identification of feature fitting categories, adaptively control the data interaction of the model training terminal, ensure security, and reduce the risk of data leakage.

[0087] Specifically, in step S3, the process of comparing the command response data of the model training terminal within the acquisition period with the data in the characteristic behavior database includes:

[0088] Record the port that currently generates data interaction. If the current port combination exists in the characteristic behavior database, determine that the port combination matches, and record the average value of the current data interaction volume corresponding to each port. It can be understood that the data interaction volume of each port in each time period within a certain time period can be synchronously recorded to further solve the average value of the data interaction volume;

[0089] The difference between the current data interaction volume mean value corresponding to each port and the data interaction volume mean value stored in the characteristic behavior database is calculated respectively, and the data interaction volume matching feature is obtained by averaging the obtained differences;

[0090] Record the currently triggered functional components, if the current functional component combination exists in the characteristic behavior database, determine that the functional component combination matches, and record the current triggering frequency of each functional component;

[0091] The difference between the current trigger frequency of each functional component and the trigger frequency stored in the characteristic behavior database is calculated, and the trigger frequency matching feature is obtained by averaging the obtained differences.

[0092] Specifically, in step S4, the process of determining the behavior fitting representation coefficient corresponding to the model training terminal according to the comparison result includes:

[0093] Calculating a ratio of a predetermined benchmark data interaction amount matching threshold to a data interaction amount matching feature to obtain a first fitting characterization factor;

[0094] Calculating a ratio of a predetermined reference trigger frequency matching threshold to a trigger frequency matching characteristic to obtain a second fit characterization factor;

[0095] The sum of the first fitting characterization factor and the second fitting characterization factor is determined as the fitting characterization coefficient.

[0096] Specifically, the interaction volume matching threshold is determined based on the average data interaction volume of the port stored in the characteristic behavior database, and is usually set to between 0.25 times and 0.35 times the average data interaction volume.

[0097] Specifically, the reference trigger frequency matching threshold is determined based on the trigger frequency of the functional component stored in the characteristic behavior database, and is usually set to between 0.15 times and 0.25 times the trigger frequency.

[0098] The present invention calculates the behavior fitting characterization coefficient and constructs the feature fitting category of the model training terminal. The fitting characterization coefficient characterizes the difference between the regularity reflected by the corresponding instruction response data when the model training terminal is accessed and the regularity reflected by the corresponding instruction response data in normal state, so as to timely classify whether the current model terminal is a weak feature fitting category, identify the possible existence of disguised authorized visitors to access the model training terminal, make subsequent interventions in time, and then adaptively control the data interaction of the model training terminal to ensure security.

[0099] Specifically, in step S4, the process of identifying the feature matching category of the current model training terminal includes:

[0100] If the fitting characterization coefficient is greater than or equal to a predetermined fitting characterization coefficient threshold, the current model training terminal is determined to be a strong feature fitting category;

[0101] If the fitting characterization coefficient is less than a predetermined fitting characterization coefficient threshold, the current model training terminal is determined to be a weak feature fitting category.

[0102] Specifically, the fitting characterization coefficient threshold is calculated based on the first fitting characterization factor and the second fitting characterization factor. It can be understood that when the first fitting characterization factor and the second fitting characterization factor are close to 1, the characterization tends to the upper limit of the acceptable difference, and the fitting characterization coefficient threshold is calculated based on the first fitting characterization factor and the second fitting characterization factor, which is the upper limit of the characterization difference. Those skilled in the art can select the fitting characterization coefficient threshold in the interval [2.15,2.25].

[0103] Specifically, in step S5, the data interaction of the terminal for controlling the model training based on the feature fitting category of the current access behavior includes:

[0104] If the current model training terminal is a weak feature fitting category, the feature behavior database is called to identify the normal feature port, the data transmission rule of the normal feature port is verified, and it is determined whether to close the corresponding normal feature port, and verification requirement information is synchronously sent to the model training terminal to determine whether to issue an early warning signal based on the feedback information of the model training terminal;

[0105] If the current model training terminal is a strong feature fitting category, the interaction data of the model training terminal during the continuous collection period and the instruction data of the corresponding instruction input device are used to determine the feature fitting category of the model training terminal.

[0106] Specifically, in step S5, the process of calling the characteristic behavior database to identify the normal characteristic port includes:

[0107] Arrange each characteristic port combination in descending order according to its corresponding first conditional probability to obtain a characteristic port combination sequence;

[0108] A predetermined proportion of characteristic port combination numbers are selected from the head end of the characteristic port combination sequence;

[0109] Mark each characteristic port combination based on the sequence number of each characteristic port combination;

[0110] Identify the ports included in the marked characteristic port combination as normal characteristic ports;

[0111] Wherein, each characteristic port combination corresponds to a unique serial number, and the characteristic port combination sequence is composed of the unique serial numbers.

