Data protection method and device for fingerprint browser and storage medium

Abstract

The invention provides a data protection method and device for a fingerprint browser and a medium, and the method comprises the steps: inputting device fingerprint data, and constructing a device fingerprint data set; calculating an entropy value of each feature according to the high-frequency feature set and the low-frequency feature set, and constructing an entropy value set of each feature; designing a low-frequency feature disturbance formula for the low-frequency feature set to obtain a disturbed low-frequency feature value and a disturbed low-frequency feature set; constructing a generative adversarial network according to the high-frequency feature set and the disturbed low-frequency feature set; and collecting dynamic feedback information of the tracking system in real time, and finally outputting an updated generative adversarial network and a verified confusion feature set. According to the method, the contradiction between privacy protection and function compatibility in the prior art is solved, and the dynamic adaptation capability of the system is remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of data protection, and in particular relates to a data protection method, device and storage medium for a fingerprint browser. Background Art

[0002] With the rapid development of the Internet and big data technology, user privacy protection is facing unprecedented challenges. As an important means of user identification and tracking, browser fingerprinting technology has been widely used in areas such as advertising push, user behavior analysis, and security verification. Browser fingerprinting generates a unique identifier by collecting specific parameters of the user's device (such as screen resolution, time zone settings, language preferences, font libraries, and Canvas drawing features, etc.) to mark users in multiple scenarios. The imperceptibility, persistence, and cross-site adaptability of this technology make it one of the difficulties in privacy protection.

[0003] In the prior art, solutions to the browser fingerprint privacy problem are mainly divided into two categories: one is to block the acquisition of fingerprint features, that is, to reduce the fingerprint generation information by prohibiting access to specific APIs or blocking certain parameters, such as limiting Canvas drawing data or blocking plug-in information. Although this type of method weakens the tracking ability to a certain extent, it will lead to partial loss of browser functions, significantly reduce user experience, and even affect the normal use of applications in some scenarios (such as online payment, identity authentication). The other is the fixed processing of fingerprint features, that is, to avoid unique identification by setting unified fingerprint parameters. For example, the time zone, resolution, font, etc. of all users are fixed to a unified value. However, this method is not only easily identified as abnormal behavior by the tracking system, but also reduces the user's functional compatibility due to the consistency of the features, especially in application scenarios that require personalized settings (such as multi-language support). More importantly, these two types of methods lack dynamism and cannot cope with the evolving technical means of the tracking system.

[0004] Faced with the above problems, existing technologies cannot achieve an effective balance between privacy protection and functional compatibility, and it is also difficult to achieve dynamic adaptation of the tracking system. This contradiction urgently requires an innovative method that can protect privacy while not affecting the user experience, and has the ability to dynamically adjust and adapt to changes in tracking technology at any time. Summary of the invention

[0005] The purpose of the present invention is to propose a data protection method, device and storage medium for a fingerprint browser, which solves the contradiction between privacy protection and functional compatibility in the prior art and significantly improves the dynamic adaptation capability of the system.

[0006] In order to achieve the above object, the present invention provides a data protection method for a fingerprint browser, the method comprising: S1. Input device fingerprint data to construct a device fingerprint data set, wherein each feature in the device fingerprint data set represents a characteristic of a user device, and the contribution of each feature to user discrimination is calculated by a variance normalization method to obtain a feature score of each feature. A stratification threshold is designed according to the feature score, and the device fingerprint features are divided into two layers, including a high-frequency feature set and a low-frequency feature set; S2. Calculate the entropy value of each feature according to the high-frequency feature set and the low-frequency feature set, and construct an entropy value set for each feature, and calculate the average entropy value of the low-frequency feature set and the high-frequency feature set according to the entropy value of each feature, and construct an optimization objective function based on the average entropy value of the low-frequency feature set and the high-frequency feature set to balance the computational overhead of privacy protection, functional integrity and feature disturbance, and obtain the expected target entropy value of each feature; S3. Design a low-frequency feature perturbation formula for the low-frequency feature set, obtain the low-frequency feature value after perturbation and the low-frequency feature set after perturbation, recalculate the entropy value set of the low-frequency feature set after perturbation to evaluate the perturbation effect, and obtain a complexity index; S4. Constructing a generative adversarial network based on the high-frequency feature set and the perturbed low-frequency feature set to enhance the interference capability to the tracking system while maintaining the integrity of the fingerprint browser function. The generative adversarial network finally outputs a confusion feature set. Among them, the target design of the generative adversarial network is as follows: Generator Receiving a high-frequency feature set and a perturbed low-frequency feature set and outputting a confused feature set; Discriminator Receive a set of confused features and output a probability value, indicating the possibility that the input features are the true feature set; Output optimized generator and the discriminator ; S5. Collect dynamic feedback information from the tracking system in real time, including the actual recognition rate and the distribution characteristics of the actual recognition samples, calculate the feedback deviation between the target recognition rate set by the system and the actual recognition rate, and dynamically adjust the optimized generator according to the feedback deviation The regular term weight of the objective function of The loss function is used, and the dynamic feedback information is added to the optimized discriminator Maintain the ability to distinguish between true feature sets and confused feature sets, thereby optimizing the discriminator The loss function is to use the optimized generator Loss function and optimized discriminator The loss function for the generator and the discriminator Iterative training is performed to finally output an updated generative adversarial network and a verified set of confusion features.

