Privacy budget allocation method, device, equipment, and medium based on property data

Abstract

The present invention proposes a privacy budget allocation method, device, equipment, and medium based on property data, the method comprising: constructing a first input data set based on multiple first property data; determining a first correlation value between each pair of first property data based on data categories; determining a second difference value from the first difference value, determining a second reasoning difficulty value from the first reasoning difficulty value, and determining a second correlation value from the first correlation value; determining a target strategy based on the second difference value, the second reasoning difficulty value, and the number of second correlation values; and publishing a first output data set after adjusting the privacy budget based on the target strategy. According to the technical solution of the embodiment of the present invention, it is possible to characterize the leakage risk of property data from three dimensions of reasoning risk, differences between users, and correlation of data categories, dynamically select a target strategy based on the number of parameter values ​​with higher risks in each dimension, and use the target strategy to achieve dynamic adjustment of the privacy budget, thereby improving data security.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a privacy budget allocation method, device, equipment, and medium based on property data. Background Art

[0002] At present, the degree of informatization in the property sector is getting higher and higher. Common property data includes household information, property service records, financial records, equipment monitoring data, and community environment data. The informatization of property management can improve the efficiency of property management, but there is also a risk of data leakage. The differential privacy protection algorithm is a commonly used data security algorithm that can allocate a privacy budget for each input data and increase noise based on the privacy budget. The smaller the privacy budget, the greater the noise, which effectively improves data security.

[0003] Reasoning attacks and reconstruction attacks are two common attack methods. After requesting non-sensitive data multiple times, the original sensitive data is deduced by using the logical relationship between non-sensitive data, resulting in privacy leakage. The related technology introduces an adaptive privacy budget allocation method based on gradient descent, which continuously reduces the privacy budget through gradient adjustment and dynamically adjusts the noise of the output data. Although it can play a certain protective role, property data is divided into household data and management data. The correlation between household data and household information is strong, while the correlation between management data and household information is weak. The gradient of the related technology is mainly adjusted by decreasing the learning rate over time. The same descent strategy is used for all output data. If a reasoning attack is launched against household data with strong correlation, the risk of privacy leakage is still high. Summary of the invention

[0004] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention proposes a privacy budget allocation method, device, equipment, and medium based on property data, which can dynamically select a target strategy according to property data with inference risks, realize dynamic adjustment of the privacy budget, and improve data security.

[0005] In a first aspect, an embodiment of the present invention provides a privacy budget allocation method based on property data, which is applied to a property management system, wherein the property management system is preset with multiple data categories and multiple household information, each of the data categories includes multiple optional property data, each of the optional property data is associated with a first difference value and a first reasoning difficulty value, the first difference value is used to indicate the degree of difference between the corresponding optional property data between different households, and the first reasoning difficulty value is used to indicate the difficulty of inferring the household information based on the corresponding optional property data, including:

[0006] In response to a first data request, determining a plurality of first property data from a plurality of the selectable property data, and constructing a first input data set based on the plurality of the first property data;

[0007] Based on each pair of the data categories to which the first property data belongs, a plurality of first correlation values ​​are obtained from a preset first mapping table, wherein the first mapping table records preset correlation values ​​between the data categories;

[0008] Based on the first input data set, determining the first difference value greater than a preset first threshold as a second difference value, determining the first reasoning difficulty value greater than a preset second threshold as a second reasoning difficulty value, and determining the first correlation value greater than a preset third threshold as a second correlation value;

[0009] Determining a target strategy from a plurality of preset strategies based on a first number of second difference values, a second number of the second reasoning difficulty values, and a third number of the second correlation values, wherein the preset strategy is used to adjust the privacy budget, and an adjustment range of the preset strategy is positively correlated with the first number, the second number, and the third number;

[0010] A first privacy budget of the first input data set is determined based on a preset privacy budget allocation model, the first privacy budget is adjusted to a second privacy budget based on the target strategy, and a first output data set is published based on the second privacy budget and the first input data set.

[0011] According to some embodiments of the present invention, the preset strategy includes a first strategy, a second strategy, and a third strategy, the maximum adjustment range indicated by the first strategy is greater than that of the second strategy, the maximum adjustment range indicated by the second strategy is greater than that of the third strategy, the maximum value of the first correlation value, the first reasoning difficulty value, and the first correlation value is 1, and the target strategy is determined from the plurality of preset strategies based on the first number of second difference values, the second number of the second reasoning difficulty values, and the third number of the second correlation values, including:

[0012] Determine a first reference value according to a fourth quantity of the first property data and a preset reference ratio, and determine a second reference value according to a fifth quantity of the first related values ​​and the reference ratio, wherein the reference ratio is less than 1 and greater than 0.5;

[0013] When at least one of the first number, the second number, and the third number is greater than a preset fourth threshold, determining the first strategy as the target strategy, wherein the fourth threshold is less than or equal to the fourth number;

[0014] Alternatively, when the first number and the second number are greater than or equal to the first reference value, and the third number is greater than or equal to the second reference value, the first strategy is determined as the target strategy;

[0015] Alternatively, when the first number and the second number are greater than or equal to the first reference value, and the third number is less than the second reference value, the second strategy is determined as the target strategy;

[0016] Alternatively, when the third number is greater than or equal to the second reference value, and either the first number or the second number is greater than or equal to the first reference value, the second strategy is determined as the target strategy;

[0017] Alternatively, when the first number and the second number are less than the first reference value, and the third number is greater than the second reference value, the third strategy is determined as the target strategy;

[0018] Alternatively, when the third number is smaller than the second reference value, and only one of the first number and the second number is larger than the first reference value, the third strategy is determined as the target strategy.

[0019] According to some embodiments of the present invention, when the target policy is the first policy, adjusting the first privacy budget to a second privacy budget based on the target policy includes:

[0020] Based on the first input data set, determining the ratio of the maximum value to the average value of the first difference value as a first initial coefficient, determining the ratio of the maximum value to the average value of the first reasoning difficulty value as a second initial coefficient; determining the ratio of the maximum value to the average value of the first correlation value as a third initial coefficient;

[0021] Based on the number of the second difference value and the first difference value being determined as a first ratio, the number ratio of the second reasoning difficulty value to the first reasoning difficulty value is determined as a second ratio, and the number ratio of the second correlation value to the first correlation value is determined as a third ratio;

[0022] Determine the product of the first ratio and the first initial coefficient as a first intermediate coefficient, determine the product of the second ratio and the second initial coefficient as a second intermediate coefficient, and determine the product of the third ratio and the third initial coefficient as a third intermediate coefficient;

[0023] determining a sum of the first intermediate coefficient, the second intermediate coefficient and the third intermediate coefficient as a target coefficient;

[0024] The first privacy budget is reduced to the second privacy budget based on the target coefficient.