[0112] Specifically, in order to filter out data with strong representativeness of the front-end data of the characteristic port combination sequence, the predetermined proportion is selected within a range of 15% to 30%.

[0113] Specifically, in step S5, the process of verifying the data transmission rule of the normal characteristic port and determining whether to close the corresponding normal characteristic port includes:

[0114] Obtain the current data interaction volume of each normal characteristic port, construct the data interaction volume time domain curve respectively, and determine the normal data interaction characteristics;

[0115] Call the characteristic behavior database to obtain the data interaction volume corresponding to each normal characteristic port, construct a sample time domain curve of the data interaction volume, and determine the sample data interaction characteristics;

[0116] Compare the normal data interaction features corresponding to the normal feature port with the sample data interaction features to determine the feature difference amount;

[0117] If the feature difference is greater than or equal to a predetermined feature difference threshold, it is determined to close the normal feature port;

[0118] The regular data interaction characteristics and the sample data interaction characteristics both include the average amplitude of the data interaction volume, the average slope of the rising segment, and the average slope of the falling segment.

[0119] Specifically, the feature difference is determined based on the comparison results of the interaction features of the normal data and the interaction features of the sample data, including

[0120] Determine the difference between the average amplitude of the data interaction amount in the constant data interaction feature and the average amplitude of the data interaction amount in the sample data interaction feature, and solve the ratio of the obtained difference to the average amplitude of the data interaction amount in the sample data interaction feature to obtain a first feature difference amount;

[0121] Determine the difference between the average slope of the rising segment in the interaction feature of the constant data and the average slope of the rising segment in the interaction feature of the sample data, and solve the ratio of the obtained difference to the average slope of the rising segment in the interaction feature of the sample data to obtain a second feature difference amount;

[0122] Determine the difference between the average slope of the descending segment in the interaction feature of the constant data and the average slope of the descending segment in the interaction feature of the sample data, and solve the ratio of the obtained difference to the average slope of the descending segment in the interaction feature of the sample data to obtain a third feature difference amount;

[0123] setting the sum of the first characteristic difference amount, the second characteristic difference amount, and the third characteristic difference amount as the characteristic difference amount;

[0124] It is understandable that both the regular data interaction features and the sample data interaction features can be extracted based on the corresponding time domain curves, which will not be elaborated here.

[0125] It is understandable that the data interaction volume of the normal feature port is allowed to have a certain deviation, and setting the feature difference threshold represents a situation with a large difference. Therefore, those skilled in the art may select the feature difference threshold within the interval [0.65, 0.85].

[0126] The present invention identifies normal characteristic ports. In actual situations, the normal characteristic ports are determined based on a characteristic behavior database, which characterizes a port combination that generates a high frequency of data interaction in normal conditions and has a high risk of being invaded. Furthermore, the present invention determines whether to close the normal characteristic ports in time based on the transmission rules of the normal characteristic ports. The data interaction volume and the change in the data interaction volume of the normal characteristic ports are considered when verifying the transmission rules. In actual situations, if a disguised operator accesses the model training terminal to steal data, interfere with data interaction, or causes pollution by calling the data input model, it will be reflected in the transmission rules. Therefore, the present invention identifies the above situation, closes the transmission port in time, ensures the security of the model training terminal, and reduces the risk of data leakage.

[0127] Specifically, the feedback information includes a key, and if the key does not match a predetermined key, it is determined to generate a warning signal.

[0128] It is understandable that the verification requirement information includes sending a key provision request to the model training terminal.

[0129] The key can be pre-set and can be in the form of asymmetric encryption or symmetric encryption. Those skilled in the art can verify whether the key complies with the corresponding encryption form by themselves, which will not be repeated here.

[0130] The generated early warning signal can be sent to the supervisory end for supervision, which will not be elaborated here.

[0131] If the large model data leakage management method of the present invention is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to 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 code.