[0007] Furthermore, the variance normalization method is used to calculate the contribution of each feature to user discrimination, and obtain the feature score of each feature, which is specifically: ; in, Representation characteristics The variance of the values ​​in the device fingerprint dataset reflects the distinguishability of the feature. The largest score indicates that the feature is most important in user identification. Score the importance of feature i and construct a score set , Representation characteristics The variance of the values ​​in the device fingerprint dataset; n is the total number of features; The high-frequency feature set includes all scores Features, is the stratification threshold; The low-frequency feature set includes all scores Features; At the same time, design dynamic layered updates: For scenarios with high real-time requirements, a sliding window dynamic update mechanism is designed to re-evaluate the feature importance score based on the user's latest interactive behavior. The updated score calculation formula is: ; in, is the weight coefficient, which is used to balance the impact of new data and historical data. It is the result of the previous round of scoring. The updated ultra-high frequency feature set and low frequency feature set adapt to the user's latest device status and improve the dynamics of feature stratification.

[0008] Furthermore, the entropy value of each feature is calculated according to the high-frequency feature set and the low-frequency feature set, which is expressed as: ; in, Representation characteristics The highest information entropy indicates that the feature has the greatest randomness and the least dependence on user identification; It is a feature The possible value set of Is the value probability; is the weight coefficient of the compensation term, which is used to balance the essential information of the entropy value and the fingerprint protection requirements; is a randomness compensation term, which is used to increase the entropy value for certain specific features and is defined as: ; in, It is a feature The number of possible values ​​of represents the inherent diversity of the feature; is a set of low-frequency features; The entropy value of each feature The average entropy value is calculated for the low-frequency feature set and the high-frequency feature set respectively, which is expressed as: ; in, It is the average entropy value of low-frequency features, which mainly reflects the randomness of the feature set with a lower protection level; It is the average entropy value of high-frequency features, which mainly reflects the randomness of the functional priority feature set; is the number of low-frequency features, is the number of high-frequency features; is a set of low-frequency features; is a set of high-frequency features; The optimization objective function , expressed as: ; in, It is the trade-off coefficient between privacy and functionality, which determines the emphasis on protection strength and functional priority; is the average entropy of low-frequency features, is the regularization weight coefficient, which is used to limit the complexity of feature perturbations and avoid unnecessary computational overhead; is the perturbation complexity regularization term, which is used to limit the overhead of low-frequency features in the randomization process and is defined as: ; in, represents the number of possible values ​​of the feature, the relative complexity required to represent the characteristic perturbation; It is a feature The variance of represents the scalability of its original distribution. The smaller the variance, the greater the difficulty of perturbation.

[0009] Furthermore, S3 also includes calculating the entropy difference of each feature according to the expected target entropy value of each feature and the actual entropy value of each feature, wherein the disturbance intensity is proportional to the entropy difference of each feature, and the feature with the largest randomness difference is preferentially assigned the largest disturbance weight.

[0010] Furthermore, the S3 specifically includes: For low-frequency feature sets The low-frequency characteristic perturbation formula is expressed as: ; in, is the perturbed eigenvalue, is the original eigenvalue; is a random disturbance term, from a normal distribution Medium sampling, Dynamically adjust to control disturbance amplitude; Representation characteristics The variance of reflects its distribution characteristics; is a regularization factor that controls the dynamic adjustment of the perturbation amplitude with the feature distribution, ensuring that more significant perturbations are applied to features with more uniform distribution; It is the feature with the largest variance in the low-frequency feature set and is used for normalization; The perturbed feature set Design perturbation complexity index , recalculate its entropy value To evaluate the perturbation effect, it is expressed as: ; in, Representation characteristics The size of the possible value set of is used to evaluate feature complexity; is the square of the disturbance amplitude and represents the disturbance energy.