[0025] According to some embodiments of the present invention, when the target policy is the second policy, adjusting the first privacy budget to the second privacy budget based on the target policy includes:

[0026] When the second strategy is determined based on the first quantity and the second quantity, the first difference value is determined as a first parameter, and the first reasoning difficulty value is determined as a second parameter;

[0027] When the second strategy is determined based on the third quantity, the first correlation value is determined as the first parameter, and the first reasoning difficulty value or the first difference value is determined as the second parameter, wherein when the second strategy is also determined based on the first quantity, the second parameter is the first difference value, and when the second strategy is also determined based on the second quantity, the second parameter is the first reasoning difficulty value;

[0028] Based on a preset clustering algorithm, clustering a plurality of the first parameters into at least two first clustering clusters, and clustering a plurality of the second parameters into at least two second clustering clusters;

[0029] Determine the cluster with the largest number of the first cluster and the second cluster as a target cluster, and determine an average value calculated based on the target cluster as a target coefficient;

[0030] The first privacy budget is reduced to the second privacy budget based on the target coefficient.

[0031] According to some embodiments of the present invention, when the target policy is the third policy, adjusting the first privacy budget to the second privacy budget based on the target policy includes:

[0032] When the third number is greater than the second reference value, determining the first correlation value as a third parameter;

[0033] Alternatively, when the first number is greater than the first reference value, the first difference value is determined as the third parameter;

[0034] Alternatively, when the second number is greater than the first reference value, the first reasoning difficulty value is determined as the third parameter;

[0035] determining the first reasoning difficulty value, the first difference value, or the first correlation value that is not determined as the third parameter as a fourth parameter;

[0036] Determine a ratio of a maximum value to an average value of the third parameter as a fourth intermediate coefficient, determine a difference between a maximum value and a minimum value of the fourth parameter as a fifth intermediate coefficient, and determine a sum of the fourth intermediate coefficient and the fifth intermediate coefficient as a target coefficient;

[0037] The first privacy budget is reduced to the second privacy budget based on the target coefficient.

[0038] According to some embodiments of the present invention, after publishing the first output data set based on the second privacy budget and the first input data set, the method further includes:

[0039] Obtaining a first learning rate for calculating the first privacy budget, and obtaining a first attenuation coefficient preset in the privacy budget allocation model;

[0040] determining a product of the target coefficient and the first attenuation coefficient as a second attenuation coefficient;

[0041] When a second data request is obtained, a plurality of second property data are determined from the plurality of optional property data, and a second input data set is constructed based on the plurality of second property data;

[0042] When the data category of the second property data is the same as the data category of the first property data, determining a second learning rate based on the first decay coefficient and the first learning rate;

[0043] Alternatively, when at least one of the data categories of the second property data does not belong to the data category of the first property data, the second learning rate is determined based on the second decay coefficient and the first learning rate;

[0044] The second learning rate is applied in the privacy budget allocation model to determine a third privacy budget of the second input dataset, and a second output dataset is published based on the third privacy budget and the second input dataset.

[0045] According to some embodiments of the present invention, the property management system is pre-associated with a plurality of resident terminals, each of the resident terminals is uniquely associated with the resident information, and after constructing a second input data set based on the plurality of second property data, the method further comprises:

[0046] When the initiating terminal of the first data request and the initiating terminal of the second data are the same resident terminal, determining the second learning rate based on the first attenuation coefficient and the first learning rate;

[0047] Alternatively, when the initiating terminal of the first data request and the second data does not belong to the resident terminal, the second learning rate is determined based on the second attenuation coefficient and the first learning rate.

[0048] In a second aspect, an embodiment of the present invention provides a privacy budget allocation device based on property data, comprising at least one control processor and a memory for communicating with the at least one control processor; the memory stores instructions executable by the at least one control processor, and the instructions are executed by the at least one control processor so that the at least one control processor can execute the privacy budget allocation method based on property data as described in the first aspect above.

[0049] In a third aspect, an embodiment of the present invention provides an electronic device, comprising the property data-based privacy budget allocation device as described in the second aspect above.

[0050] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions are used to execute the privacy budget allocation method based on property data as described in the first aspect above.

[0051] The privacy budget allocation method based on property data according to an embodiment of the present invention has at least the following beneficial effects: in response to a first data request, a plurality of first property data are determined from a plurality of the optional property data, and a first input data set is constructed based on the plurality of the first property data; based on the data categories to which each pair of the first property data belongs, a plurality of first correlation values ​​are obtained from a preset first mapping table, wherein the first mapping table records the preset correlation values ​​between the data categories; based on the first input data set, the first difference value greater than a preset first threshold is determined as a second difference value, the first reasoning difficulty value greater than a preset second threshold is determined as a second reasoning difficulty value, and the first correlation value greater than a preset third threshold is determined as a second correlation value; based on the first number of second difference values, the second number of second reasoning difficulty values, and the third number of second correlation values, a target strategy is determined from a plurality of preset strategies, wherein the preset strategy is used to adjust the privacy budget; based on a preset privacy budget allocation model, a first privacy budget of the first input data set is determined, based on the target strategy, the first privacy budget is adjusted to the second privacy budget, and a first output data set is published based on the second privacy budget and the first input data set. According to the technical solution of the embodiment of the present invention, it is possible to characterize the leakage risk of property data from three dimensions: inference risk, differences between users, and correlation of data categories. The target strategy can be dynamically selected based on the number of parameter values ​​with higher risks in each dimension, and the target strategy can be used to achieve dynamic adjustment of the privacy budget to improve data security. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 is a schematic diagram of a privacy budget allocation method based on property data provided by an embodiment of the present invention;

[0053] Figure 2 is a flow chart of a privacy budget allocation method based on property data provided by another embodiment of the present invention;

[0054] Figure 3 is a schematic diagram of target key points and facial features areas provided by another embodiment of the present invention;

[0055] Figure 4 It is a structural diagram of a privacy budget allocation device based on property data provided by another embodiment of the present invention. DETAILED DESCRIPTION

[0056] 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.

[0057] In the description of the present invention, it should be understood that descriptions involving orientations, such as up, down, front, back, left, right, etc., and orientations or positional relationships indicated are based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the present invention.

[0058] In the description of the present invention, "several" means one or more, "more" means more than two, "greater than", "less than", "exceed" etc. are understood as not including the number itself, and "above", "below", "within" etc. are understood as including the number itself. If there is a description of "first" or "second", it is only used for the purpose of distinguishing the technical features, and cannot be understood as indicating or implying the relative importance or implicitly indicating the number of the indicated technical features or implicitly indicating the order of the indicated technical features.