[0132] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

Claims

1. A method for managing large model data leakage, characterized in that: include: Obtaining command response data within a historical period of the model training terminal, including the amount of data interaction between the model training terminal and each port and the command trigger target of the corresponding command input device; Screening regular features based on the instruction response data to establish a characteristic behavior database for the model training terminal, wherein the regular features include characteristic port combinations and characteristic function component combinations; In response to the model training terminal being accessed, the instruction response data of the model training terminal during the acquisition period is compared with the data in the characteristic behavior database; Determine the behavior fit representation coefficient corresponding to the model training terminal according to the comparison result to identify the feature fit category of the current model training terminal; Based on the characteristics of the current access behavior, the category control model is trained to control the data interaction of the terminal, including: Calling the characteristic behavior database to identify the normal characteristic port, verifying the data transmission rule of the normal characteristic port, determining whether to close the corresponding normal characteristic port, and synchronously sending verification requirement information to the model training terminal to determine whether to issue an early warning signal based on the feedback information of the model training terminal; Or, the interaction data of the model training terminal and the instruction data of the corresponding instruction input device within the continuous collection period are used to determine the feature matching category of the model training terminal; The process of screening the characteristic port combination based on the instruction response data includes: Determine a first conditional probability that a model training terminal interacts with different port combinations during a number of accesses; Determine a characteristic port combination based on the first conditional probability corresponding to each port combination, record the average value of the data interaction amount of each port in the characteristic port combination during data interaction, and store it in a characteristic behavior database; If the first conditional probability corresponding to the port combination is greater than a predetermined port combination conditional probability threshold, the port combination is determined to be a characteristic port combination; The process of screening the characteristic function component combination based on the instruction response data includes: Obtaining a second conditional probability that each combination of functional components is triggered during a number of accesses to the model training terminal; Determine a characteristic function component combination based on a second conditional probability corresponding to the function component combination, and record a trigger frequency of each function component in the characteristic function component combination; If the second conditional probability corresponding to the functional component combination is greater than a predetermined functional component combination probability threshold, the current functional component combination is determined to be a characteristic functional component combination; The process of comparing the command response data of the model training terminal with the data in the characteristic behavior database during the collection period includes: Record the ports that currently generate data interaction. If the current port combination exists in the characteristic behavior database, determine that the port combination matches, and record the average value of the current data interaction volume corresponding to each port; The difference between the current data interaction volume mean value corresponding to each port and the data interaction volume mean value stored in the characteristic behavior database is calculated respectively, and the data interaction volume matching feature is obtained by averaging the obtained differences; Record the currently triggered functional components, if the current functional component combination exists in the characteristic behavior database, determine that the functional component combination matches, and record the current triggering frequency of each functional component; Calculate the difference between the current trigger frequency of each functional component and the trigger frequency stored in the characteristic behavior database, and average the obtained differences to obtain the trigger frequency matching feature; The process of determining the behavior fit representation coefficient corresponding to the model training terminal based on the comparison results includes: Calculating a ratio of a predetermined benchmark data interaction amount matching threshold to a data interaction amount matching feature to obtain a first fitting characterization factor; Calculating a ratio of a predetermined reference trigger frequency matching threshold to a trigger frequency matching characteristic to obtain a second fit characterization factor; Determine the sum of the first fitting characterization factor and the second fitting characterization factor as a fitting characterization coefficient; The process of identifying the feature fit category of the current model training terminal includes: If the fitting characterization coefficient is greater than or equal to a predetermined fitting characterization coefficient threshold, the current model training terminal is determined to be a strong feature fitting category; If the fitting characterization coefficient is less than a predetermined fitting characterization coefficient threshold, the current model training terminal is determined to be a weak feature fitting category.

2. The method for managing large model data leakage according to claim 1, characterized in that: Based on the feature matching category control model training terminal data interaction of current access behavior includes: If the current model training terminal is a weak feature fitting category, the feature behavior database is called to identify the normal feature port, the data transmission rule of the normal feature port is verified, and it is determined whether to close the corresponding normal feature port, and verification requirement information is synchronously sent to the model training terminal to determine whether to issue an early warning signal based on the feedback information of the model training terminal; If the current model training terminal is a strong feature fitting category, the interaction data of the model training terminal during the continuous collection period and the instruction data of the corresponding instruction input device are used to determine the feature fitting category of the model training terminal.

3. The method for managing large model data leakage according to claim 1, characterized in that: The process of calling the characteristic behavior database to identify normal characteristic ports includes: Arrange each characteristic port combination in descending order according to its corresponding first conditional probability to obtain a characteristic port combination sequence; A predetermined proportion of characteristic port combination numbers are selected from the head end of the characteristic port combination sequence; Mark each characteristic port combination based on the sequence number of each characteristic port combination; Identify the ports included in the marked characteristic port combination as normal characteristic ports; Wherein, each characteristic port combination corresponds to a unique serial number, and the characteristic port combination sequence is composed of the unique serial numbers.

4. The method for managing large model data leakage according to claim 1, characterized in that: The process of verifying the data transmission rules of the normal characteristic port and determining whether to close the corresponding normal characteristic port includes: Obtain the current data interaction volume of each normal characteristic port, construct the data interaction volume time domain curve respectively, and determine the normal data interaction characteristics; Call the characteristic behavior database to obtain the data interaction volume corresponding to each normal characteristic port, construct a sample time domain curve of the data interaction volume, and determine the sample data interaction characteristics; Compare the normal data interaction features corresponding to the normal feature port with the sample data interaction features to determine the feature difference amount; If the feature difference is greater than or equal to a predetermined feature difference threshold, it is determined to close the normal feature port; The regular data interaction characteristics and the sample data interaction characteristics both include the average amplitude of the data interaction volume, the average slope of the rising segment, and the average slope of the falling segment.

5. The method for managing large model data leakage according to claim 1, characterized in that: The feedback information includes a key, and if the key does not match a predetermined key, it is determined that an early warning signal is generated.

Images