[0011] Furthermore, the target design of the generative adversarial network is specifically: Generator Optimization goals design: ; Item 1 Represents the ability of the generator to confuse the discriminator. The generator hopes to confuse the feature set Misidentified by the discriminator as a real feature; Represents the probability value of the discriminator judging whether the confused feature is a real feature; Item 2 is a regularization term used to improve the randomness of the obfuscated features. represents the weight coefficient of the regularization term, where Represents the average entropy value of the confusion feature set: ; in, is the entropy value of each feature; Discriminator D optimization goal design: ; in, represents the real feature set, Represents the generated confusion feature set.

[0012] Furthermore, the generator after dynamic adjustment of the optimization according to the feedback deviation The regular term weight of the objective function of The loss function includes: ; in, For the optimized generator The loss function of Item 1 is the main GAN loss of the generator, ensuring that the obfuscated features are difficult for the discriminator to distinguish; Item 2 It is a feedback-driven random regularization term that improves the entropy and randomness of the obfuscated features through dynamically adjusted regularization weights: ; in, It is a confusing feature The entropy value of , which measures the randomness of features; It is a confusing feature The distribution variance of ensures the diversity of features; is the characteristic distribution diversity adjustment coefficient; At the same time, dynamic feedback information is added to the optimized discriminator Maintain the ability to distinguish between true feature sets and confused feature sets, thereby optimizing the discriminator The loss function is expressed as: ; in, The optimized discriminator The loss function of The first and second terms are the standard GAN losses, which measure the probability that a true feature is judged as true and the probability that a confused feature is judged as generated, respectively; Item 3 is a feedback control regularization term used to limit the overfitting of the discriminator to the confusing features: ; This term controls the output of the discriminator to deviate from the central value. to avoid overfitting and simplification.

[0013] Furthermore, the S5 also includes verifying the updated generator and the discriminator Are the following conditions met: Recognition rate of confusing feature set ; The average entropy of the confusion feature set Reach the set range; If the set range is not reached or the recognition rate is not reached Then the feedback bias is recalculated to train the generative adversarial network.

[0014] In addition, to achieve the above-mentioned purpose, the present invention also proposes a data protection device for a fingerprint browser, the device comprising: a memory, a processor, and a data protection program for a fingerprint browser stored on the memory and executable on the processor, the data protection configuration for a fingerprint browser being configured to implement the steps of a data protection method for a fingerprint browser as described above.

[0015] In addition, to achieve the above-mentioned purpose, the present invention also proposes a storage medium, on which is stored a data protection program for a fingerprint browser. When the data protection program for a fingerprint browser is executed by a processor, the steps of a data protection method for a fingerprint browser as described above are implemented.

[0016] The beneficial technical effects of the present invention are at least as follows: (1) This invention introduces a dynamic hierarchical modeling method for fingerprint features for the first time, dividing fingerprint features into high-frequency layers and low-frequency layers according to the importance and frequency of use of the features. By maintaining dynamic legitimacy output for high-frequency features, the integrity of browser functions is ensured; and by injecting fake features or obfuscated parameters into low-frequency features, the privacy protection effect is improved. This dynamic hierarchical mechanism can effectively balance privacy protection and functional compatibility, and solves the compatibility problem caused by fingerprint fixation in the prior art.

[0017] (2) The present invention dynamically adjusts the obfuscation strategy of different features by calculating the entropy value of fingerprint features. For features with low information entropy (such as plug-in order or font information), their randomness and diversity are increased first, thereby interfering with the analysis of the tracking system; for features with high information entropy (such as screen resolution or Canvas drawing), moderate disturbance is maintained to ensure normal function. The entropy optimization method makes the obfuscated features closer to the real data in statistical distribution, thereby effectively improving the protection effect and solving the problem that the single obfuscated features in the prior art are easily identified.

[0018] (3) The present invention designs a dynamically updated Generative Adversarial Network (GAN) to generate obfuscated fingerprint features in real time. The generator dynamically generates high-entropy forged fingerprints based on user behavior and tracking scenarios, while the discriminator simulates the recognition logic of the tracking system to ensure that the generated features achieve the best balance between privacy protection and authenticity. The network has adaptive capabilities and can automatically adjust the obfuscation strategy according to the technical evolution of the tracking system, solving the problem that static protection methods in the prior art are difficult to cope with dynamic threats. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The present invention is further described using the accompanying drawings, but the embodiments in the accompanying drawings do not constitute any limitation to the present invention. A person skilled in the art can obtain other drawings based on the following drawings without creative work.