[0059] In the description of the present invention, unless otherwise clearly defined, terms such as setting, installing, connecting, etc. should be understood in a broad sense, and technicians in the relevant technical field can reasonably determine the specific meanings of the above terms in the present invention based on the specific content of the technical solution.

[0060] The embodiments of the present invention provide a privacy budget allocation method, apparatus, device, and medium based on property data, wherein the privacy budget allocation method based on property data includes: in response to a first data request, determining multiple first property data from multiple optional property data, and constructing a first input data set based on the multiple first property data; obtaining multiple first correlation values ​​from a preset first mapping table based on the data category to which each pair of the first property data belongs, wherein the first mapping table records the preset correlation values ​​between the data categories; based on the first input data set, determining the first difference value greater than a preset first threshold as a second difference value, and converting the first inference difficulty value greater than a preset second threshold into a second difference value; The first input data set is determined as the second reasoning difficulty value, and the first correlation value greater than the preset third threshold is determined as the second correlation value; based on the first number of second difference values, the second number of the second reasoning difficulty values ​​and the third number of the second correlation values, a target strategy is determined from a plurality of preset strategies, wherein the preset strategy is used to adjust the privacy budget, and the adjustment range of the preset strategy is positively correlated with the first number, the second number and the third number; based on a preset privacy budget allocation model, a first privacy budget of the first input data set is determined, based on the target strategy, the first privacy budget is adjusted to the second privacy budget, and a first output data set is published based on the second privacy budget and the first input data set. According to the technical solution of the embodiment of the present invention, the leakage risk of property data can be characterized from three dimensions of reasoning risk, difference between users and correlation of data categories, and the target strategy can be dynamically selected in combination with the number of parameter values ​​with higher risks in each dimension, and the dynamic adjustment of the privacy budget can be achieved by using the target strategy to improve data security.

[0061] The technical solution of the embodiment of the present invention is further described below based on the accompanying drawings.

[0062] Reference Figure 1 and Figure 2 The property management system of the embodiment of the present invention is preset with multiple data categories and multiple household information, each data category includes multiple optional property data, each optional property data is associated with a first difference value and a first reasoning difficulty value, the first difference value is used to indicate the degree of difference between the corresponding optional property data of different households, the first reasoning difficulty value is used to indicate the difficulty of inferring the household information based on the corresponding optional property data, and the privacy budget allocation method based on property data includes but is not limited to the following steps:

[0063] S10, in response to a first data request, determining a plurality of first property data from a plurality of optional property data, and constructing a first input data set based on the plurality of first property data;

[0064] S20, acquiring a plurality of first correlation values ​​from a preset first mapping table based on the data category to which each pair of first property data belongs, wherein the first mapping table records preset correlation values ​​between data categories;

[0065] S30, based on the first input data set, determining a first difference value greater than a preset first threshold as a second difference value, determining a first reasoning difficulty value greater than a preset second threshold as a second reasoning difficulty value, and determining a first correlation value greater than a preset third threshold as a second correlation value;

[0066] S40, determining a target strategy from a plurality of preset strategies based on a first number of second difference values, a second number of second reasoning difficulty values, and a third number of second correlation values, wherein the preset strategy is used to adjust the privacy budget, and an adjustment range of the preset strategy is positively correlated with the first number, the second number, and the third number;

[0067] S50, determining a first privacy budget of a first input data set based on a preset privacy budget allocation model, adjusting the first privacy budget to a second privacy budget based on a target strategy, and publishing a first output data set based on the second privacy budget and the first input data set.

[0068] It should be noted that the data category can be Figure 1 The resident information, property service records, financial information, and monitoring data shown may also include environmental monitoring data detected by sensors installed in the community, which is not limited here. Each data category includes multiple optional property data, such as Figure 1 As shown in , the optional property data corresponding to the resident information may be the resident's name, contact information, and resident address, etc., and no limitation is made on the specific data types and names here.

[0069] It should be noted that the first data request can be initiated through the property management APP. For example, the fee inquiry of a certain unit is carried out through the property management APP. The requested output data includes the resident's name, resident's address, property fee payment record, fee details, etc. The property management system can determine the above-mentioned optional property data by parsing the first data request, use it as the first property data, and construct a first input data set.

[0070] It should be noted that each optional property data is pre-set with a first difference value and a first inference difficulty value according to the actual situation. The first difference value represents the degree of difference between different households, and the first inference difficulty value represents the difficulty of inferring household information. In addition, a first correlation is pre-set between each data category, such as Figure 1As shown, the optional property data is classified according to the data category, so the optional property data under the same data category are usually highly correlated. Therefore, there is no need to set a preset correlation for each pair of optional property data to avoid the first mapping table being too bloated. In this embodiment, the first correlation value is determined only by category. The specific values ​​of the first difference value, the first reasoning difficulty value, and the preset correlation value can be normalized to ensure that the values ​​are unified in the subsequent calculation process.

[0071] For example, the first difference value has 1 as the maximum value, and the first difference degree of 1 indicates that the data has the largest difference between different households, such as Figure 1 As shown, for example, the optional property data is the name of the resident. The names of different residents are usually different. Therefore, the optional property data of the resident name of two residents is relatively different, and the first difference value corresponding to the resident name is 1. For another example, the resident addresses of different residents in the same community are usually similar, and the difference mainly lies in the specific residential units. Therefore, the optional property data of the resident address is relatively small, and the first difference value corresponding to the resident address can be set to 0.1. The first difference value of each optional property data can be set according to the above principles and actual needs.

[0072] For example, the first reasoning difficulty value also takes 1 as the maximum value, and the first reasoning difficulty value of 1 indicates that the optional property data can easily be used as the reasoning basis for the resident information, such as Figure 1 As shown, for example, the optional property data is the expense details in the financial information. It is difficult to infer the household information such as the household name or contact information from the expense details, so the first reasoning difficulty value corresponding to the expense details can be 0.1; for another example, the household information data category is directly the household information itself, so the first reasoning difficulty of each optional property data under it can be set to 1; for another example, the property fee payment record can be used to infer the household information such as the household address, and can be used as the basis for reasoning, so the first reasoning difficulty value of the property fee payment record can be set to 0.7. The first reasoning difficulty value of each optional property data can be set according to the above principles and actual needs.

[0073] Exemplarily, the preset correlation value takes 1 as the maximum value, and a preset correlation value of 1 indicates that the correlation between data categories is strong, and the optional property data of one data category can be inferred from the optional property data of another data category. Of course, multiple optional property data can come from the same data category. For example, when initiating a household details query, the optional property data are the household name, contact information and household address, all of which come from the data category of household information. Therefore, the preset correlation value can be set to 1 in the first mapping table; for another example, the content recorded in the various optional property data under the property service record is highly correlated with the household information, and the preset correlation value between the property service record and the household information can be set to 0.8; for another example, the correlation between monitoring data and financial information is weak, and the preset correlation value between the two is set to 0.1.

[0074] It should be noted that the preset correlation value is different from the first difference value and the first reasoning difficulty value. The first difference value and the first reasoning difficulty value are preset in each optional property data. Therefore, after determining the first input data set, the number of the first difference value and the first reasoning difficulty value is the same as the number of the optional property data, while the preset correlation value is a parameter between two optional property data, and the number of the first correlation value is a combination of the optional property data. Calculated, n is the number of optional property data, for example Figure 1 As shown in Figure 2, when the first input data set includes household names, complaint records, and property payment records, n=3, then the number of first related values ​​is =3, based on the data categories to which they belong, correlation value 1 is the preset correlation value between household information and property service records, correlation value 2 is the preset correlation value between household information and financial information, and correlation value 3 is the preset correlation value between financial information and property service records; for another example, when the first input data set includes 4 optional property data, n=4, and the number of first correlation values ​​is =6, which will not be elaborated here.

[0075] It should be noted that, according to the definitions of the above-mentioned first difference value, the first inference difficulty value and the preset correlation value, the larger the value is, the higher the risk of inferring the resident information. Therefore, this embodiment sets the first threshold, the second threshold and the third threshold to screen out the components with inference risk from each value. The second difference value is screened out from the first difference value by the first threshold. The first property data corresponding to the second difference value has large differences between each household and can be used as an exclusion condition during inference, which has a large risk of inference attack. For example, the first property data is a fee detail. After obtaining the payment records of multiple households, different residents are distinguished based on the fee details, which has a large risk of inference attack. The second inference difficulty value is screened out from the first inference difficulty value by the second threshold. The first property data corresponding to the second inference difficulty value can easily infer the resident information. For example, the first property information is the resident information itself, so the second thrust difficulty value is high, which has a high risk of inference attack. The second correlation value is screened out from the first correlation value by the third threshold. The correlation between the two data categories is high, and the first property data of one data category can be inferred from the first property data of another data category, which has a high risk of inference attack.

[0076] It should be noted that after determining the second difference value, the second reasoning difficulty value, and the second correlation value of each first property data, this embodiment selects a strategy based on the number of each value. The more the second difference value, the second reasoning difficulty value, and the second correlation value, the stronger the logic available during the reasoning attack, and therefore the greater the risk, and more noise needs to be added to the output data to ensure data security. Therefore, this embodiment sets a strategy library in the property management system, and a plurality of preset strategies are pre-set in the strategy library, each of which is used to adjust the privacy budget, and the adjustment range of the privacy budget of the plurality of preset strategies is positively correlated with the first number, the second number, and the third number, that is, the larger the first number, the second number, and the third number, the larger the adjustment range of the preset strategy matched, ensuring that appropriate noise is added to the first output data set to improve data security.

[0077] It should be noted that a corresponding number of second difference values, second thrust difficulty values ​​and second related values ​​can be pre-set for each preset strategy. After determining the first number, the second number and the third number based on the first input data set, strategy matching is performed according to the first number, the second number and the third number to ensure that the target strategy can be matched.

[0078] It should be noted that the privacy budget allocation model is an existing technology. This embodiment is based on the common privacy budget allocation model. After inputting the first input data set into the privacy budget allocation model, the common learning rate descent and gradient descent methods are used to set the initial learning rate and the initial gradient. Then, the first privacy budget is calculated based on the initial learning rate and the initial gradient. Those skilled in the art are familiar with how to calculate the privacy budget, and will not go into details here.

[0079] In addition, in one embodiment, the preset strategy includes a first strategy, a second strategy, and a third strategy, the maximum adjustment range indicated by the first strategy is greater than the second strategy, the maximum adjustment range indicated by the second strategy is greater than the third strategy, the maximum value of the first correlation value, the first reasoning difficulty value, and the first correlation value is 1, and the reference Figure 3 Step S40 specifically includes but is not limited to the following steps:

[0080] S41, determining a first reference value according to a fourth quantity of the first property data and a preset reference ratio, and determining a second reference value according to a fifth quantity of the first correlation value and the reference ratio, wherein the reference ratio is less than 1 and greater than 0.5;

[0081] S42, when at least one of the first number, the second number, and the third number is greater than a preset fourth threshold, determining the first strategy as the target strategy, wherein the fourth threshold is less than or equal to the fourth number;

[0082] S43, when the first number and the second number are greater than or equal to the first reference value, and the third number is greater than or equal to the second reference value, determining the first strategy as the target strategy;

[0083] S44, when the first number and the second number are greater than or equal to the first reference value, and the third number is less than the second reference value, determining the second strategy as the target strategy;

[0084] S45, when the third quantity is greater than or equal to the second reference value, and either the first quantity or the second quantity is greater than or equal to the first reference value, determining the second strategy as the target strategy;

[0085] S46, when the first number and the second number are less than the first reference value, and the third number is greater than the second reference value, determining the third strategy as the target strategy;

[0086] S47: When the third number is less than the second reference value, and only one of the first number and the second number is greater than the first reference value, determine the third strategy as the target strategy.

[0087] It should be noted that the present embodiment presets the first strategy, the second strategy and the third strategy, and the maximum adjustment range decreases successively. The present embodiment does not limit the adjustment range of the first strategy to be greater than the second strategy and the third strategy, but the maximum adjustment range. The adjustment range of the preset strategy of the present embodiment is positively correlated with the first quantity, the second quantity and the third quantity. The present embodiment sets multiple strategies for the purpose of screening risk scenarios through the first quantity, the second quantity and the third quantity. Each risk scenario corresponds to a prediction strategy. For example, the first strategy corresponds to a scenario with a higher risk, and a larger adjustment needs to be made to the privacy budget. However, it does not mean that the adjustment range of the privacy budget by the second strategy is necessarily smaller than that of the first strategy. The calculation methods of the first strategy and the second strategy are different, and the adjustment ranges calculated by the two can have overlapping intervals in the value range, but the maximum value that can be calculated by the first strategy is necessarily greater than that of the second strategy. Different risk scenarios can be distinguished by different preset strategies to ensure that the maximum adjustment range that can be executed in a higher risk scenario is greater than that in a lower risk scenario.