[0020] Figure 1 The present invention is a flow chart of a data protection method for a fingerprint browser. DETAILED DESCRIPTION

[0021] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0022] like Figure 1 As shown, an embodiment of the present invention provides a data protection method for a fingerprint browser, the method comprising the following steps S1-S5: S1. Input device fingerprint data and construct a device fingerprint data set, wherein each feature in the device fingerprint data set represents a characteristic of the user device, and the contribution of each feature to user discrimination is calculated by the variance normalization method to obtain a feature score of each feature. A stratification threshold is designed according to the feature score, and the device fingerprint features are divided into two layers, including a high-frequency feature set and a low-frequency feature set.

[0023] Specifically, this step aims to collect device fingerprint features and provide support for subsequent steps through dynamic layering modeling. Feature standardization eliminates differences in data ranges, combines feature importance scoring models to quantify feature sensitivity, and then layers features based on the scoring results. The design of dynamic layering not only ensures the identification of key features for protection, but also provides clear input for subsequent entropy optimization and obfuscation strategies.

[0024] Input is device fingerprint dataset , where each feature Represents a feature of the user's device, such as screen resolution, time zone, language setting, font library, etc. First, each feature is standardized to eliminate the impact of different feature value ranges. The standardization formula is: ; in, is the standardized feature, and Characteristics The mean and standard deviation of are calculated from the training data set. It is the basic data for subsequent feature scoring.

[0025] Furthermore, the feature scoring model is used to calculate the relative importance of each standardized feature, and the contribution of each feature to user discrimination is quantified by the variance normalization method: ; in, Representation characteristics The variance of the values ​​in the device fingerprint dataset reflects the distinguishability of the feature. The larger the score value, the more important the feature is in user identification. The calculated score set It is the core basis for dividing high-frequency and low-frequency features.

[0026] Furthermore, according to the importance score Set a stratification threshold , the device fingerprint features are divided into two layers: High frequency feature set Include all ratings These features are usually core features related to user functions, such as resolution, language options, etc.

[0027] Low frequency feature set Include all ratings These features contribute relatively little to identifying users and can be perturbed by subsequent obfuscation operations.

[0028] After stratification, output high-frequency feature set and low-frequency feature set , which is used for the next step of entropy calculation.

[0029] Furthermore, for scenarios with high real-time requirements, a sliding window dynamic update mechanism is designed to re-evaluate the feature importance score based on the user's latest interactive behavior (such as device update, browsing frequency). The updated score calculation formula is: ; in, is the weight coefficient, which is used to balance the impact of new data and historical data. is the result of the last round of scoring. Updated feature stratification results and Adapt to the user's latest device status and improve the dynamics of feature stratification.

[0030] S2. Calculate the entropy value of each feature according to the high-frequency feature set and the low-frequency feature set, and construct an entropy value set for each feature, and calculate the average entropy value of the low-frequency feature set and the high-frequency feature set according to the entropy value of each feature, and construct an optimization objective function based on the average entropy value of the low-frequency feature set and the high-frequency feature set to balance the computational overhead of privacy protection, functional integrity and feature disturbance, so as to obtain the expected target entropy value of each feature.

[0031] Furthermore, for the input and , calculate the entropy value of each feature respectively. The entropy value is defined as: ; in, Representation characteristics The higher the information entropy, the greater the randomness of the feature and the less dependent it is on user identification. It is a feature The possible value set of Is the value probability. It is the weight coefficient of the compensation term, which is used to balance the essential information of the entropy value and the fingerprint protection requirements. It is an additional randomness compensation item designed to increase the entropy value for certain specific features (such as fonts and plug-in lists). These features are more easily exploited in the tracking system and need to further improve their randomness. It is defined as: ; in, It is a feature The number of possible values ​​of , indicating the inherent diversity of the feature.

[0032] By introducing compensation items, the entropy calculation is made more adaptable to the scenario requirements of fingerprint protection, especially low-frequency features require higher randomness to interfere with tracking.