[0088] This embodiment makes adjustments based on scenarios with higher risks, so the adjustment direction is to reduce, that is, to reduce the first privacy budget to the second privacy budget. The smaller the privacy budget, the greater the noise added to the output data, and the better the data security is protected. However, excessive noise will also lead to poor data readability. Therefore, this embodiment performs strategy matching based on the parameter values ​​of the three dimensions based on the first quantity, the second quantity, and the third quantity, to ensure that the matched target strategy meets the current risk scenario.

[0089] It should be noted that, according to the description of the above embodiment, the first correlation value is calculated based on the fourth quantity of the first property data. The combination is calculated, so the number of the first correlation value is equal to the first number and the second number only when n=3. Therefore, the third number of the second correlation value is likely to be greater than the first number and the second number. On this basis, this embodiment uses the reference ratio to determine the first reference value according to the fourth number and the second reference value according to the fifth number. It can dynamically adjust the first reference value and the second reference value when the fourth number is different, ensuring that the strategy matching can be carried out based on the difference value, reasoning difficulty value and correlation value with the correct benchmark, ensuring that the target strategy is applicable to the current risk scenario.

[0090] It should be noted that, in this embodiment, a fourth threshold is further set based on the reference ratio, and the fourth threshold is used as a condition for directly triggering the first strategy with the highest risk. Therefore, the value of the fourth threshold can be equal to the fourth number. For example, when the fourth number is 4, the fourth threshold can be 4 or 3. When any of the first number, the second number and the third number is greater than the fourth threshold, the parameters of one dimension (such as the difference value) are almost all high-risk data. Even if the parameters of other dimensions (such as the reasoning difficulty value and the correlation value) are less, there will be a relatively high risk of reverse inference, and the first strategy will be determined as the target strategy.

[0091] It is worth noting that the reference ratio is greater than 0.5 and less than 1, which can ensure that the values ​​of the first reference value and the second reference value are greater than half of the corresponding quantities, thereby representing a higher-risk scenario through the first reference value and the second reference value. For example, when the first quantity and the second quantity are greater than the first reference value, the proportion of the second difference value and the second reasoning difficulty value both exceeds 50%, indicating a higher risk. The same is true when the third quantity is greater than the second reference value.

[0092] It should be noted that in step S43, if the first number, the second number, and the third number are all less than the fourth threshold, then no type of data can directly trigger the first strategy, but this does not mean that the risk is small. When the first number and the second number are greater than the first reference value, and the third number is greater than the second reference value, it can be determined that the data in each dimension has a high risk and needs to adopt the first strategy with the largest adjustment range. Therefore, this embodiment is based on the numerical comparison of the first number, the second number, and the third number with the dynamically determined first reference value and the second reference value, and uses the number of high-risk data as a condition for selecting a strategy. When there are more high-risk data, a strategy with a larger adjustment range can be adopted to improve data security.

[0093] It should be noted that in step S44 and step S45, when two of the first quantity, the second quantity and the third quantity meet the conditions, it can be determined that the data in two dimensions have higher risks and the data in the other dimension has lower risks. For example, in step S44, the first quantity and the second quantity meet the condition of being greater than the first reference value, the first property data has a higher risk in the difference value dimension and the reasoning difficulty value dimension, while the third quantity does not meet the condition of being greater than the second reference value, and the risk in the relevant value dimension is controllable. Therefore, the second strategy with a smaller adjustment range than the first strategy can be adopted as the target strategy to avoid excessive adjustment based on the first strategy.

[0094] Similarly, in steps S46 and S47, any one of the first quantity, the second quantity and the third quantity meets the condition, and the other two do not meet the condition, that is, only one dimension of data has a higher risk, and the data of the other two dimensions have a lower risk. For example, in step S46, the first quantity and the second quantity do not meet the condition of being greater than the first reference value, and the third quantity meets the condition of being greater than the second reference value. Therefore, the risk of the first property data in the difference value dimension and the reasoning difficulty value dimension is controllable, and the risk in the related value dimension is high. Although the first privacy budget can be adjusted, the adjustment range is small, which improves data readability while ensuring data security.

[0095] In addition, in one embodiment, referring to Figure 3 When the target strategy is the first strategy, step S50 specifically includes but is not limited to the following steps:

[0096] S511, based on the first input data set, determining the ratio of the maximum value to the average value of the first difference value as the first initial coefficient, determining the ratio of the maximum value to the average value of the first reasoning difficulty value as the second initial coefficient; determining the ratio of the maximum value to the average value of the first correlation value as the third initial coefficient;

[0097] S512, based on the number of the second difference value and the first difference value being determined as the first ratio, the number ratio of the second reasoning difficulty value to the first reasoning difficulty value is determined as the second ratio, and the number ratio of the second correlation value to the first correlation value is determined as the third ratio;

[0098] S513, determining the product of the first ratio and the first initial coefficient as the first intermediate coefficient, determining the product of the second ratio and the second initial coefficient as the second intermediate coefficient, and determining the product of the third ratio and the third initial coefficient as the third intermediate coefficient;

[0099] S514, determining the sum of the first intermediate coefficient, the second intermediate coefficient and the third intermediate coefficient as the target coefficient;

[0100] S515: Reduce the first privacy budget to a second privacy budget based on the target coefficient.

[0101] It should be noted that, in the first strategy, the initial coefficient is first determined based on the ratio of the maximum value to the average value. Taking the first difference value as an example, according to the definition of the above embodiment, the maximum value that can be set for the first difference value can be 1. Therefore, the average value of the first difference value must be less than 1, and the maximum value must be greater than the average value. Therefore, the calculated first initial coefficient must be greater than 1. The first initial coefficient represents the degree of difference between the maximum risk and the average risk. The second initial coefficient and the third initial coefficient are the same. The calculation method of the initial coefficient is used to realize the dynamic determination of the target coefficient, which makes the target coefficient more flexible and accurate.

[0102] It should be noted that the number of second difference values ​​is the above-mentioned first number, and the first number is less than or equal to the number of first difference values. Therefore, the calculated first ratio must be less than 1. However, the basis for adopting the first strategy is that the number of second difference values ​​is large. Therefore, although the first ratio is less than 1, it will at least be greater than 0.5. This can not only avoid the target coefficient being too large, resulting in a large decrease in the second privacy budget and affecting data readability, but also can correct the first initial coefficient through the first ratio that represents the proportion of high-risk data, so that the corrected first intermediate coefficient can be strongly correlated with the degree of risk, thereby improving reliability. The same applies to the second intermediate coefficient and the third intermediate coefficient.

[0103] It should be noted that after the first intermediate coefficient, the second intermediate coefficient and the third intermediate coefficient are corrected and determined using the numerical difference and the proportion of high-risk data, the values ​​of the three may be less than 1, but at least greater than 0.5. Therefore, this embodiment determines the sum of the three as the target coefficient, so that the target coefficient can better reflect the risks of the three dimensions, and the target coefficient is at least greater than 1.5. The second privacy budget can be the product of the first privacy budget and the target coefficient. Noise is added with a smaller second privacy budget to improve data security.

[0104] In addition, in one embodiment, referring to Figure 3 When the target strategy is the second strategy, step S50 specifically includes but is not limited to the following steps:

[0105] S521, when the second strategy is determined based on the first quantity and the second quantity, the first difference value is determined as the first parameter, and the first reasoning difficulty value is determined as the second parameter;

[0106] S522, when the second strategy is determined based on the third quantity, the first correlation value is determined as the first parameter, and the first reasoning difficulty value or the first difference value is determined as the second parameter, wherein when the second strategy is also determined based on the first quantity, the second parameter is the first difference value, and when the second strategy is also determined based on the second quantity, the second parameter is the first reasoning difficulty value;

[0107] S523, clustering the plurality of first parameters into at least two first clustering clusters, and clustering the plurality of second parameters into at least two second clustering clusters based on a preset clustering algorithm;

[0108] S524, determining the cluster with the largest number of clusters in the first cluster and the second cluster as the target cluster, and determining the average value calculated based on the target cluster as the target coefficient;

[0109] S525: Reduce the first privacy budget to a second privacy budget based on the target coefficient.

[0110] It should be noted that the first parameter of this embodiment is a value used to determine the second strategy. For example, when determining the second strategy according to the above step S44, the first parameter is the first correlation value corresponding to the first quantity and the first reasoning difficulty value corresponding to the second quantity, and so on.

[0111] It should be noted that in the second strategy, the first number, the second number, and the third number are less than the triggering condition of the first strategy. Therefore, the risk corresponding to the first strategy is lower. This embodiment performs clustering in the first parameter and the second parameter, and clusters the first parameter into at least two first clustering clusters by setting the number of clustering clusters. Similarly, at least two second clustering clusters are obtained based on the second parameter. The cluster with the largest number is determined as the target cluster, and the target cluster is used to characterize the interval with the highest risk of inference attack, so as to calculate the average value as the target coefficient. The second strategy belongs to a medium-risk scenario, and does not necessarily need to perform aggressive adjustments based on the maximum adjustment range, but needs to consider the risk distribution. Therefore, the use of a clustering algorithm can better perform adjustments based on the risk distribution. Clustering algorithms are technologies well known to those skilled in the art, and will not be elaborated here.

[0112] In addition, in one embodiment, referring to Figure 3 When the target strategy is the third strategy, step S50 specifically includes but is not limited to the following steps:

[0113] S531, when the third number is greater than the second reference value, determining the first correlation value as a third parameter;

[0114] S532, when the first number is greater than the first reference value, determining the first difference value as a third parameter;

[0115] S533, when the second number is greater than the first reference value, determining the first reasoning difficulty value as the third parameter;

[0116] S534, determining the first reasoning difficulty value, the first difference value, or the first correlation value that is not determined as the third parameter as the fourth parameter;

[0117] S535, determining the ratio of the maximum value to the average value of the third parameter as the fourth intermediate coefficient, determining the difference between the maximum value and the minimum value of the fourth parameter as the fifth intermediate coefficient, and determining the sum of the fourth intermediate coefficient and the fifth intermediate coefficient as the target coefficient;

[0118] S536: Reduce the first privacy budget to a second privacy budget based on the target coefficient.

[0119] It should be noted that only one parameter in the third strategy is greater than the first reference value, so the reasoning risk is lower than the scenarios corresponding to the first and second strategies. It can be determined that only one of the parameters in the three dimensions is at risk, and the other two dimensions will not cause a large reasoning risk. In this embodiment, the parameter used to trigger the third strategy is determined as the third parameter. Referring to the principle of the first initial coefficient, the fourth intermediate coefficient is determined by the ratio of the maximum value and the average value of the third parameter. The fourth intermediate coefficient represents the risk span of the third parameter. Calculating the target coefficient with the fourth intermediate coefficient can better reflect the reasoning risk corresponding to the third parameter.

[0120] Exemplarily, when the third quantity is greater than the second reference value, the first correlation value is determined as the third parameter, the first difference value and the first inference difficulty value are determined as the fourth parameter, the first property data has a strong correlation in the data category, but the inference risk of each data itself is low, for example, the first property data is Figure 1 The maintenance records, expense details, and surveillance videos shown in the figure can be used to infer the information of the resident who requested the maintenance by combining the above three parameters. The first correlation value is relatively high, but the first difference value and the first inference difficulty of the maintenance records, expense details, and surveillance videos are relatively high. It is only necessary to add noise to eliminate the correlation between data categories through a lower privacy budget.

[0121] It should be noted that this embodiment further determines the remaining two parameters as the fourth parameter, and jointly determines the target coefficient based on the third parameter and the fourth parameter. Since the maximum value of the first difference value, the first correlation value and the first reasoning difficulty value is 1, the calculated fifth intermediate coefficient must be less than 1. The fifth intermediate coefficient calculated by the fourth parameter is used as a correction coefficient for the fourth intermediate coefficient calculated based on the third parameter, and a lower-risk dimension is simultaneously introduced to modify the privacy budget to improve the reliability of the target coefficient.

[0122] In addition, in one embodiment, referring to Figure 3 After executing step S50, the method further includes but is not limited to the following steps:

[0123] S61, obtaining a first learning rate for calculating a first privacy budget, and obtaining a first attenuation coefficient preset in a privacy budget allocation model;

[0124] S62, determining the product of the target coefficient and the first attenuation coefficient as the second attenuation coefficient;

[0125] S63, when a second data request is obtained, determining a plurality of second property data from a plurality of optional property data, and constructing a second input data set based on the plurality of second property data;

[0126] S64, when the data category of the second property data is the same as the data category of the first property data, determining a second learning rate based on the first decay coefficient and the first learning rate;

[0127] S65, when at least one of the data categories of the second property data does not belong to the data category of the first property data, determining a second learning rate based on the second decay coefficient and the first learning rate;

[0128] S66, applying the second learning rate in the privacy budget allocation model to determine a third privacy budget for the second input data set, and publishing a second output data set based on the third privacy budget and the second input data set.

[0129] It should be noted that those skilled in the art know that it is necessary to introduce a learning rate when calculating the privacy budget, and to prevent oscillations caused by excessive step lengths by gradually reducing the learning rate. The privacy budget allocation model of this embodiment pre-sets the first learning rate and the corresponding first attenuation coefficient. After protecting the first data request based on the target coefficient, this embodiment further adjusts the first learning rate with the target coefficient, so that the subsequent calculation of the privacy budget can also consider the initially determined inference risk, thereby improving data security.