[0033] Furthermore, after calculating the entropy value of each feature, the average entropy value is calculated for the low-frequency feature set and the high-frequency feature set respectively, and the formula is: ; in, It is the average entropy value of low-frequency features, which mainly reflects the randomness of feature sets with lower protection levels. It is the average entropy value of high-frequency features, which mainly reflects the randomness of the functional priority feature set. is the number of low-frequency features, is the number of high frequency features. Corresponding to higher privacy protection effect, A certain level should be maintained to ensure functional integrity.

[0034] Furthermore, in order to balance privacy protection and functional requirements, an optimization objective function is constructed , while considering global randomness, protection requirements and tracking interference capabilities. The optimization objective function is: ; in, It is the trade-off between privacy and functionality, which determines the emphasis on protection strength and functional priority. is the regularization weight coefficient, which is used to limit the complexity of feature perturbations and avoid unnecessary computational overhead. is an innovatively designed perturbation complexity regularization term used to limit the overhead of low-frequency features in the randomization process, which is defined as: ; in, represents the number of possible values ​​of the feature, Relative complexity required to represent feature perturbations. It is a feature The variance of indicates the scalability of its original distribution. The smaller the variance, the greater the difficulty of perturbation. By introducing ,Optimizing the objective function can more effectively balance the computational overhead of privacy protection, functional integrity, and feature perturbation.

[0035] Understandably, the final output is as follows: The entropy value set for each feature , clarify the randomness level of each feature.

[0036] Average entropy of low-frequency and high-frequency features and , providing global characteristic randomness indicators for subsequent protection strategies.

[0037] Optimizing the objective function , as an important reference for subsequent feature perturbation and confusion strategies.

[0038] S3. Design a low-frequency feature perturbation formula for the low-frequency feature set to obtain the perturbed low-frequency feature value and the perturbed low-frequency feature set, recalculate the entropy value set of the perturbed low-frequency feature set to evaluate the perturbation effect, and obtain a complexity index.

[0039] Specifically, according to the optimization objective function calculated in step S2 The entropy difference between each feature , is the low-frequency feature set Assign perturbation strength. Define each feature The entropy difference is: ; in, is the desired target entropy value, usually obtained by optimizing the objective function Decision, for example, selecting the target average entropy value of the low-frequency features as a benchmark. is the current entropy value calculated in step S2.

[0040] The disturbance intensity is The larger the randomness gap, the features with larger perturbation weights are assigned preferentially.

[0041] Furthermore, for the low-frequency feature set , using the following innovative perturbation formula: ; in, is the perturbed eigenvalue, is the original eigenvalue. is a random disturbance term, from a normal distribution Medium sampling, Dynamically adjust to control the magnitude of disturbances. Representation characteristics The variance reflects its distribution characteristics. is a regularization factor that controls the dynamic adjustment of the perturbation amplitude with the feature distribution, ensuring that more significant perturbations are applied to features with more uniform distribution. is the feature with the largest variance in the low-frequency feature set and is used for normalization. This formula is introduced by This special item adaptively adjusts the disturbance amplitude according to the distribution characteristics of the features, effectively improving the perturbation's targeting and effectiveness.

[0042] Furthermore, the perturbed feature set , recalculate its entropy value To evaluate the disturbance effect. To reduce formula duplication, the entropy calculation formula here follows the definition of step S2, but the distribution of the features after disturbance is recalculated.

[0043] In addition, according to the control requirements of complexity, the disturbance complexity index is designed: ; in, Representation characteristics The size of the possible value set of , which is used to evaluate feature complexity. is the square of the perturbation amplitude, indicating the perturbation energy. Complexity index It can be used in subsequent steps to limit the negative effects of excessive perturbations.

[0044] Understandably, the output perturbed low-frequency feature set , providing input for subsequent confusion feature generation. Output updated entropy value set , used to evaluate the optimization results. Output complexity index , providing a basis for controlling disturbance resource consumption.

[0045] S4. Construct a generative adversarial network based on the high-frequency feature set and the perturbed low-frequency feature set to enhance the interference capability to the tracking system while maintaining the integrity of the fingerprint browser function. The generative adversarial network finally outputs a confusion feature set.

[0046] Specifically, we construct a generative adversarial network (GAN), where the generator and the discriminator The input and target are designed as follows: Generator :take over and , output confusion feature set The goal of the generator is to generate features that are highly random and indistinguishable while preserving key functional characteristics.

[0047] Discriminator : Receive a feature set and output a probability value, indicating that the input feature is a true feature set (such as ). The goal of the discriminator is to maximize the ability to distinguish between real features and generated features.