[0130] For example, a common learning rate expression is: ,in, is the current learning rate, is the preset initial learning rate, RATE is the first attenuation coefficient, and t is the number of attenuation times. In this embodiment, RATE is adjusted with the target coefficient. Since the target coefficient is a value greater than 1, RATE is larger after adjustment (that is, the first attenuation coefficient is smaller than the second attenuation coefficient), so that the second learning rate is smaller than the first learning rate before adjustment. The common budget expression is: ,in, is the t-th privacy budget, t is a natural number, is the preset descent gradient, is the preset loss function of the privacy budget model. When the learning rate The smaller the calculated The larger the value, the smaller the privacy budget can be.

[0131] It is worth noting that after obtaining the second data request, when the data category of the second property data is the same as the first property data, two consecutive data requests are initiated for the same category, and the data difference is small. If the data category of the second property data is different from the first property data, then the data of the two consecutive requests have certain data differences. The risk of using data differences for data inference is higher than the scenario in which two consecutive data requests are made for the same data category. Therefore, in the scenario corresponding to step S64, this embodiment normally decays the first learning rate to obtain the second learning rate, and in the scenario corresponding to step S65, the target coefficient is used to accelerate the decay of the first learning rate, so that the third privacy budget calculated based on the second learning rate in step S64 is less than the third privacy budget obtained in step S65, and the learning rate is adaptively decayed according to the data categories of multiple data requests to improve data security.

[0132] In addition, in one embodiment, the property management system is pre-associated with multiple resident terminals, and each resident terminal is uniquely associated with resident information.

[0133] Reference Figure 3 After step S63, the following steps are also included but not limited to:

[0134] S71, when the initiating terminal of the first data request and the second data request is the same resident terminal, determining a second learning rate based on the first attenuation coefficient and the first learning rate;

[0135] S72: When the initiating terminal of the first data request and the second data request is not a resident terminal, determine a second learning rate based on the second attenuation coefficient and the first learning rate.

[0136] It should be noted that the resident terminal can be bound in the property management system, and the specific process will not be described here. After obtaining the first data request and the second data request, it can be determined whether it is a resident terminal based on the terminal information carried in the data request, or whether it is the same terminal. If it belongs to the same resident terminal, it is likely that the operation was wrong when requesting data for the first time, and the possibility of the resident launching an inference attack is small. The second learning rate can be determined normally according to the first attenuation coefficient. If it does not belong to a resident terminal, there is a certain risk of inference attack. The second learning rate is determined with a larger second attenuation coefficient, and noise is added through a smaller third privacy budget to improve data security.

[0137] like Figure 4 As shown, Figure 4 : is a structural diagram of a privacy budget allocation device based on property data provided by an embodiment of the present invention. The present invention also provides a privacy budget allocation device based on property data, including:

[0138] The processor 401 may be implemented by a general-purpose central processing unit (CPU), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application;

[0139] The memory 402 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 402 can store an operating system and other applications. When the technical solution provided in the embodiment of this specification is implemented by software or firmware, the relevant program code is stored in the memory 402, and the processor 401 calls and executes the privacy budget allocation method based on property data in the embodiment of this application;

[0140] Input / output interface 403, used to implement information input and output;

[0141] Communication interface 404, used to realize communication interaction between the device and other devices, which can be realized through wired mode (such as USB, network cable, etc.) or wireless mode (such as mobile network, WIFI, Bluetooth, etc.);

[0142] A bus 405 that transmits information between various components of the device (e.g., processor 401, memory 402, input / output interface 403, and communication interface 404);

[0143] The processor 401 , the memory 402 , the input / output interface 403 and the communication interface 404 are connected to each other in communication within the device via the bus 405 .

[0144] An embodiment of the present application also provides an electronic device, including the property data-based privacy budget allocation device as described above.

[0145] An embodiment of the present application further provides a storage medium, which is a computer-readable storage medium and stores a computer program. When the computer program is executed by a processor, the above-mentioned privacy budget allocation method based on property data is implemented.

[0146] As a non-transient computer-readable storage medium, the memory can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage devices. In some embodiments, the memory may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned networks include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof. The device embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and are implemented to be located in one place, or may also be distributed to multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment.

[0147] It will be appreciated by those skilled in the art that all or some of the steps and systems in the methods disclosed above may be implemented as software, firmware, hardware, and appropriate combinations thereof. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include a computer storage medium (or non-transitory medium) and a communication medium (or transient medium). As known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disk (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and can be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically include computer readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0148] The above is a specific description of the preferred implementation of the present invention, but the present invention is not limited to the above-mentioned implementation mode. Technical personnel familiar with the field can also make various equivalent deformations or substitutions under the shared conditions without violating the spirit of the present invention. These equivalent deformations or substitutions are all included in the scope defined by the claims of the present invention.

Claims

1. A privacy budget allocation method based on property data, characterized in that: Applied to a property management system, the property management system is preset with multiple data categories and multiple household information, each of the data categories includes multiple optional property data, each of the optional property data is associated with a first difference value and a first reasoning difficulty value, the first difference value is used to indicate the degree of difference between the corresponding optional property data between different households, and the first reasoning difficulty value is used to indicate the difficulty of inferring the household information based on the corresponding optional property data, including: In response to a first data request, determining a plurality of first property data from a plurality of the selectable property data, and constructing a first input data set based on the plurality of the first property data; Based on each pair of the data categories to which the first property data belongs, a plurality of first correlation values ​​are obtained from a preset first mapping table, wherein the first mapping table records preset correlation values ​​between the data categories; Based on the first input data set, determining the first difference value greater than a preset first threshold as a second difference value, determining the first reasoning difficulty value greater than a preset second threshold as a second reasoning difficulty value, and determining the first correlation value greater than a preset third threshold as a second correlation value; Determining a target strategy from a plurality of preset strategies based on a first number of second difference values, a second number of the second reasoning difficulty values, and a third number of the second correlation values, wherein the preset strategy is used to adjust the privacy budget, and an adjustment range of the preset strategy is positively correlated with the first number, the second number, and the third number; Determine a first privacy budget for the first input data set based on a preset privacy budget allocation model, adjust the first privacy budget to a second privacy budget based on the target strategy, and publish a first output data set based on the second privacy budget and the first input data set; The preset strategy includes a first strategy, a second strategy, and a third strategy, the maximum adjustment range indicated by the first strategy is greater than that of the second strategy, the maximum adjustment range indicated by the second strategy is greater than that of the third strategy, the maximum value of the first difference value, the first reasoning difficulty value, and the first correlation value is 1, and a target strategy is determined from a plurality of preset strategies based on a first number of second difference values, a second number of second reasoning difficulty values, and a third number of second correlation values, including: Determine a first reference value according to a fourth quantity of the first property data and a preset reference ratio, and determine a second reference value according to a fifth quantity of the first related values ​​and the reference ratio, wherein the reference ratio is less than 1 and greater than 0.5; When at least one of the first number, the second number, and the third number is greater than a preset fourth threshold, determining the first strategy as the target strategy, wherein the fourth threshold is less than or equal to the fourth number; Alternatively, when the first number and the second number are greater than or equal to the first reference value, and the third number is greater than or equal to the second reference value, the first strategy is determined as the target strategy; Alternatively, when the first number and the second number are greater than or equal to the first reference value, and the third number is less than the second reference value, the second strategy is determined as the target strategy; Alternatively, when the third number is greater than or equal to the second reference value, and either the first number or the second number is greater than or equal to the first reference value, the second strategy is determined as the target strategy; Alternatively, when the first number and the second number are less than the first reference value, and the third number is greater than the second reference value, the third strategy is determined as the target strategy; Alternatively, when the third number is smaller than the second reference value, and only one of the first number and the second number is larger than the first reference value, the third strategy is determined as the target strategy.