[0048] Furthermore, in order to balance the authenticity and randomness of the generated features, an innovative optimization objective function is introduced: ; Item 1 Represents the ability of the generator to confuse the discriminator. The generator hopes to confuse the feature set Misclassified by the discriminator as a real feature.

[0049] Item 2 is a regularization term used to improve the randomness of the obfuscation feature, where Represents the average entropy value of the confusion feature set: ; This guides the generator to output high entropy features, making the generated features more difficult to be exploited by the tracking system.

[0050] Furthermore, the goal of the discriminator is to distinguish between real features and generated confusing features, and its optimization goal is: ; in, Represents a real feature set (such as and part ). Represents the generated confusion feature set.

[0051] Furthermore, in order to adapt to the dynamic changes of the tracking system, a dynamic feedback mechanism is designed: Collect behavioral feedback from the tracking system, such as The recognition rate changes.

[0052] Using feedback information to dynamically adjust the weights in the generator optimization objective , to adapt to the evolution of tracking algorithms and improve the robustness of obfuscated features.

[0053] Understandably, the output confusion feature set , providing input for subsequent dynamic feedback and privacy protection deployment. Output optimized generator and the discriminator , as a core module, supports long-term operation.

[0054] S5. Collect dynamic feedback information from the tracking system in real time, including the actual recognition rate and the distribution characteristics of the actual recognition samples, calculate the feedback deviation between the target recognition rate set by the system and the actual recognition rate, and dynamically adjust the optimized generator according to the feedback deviation The regular term weight of the objective function of The loss function is used, and the dynamic feedback information is added to the optimized discriminator Maintain the ability to distinguish between true feature sets and confused feature sets, thereby optimizing the discriminator The loss function is to use the optimized generator Loss function and optimized discriminator The loss function for the generator and the discriminator Iterative training is performed to finally output an updated generative adversarial network and a verified set of confusion features.

[0055] Specifically, collect dynamic feedback information from the tracking system, including recognition rate (The tracking system correctly identifies the obfuscated features probability) and identification sample distribution characteristics .

[0056] Where, feedback deviation is defined : ; in, is the target recognition rate set by the system (e.g. ), which represents the upper limit of the expected confusion feature set to be identified. It is the optimization requirement of the current state of the model. The larger the deviation, the insufficient the current confusion effect, and the need to increase randomness and complexity to interfere with the tracking system.

[0057] Furthermore, according to the feedback bias Dynamically adjust the regularization term weight of the generator objective function, and the optimized generator objective function is: ; Item 1 is the main GAN loss for the generator, ensuring that the obfuscated features are difficult to distinguish for the discriminator.

[0058] Item 2 It is a feedback-driven random regularization term that improves the entropy and randomness of the obfuscated features through dynamically adjusted regularization weights: ; in, It is a confusing feature The entropy value of is used to measure the randomness of features. It is a confusing feature The distribution variance of ensures the diversity of features. is the characteristic distribution diversity adjustment coefficient. This design combines entropy and variance. Dynamic adjustment enables the generator to adapt to the evolution of the tracking system in real time.

[0059] Furthermore, the discriminator objective maintains the ability to distinguish between the true feature set and the confused feature set, while adding feedback information to optimize the loss function: ; The first and second terms are standard GAN losses, which measure the probability that a real feature is judged to be real and the probability that a confused feature is judged to be generated, respectively.

[0060] Item 3 is a feedback control regularization term used to limit the overfitting of the discriminator to the confusing features: ; This term controls the output of the discriminator to deviate from the central value. to avoid overfitting and simplification.

[0061] Understandably, using the updated and , for the generator and the discriminator Perform iterative training: In each round of training, sample the latest feedback data and , dynamically update weights and .

[0062] The outputs of the optimized generator and discriminator are reused to verify the feedback bias , ensuring that the update effect gradually approaches the target state.

[0063] Finally, verify the updated generator and the discriminator Are the following conditions met: Recognition rate of confusing feature set .

[0064] The average entropy of the confusion feature set Reach the set range.

[0065] The final output is the optimized generator and the discriminator , and a collection of verified obfuscation features , used for subsequent privacy protection deployment.

[0066] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of the present invention. In practical applications, technicians in this field can select part or all of them according to actual needs to achieve the purpose of the present embodiment, and no limitation is made here.

[0067] In addition, for technical details that are not described in detail in this embodiment, reference can be made to the parameter operation method provided in any embodiment of the present invention, and will not be repeated here.

[0068] Other embodiments or specific implementations of the data protection device for a fingerprint browser of the present invention can refer to the above-mentioned method embodiments and will not be described in detail here.

[0069] In addition, an embodiment of the present invention further proposes a storage medium, on which is stored a data protection program for a fingerprint browser. When the data protection program for a fingerprint browser is executed by a processor, the steps of a data protection method for a fingerprint browser as described above are implemented.

[0070] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or system including the element.

[0071] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0072] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as a read-only memory / random access memory, a magnetic disk, or an optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0073] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A data protection method for a fingerprint browser, characterized in that: The method comprises: S1. Input device fingerprint data to construct a device fingerprint data set, wherein each feature in the device fingerprint data set represents a characteristic of a user device, and the contribution of each feature to user discrimination is calculated by a variance normalization method to obtain a feature score of each feature. A stratification threshold is designed according to the feature score, and the device fingerprint features are divided into two layers, including a high-frequency feature set and a low-frequency feature set; S2. Calculate the entropy value of each feature according to the high-frequency feature set and the low-frequency feature set, and construct an entropy value set for each feature, and calculate the average entropy value of the low-frequency feature set and the high-frequency feature set according to the entropy value of each feature, and construct an optimization objective function based on the average entropy value of the low-frequency feature set and the high-frequency feature set to balance the computational overhead of privacy protection, functional integrity and feature disturbance, and obtain the expected target entropy value of each feature; S3. Design a low-frequency feature perturbation formula for the low-frequency feature set, obtain the low-frequency feature value after perturbation and the low-frequency feature set after perturbation, recalculate the entropy value set of the low-frequency feature set after perturbation to evaluate the perturbation effect, and obtain a complexity index; S4. Constructing a generative adversarial network based on the high-frequency feature set and the perturbed low-frequency feature set to enhance the interference capability to the tracking system while maintaining the integrity of the fingerprint browser function. The generative adversarial network finally outputs a confusion feature set. Among them, the target design of the generative adversarial network is as follows: Generator Receiving a high-frequency feature set and a perturbed low-frequency feature set and outputting a confused feature set; Discriminator Receive a set of confused features and output a probability value, indicating the possibility that the input features are the true feature set; Output optimized generator and the discriminator ; S5. Collect dynamic feedback information from the tracking system in real time, including the actual recognition rate and the distribution characteristics of the actual recognition samples, calculate the feedback deviation between the target recognition rate set by the system and the actual recognition rate, and dynamically adjust the optimized generator according to the feedback deviation The regular term weight of the objective function of The loss function is used, and the dynamic feedback information is added to the optimized discriminator Maintain the ability to distinguish between true feature sets and confused feature sets, thereby optimizing the discriminator The loss function is to use the optimized generator The loss function and optimized discriminator The loss function for the generator and the discriminator Iterative training is performed to finally output an updated generative adversarial network and a verified set of confusion features.

2. A data protection method for a fingerprint browser according to claim 1, characterized in that: The variance normalization method is used to calculate the contribution of each feature to the user discrimination, and obtain the feature score of each feature, which is specifically: ； in, Representation characteristics The variance of the values ​​in the device fingerprint dataset reflects the distinguishability of the feature. The largest score indicates that the feature is most important in user identification. Score the importance of feature i and construct a score set , Representation characteristics The variance of the values ​​in the device fingerprint dataset; n is the total number of features; The high-frequency feature set includes all scores Features, is the stratification threshold; The low-frequency feature set includes all scores Features; At the same time, design dynamic layered updates: For scenarios with high real-time requirements, a sliding window dynamic update mechanism is designed to re-evaluate the feature importance score based on the user's latest interactive behavior. The updated score calculation formula is: ； in, is the weight coefficient, which is used to balance the impact of new data and historical data. It is the result of the previous round of scoring; the updated ultra-high frequency feature set and low-frequency feature set adapt to the user's latest device status and improve the dynamics of feature stratification.