2. The privacy budget allocation method based on property data according to claim 1 is characterized in that: When the target policy is the first policy, adjusting the first privacy budget to a second privacy budget based on the target policy includes: Based on the first input data set, determining the ratio of the maximum value to the average value of the first difference value as a first initial coefficient, determining the ratio of the maximum value to the average value of the first reasoning difficulty value as a second initial coefficient; determining the ratio of the maximum value to the average value of the first correlation value as a third initial coefficient; Based on the number of the second difference value and the first difference value being determined as a first ratio, the number ratio of the second reasoning difficulty value to the first reasoning difficulty value is determined as a second ratio, and the number ratio of the second correlation value to the first correlation value is determined as a third ratio; Determine the product of the first ratio and the first initial coefficient as a first intermediate coefficient, determine the product of the second ratio and the second initial coefficient as a second intermediate coefficient, and determine the product of the third ratio and the third initial coefficient as a third intermediate coefficient; determining a sum of the first intermediate coefficient, the second intermediate coefficient and the third intermediate coefficient as a target coefficient; The first privacy budget is reduced to the second privacy budget based on the target coefficient.

3. The privacy budget allocation method based on property data according to claim 1 is characterized in that: When the target policy is the second policy, adjusting the first privacy budget to the second privacy budget based on the target policy includes: When the second strategy is determined based on the first quantity and the second quantity, the first difference value is determined as a first parameter, and the first reasoning difficulty value is determined as a second parameter; When the second strategy is determined based on the third quantity, the first correlation value is determined as the first parameter, and the first reasoning difficulty value or the first difference value is determined as the second parameter, wherein when the second strategy is also determined based on the first quantity, the second parameter is the first difference value, and when the second strategy is also determined based on the second quantity, the second parameter is the first reasoning difficulty value; Based on a preset clustering algorithm, clustering a plurality of the first parameters into at least two first clustering clusters, and clustering a plurality of the second parameters into at least two second clustering clusters; Determine the cluster with the largest number of the first cluster and the second cluster as a target cluster, and determine an average value calculated based on the target cluster as a target coefficient; The first privacy budget is reduced to the second privacy budget based on the target coefficient.

4. The privacy budget allocation method based on property data according to claim 1 is characterized in that: When the target policy is the third policy, adjusting the first privacy budget to the second privacy budget based on the target policy includes: When the third number is greater than the second reference value, determining the first correlation value as a third parameter; Alternatively, when the first number is greater than the first reference value, the first difference value is determined as the third parameter; Alternatively, when the second number is greater than the first reference value, the first reasoning difficulty value is determined as the third parameter; determining the first reasoning difficulty value, the first difference value, or the first correlation value that is not determined as the third parameter as a fourth parameter; Determine a ratio of a maximum value to an average value of the third parameter as a fourth intermediate coefficient, determine a difference between a maximum value and a minimum value of the fourth parameter as a fifth intermediate coefficient, and determine a sum of the fourth intermediate coefficient and the fifth intermediate coefficient as a target coefficient; When the value of the fifth intermediate coefficient is greater than or equal to 1, determining the fourth intermediate coefficient as the target coefficient; The first privacy budget is reduced to the second privacy budget based on the target coefficient.

5. The privacy budget allocation method based on property data according to any one of claims 2 to 4, characterized in that: After publishing a first output data set based on the second privacy budget and the first input data set, the method further includes: Obtaining a first learning rate for calculating the first privacy budget, and obtaining a first attenuation coefficient preset in the privacy budget allocation model; determining a product of the target coefficient and the first attenuation coefficient as a second attenuation coefficient; When a second data request is obtained, a plurality of second property data are determined from the plurality of optional property data, and a second input data set is constructed based on the plurality of second property data; When the data category of the second property data is the same as the data category of the first property data, determining a second learning rate based on the first decay coefficient and the first learning rate; Alternatively, when at least one of the data categories of the second property data does not belong to the data category of the first property data, the second learning rate is determined based on the second decay coefficient and the first learning rate; The second learning rate is applied in the privacy budget allocation model to determine a third privacy budget of the second input dataset, and a second output dataset is published based on the third privacy budget and the second input dataset.

6. The privacy budget allocation method based on property data according to claim 5 is characterized in that: The property management system is pre-associated with a plurality of resident terminals, each of the resident terminals is uniquely associated with the resident information, and after constructing a second input data set based on the plurality of second property data, the method further includes: When the initiating terminal of the first data request and the initiating terminal of the second data are the same resident terminal, determining the second learning rate based on the first attenuation coefficient and the first learning rate; Alternatively, when the initiating terminal of the first data request and the second data does not belong to the resident terminal, the second learning rate is determined based on the second attenuation coefficient and the first learning rate.

7. A privacy budget allocation device based on property data, characterized in that: It includes at least one control processor and a memory for communicating with the at least one control processor; the memory stores instructions that can be executed by the at least one control processor, and the instructions are executed by the at least one control processor to enable the at least one control processor to execute the privacy budget allocation method based on property data as described in any one of claims 1 to 6.

8. An electronic device, characterized in that: Includes the privacy budget allocation device based on property data as described in claim 7.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the privacy budget allocation method based on property data as described in any one of claims 1 to 6.

Images