3. A data protection method for a fingerprint browser according to claim 2, characterized in that: The entropy value of each feature is calculated according to the high-frequency feature set and the low-frequency feature set, and is expressed as: ； in, Representation characteristics The highest information entropy indicates that the feature has the greatest randomness and the least dependence on user identification; It is a feature The possible value set of Is the value probability; is the weight coefficient of the compensation term, which is used to balance the essential information of the entropy value and the fingerprint protection requirements; is a randomness compensation term, which is used to increase the entropy value for certain specific features and is defined as: ； in, It is a feature The number of possible values ​​of represents the inherent diversity of the feature; is a set of low-frequency features; The entropy value of each feature The average entropy value is calculated for the low-frequency feature set and the high-frequency feature set respectively, which is expressed as: ； in, It is the average entropy value of low-frequency features, which mainly reflects the randomness of the feature set with a lower protection level; It is the average entropy value of high-frequency features, which mainly reflects the randomness of the functional priority feature set; is the number of low-frequency features, is the number of high-frequency features; is a set of low-frequency features; is a set of high-frequency features; The optimization objective function , expressed as: ； in, It is the trade-off coefficient between privacy and functionality, which determines the emphasis on protection strength and functional priority; is the average entropy of low-frequency features, is the regularization weight coefficient, which is used to limit the complexity of feature perturbations and avoid unnecessary computational overhead; is the perturbation complexity regularization term, which is used to limit the overhead of low-frequency features in the randomization process and is defined as: ； in, represents the number of possible values ​​of the feature, the relative complexity required to represent the characteristic perturbation; It is a feature The variance of represents the scalability of its original distribution. The smaller the variance, the greater the difficulty of perturbation.

4. A data protection method for a fingerprint browser according to claim 1, characterized in that: The S3 also includes calculating the entropy difference of each feature according to the expected target entropy value of each feature and the actual entropy value of each feature, wherein the disturbance intensity is proportional to the entropy difference of each feature, and the feature with the largest randomness difference is preferentially assigned the largest disturbance weight.

5. A data protection method for a fingerprint browser according to claim 4, characterized in that: The S3 specifically includes: For low-frequency feature sets The low-frequency characteristic perturbation formula is expressed as: ； in, is the perturbed eigenvalue, is the original eigenvalue; is a random disturbance term, from a normal distribution Medium sampling, Dynamically adjust to control disturbance amplitude; Representation characteristics The variance of reflects its distribution characteristics; It is a patented regularization factor that controls the dynamic adjustment of the perturbation amplitude with the feature distribution, ensuring that more significant perturbations are applied to features with more uniform distribution; It is the feature with the largest variance in the low-frequency feature set and is used for normalization; The perturbed feature set Design perturbation complexity index , recalculate its entropy value To evaluate the perturbation effect, it is expressed as: ； in, Representation characteristics The size of the possible value set of is used to evaluate feature complexity; is the square of the disturbance amplitude and represents the disturbance energy.

6. A data protection method for a fingerprint browser according to claim 1, characterized in that: The target design of the generative adversarial network is specifically: Generator Optimization goals design: ； Item 1 Represents the ability of the generator to confuse the discriminator. The generator hopes to confuse the feature set Misidentified by the discriminator as a real feature; Represents the probability value of the discriminator judging whether the confused feature is a real feature; Item 2 is a regularization term used to improve the randomness of the obfuscated features. represents the weight coefficient of the regularization term, where Represents the average entropy value of the confusion feature set: ； in, is the entropy value of each feature; Discriminator D optimization goal design: ； in, represents the real feature set, Represents the generated confusion feature set.

7. A data protection method for a fingerprint browser according to claim 6, characterized in that: The optimized generator is dynamically adjusted according to the feedback deviation The regular term weight of the objective function of The loss function includes: ； in, For the optimized generator The loss function of Item 1 is the main GAN loss of the generator, ensuring that the obfuscated features are difficult for the discriminator to distinguish; Item 2 It is a feedback-driven random regularization term that improves the entropy and randomness of the obfuscated features through dynamically adjusted regularization weights: ； in, It is a confusing feature The entropy value of , which measures the randomness of features; It is a confusing feature The distribution variance of ensures the diversity of features; is the characteristic distribution diversity adjustment coefficient; At the same time, dynamic feedback information is added to the optimized discriminator Maintain the ability to distinguish between true feature sets and confused feature sets, thereby optimizing the discriminator The loss function is expressed as: ； in, is the optimized discriminator The loss function of The first and second terms are the standard GAN losses, which measure the probability that a true feature is judged as true and the probability that a confused feature is judged as generated, respectively; Item 3 is a feedback control regularization term used to limit the overfitting of the discriminator to the confusing features: ； The term controls the output of the discriminator to deviate from the central value to avoid overfitting and simplification.

8. A data protection method for a fingerprint browser according to claim 7, characterized in that: The S5 also includes verifying the updated generator and the discriminator Are the following conditions met: Recognition rate of confusing feature set ; The average entropy of the confusion feature set Reach the set range; If the set range is not reached or the recognition rate is not reached Then the feedback bias is recalculated to train the generative adversarial network.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images