Two-party decision tree model privacy prediction method and system based on homomorphic encryption nesting

Abstract

The invention discloses a two-party decision tree model privacy prediction method and system based on homomorphic encryption nesting, and belongs to the technical field of homomorphic encryption. Performing path traversal on an encryption model preliminarily encrypted by a model provider through a data owner according to a prediction sample set, performing path disruption based on a comparison result of a preset threshold value and a feature value, generating an encryption path set, and transmitting the encryption path set to the model provider; the model provider performs path evaluation based on the encrypted path set, converts the encrypted path into a ciphertext state based on an evaluation result and sends the ciphertext state to the data owner; the data owner performs weight calculation according to the path evaluation result of the ciphertext path, generates a ciphertext prediction result in combination with a prediction result confidentiality demand, and sends the ciphertext prediction result to the model provider, so that privacy protection of the model and the data is ensured, and high efficiency is maintained at the same time; the problem of high privacy protection communication and calculation overhead in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the field of homomorphic encryption technology, and in particular to a privacy prediction method and system based on a two-party decision tree model. Background Art

[0002] The statements in this section merely mention background art related to the present invention and do not necessarily constitute prior art.

[0003] With the popularity of machine learning inference services, decision tree models play an important role in many fields because they can make decisions in a structured and explainable way. Decision trees divide data into multiple branches according to specific conditions to form a flowchart-like structure, guiding decisions from the root node to the leaf node. In decision tree inference services, the provider has a trained machine learning model, and the customer submits data to obtain prediction results, which involves both the privacy of the model itself and the privacy of the input data.

[0004] However, these decision tree inference services also bring the risk of privacy leakage. Once the model information is leaked, it will cause significant losses. At the same time, the data involved in decision tree inference services often contains sensitive personal information. Data owners are usually interested in obtaining service results, and at the same time hope to ensure that model providers cannot obtain their data.

[0005] In a typical service-centric privacy-preserving decision tree framework, there are two main computing entities, including the model provider with a trained model and the data owner. Traditional privacy-preserving methods have an obvious trade-off between privacy protection and performance, especially in large-scale distributed environments, where the computing and communication overheads are too high, resulting in limited real-time reasoning in practical applications.

[0006] In addition, the traditional service-centric model limits the flexibility of decision tree reasoning. When deploying decision tree services between different companies, the goals of each party are usually different. For example, for the model provider, it hopes to continuously obtain data through reasoning results to support decision-making or optimize its model performance; at the same time, the amount of data information of the data owner grows steadily over time.

[0007] With the powerful computing and storage capabilities of cloud computing, cloud-based outsourcing frameworks have become the main design choice. A major advantage of introducing a third party is that it simplifies the design of privacy protection protocols. For example, if the third party does not understand the decision tree model, the data can be used more effectively to prevent the third party from obtaining private information through the data.

[0008] However, reliance on a trusted (at least semi-honest) third party introduces important security assumptions. While using a trusted third party simplifies certain aspects of the framework, it also introduces a number of problems. Designs often come with significant performance tradeoffs to protect against attacks by malicious adversaries, and these protocols are also vulnerable to collision attacks. Furthermore, reliance on a third party may hinder real-time collaboration between model providers and data owners.

[0009] In order to provide services, they first need to identify a trusted third party that can perform the necessary computing and communication tasks and install the required software. Hiring a third party usually incurs additional financial costs. In addition, frameworks that rely on third-party services have potential risks in privacy protection and cannot meet data privacy requirements in certain specific application scenarios. Summary of the invention

[0010] In order to address the deficiencies in the prior art, the present invention provides a privacy prediction method, system, electronic device, computer-readable storage medium and computer program product for a two-party decision tree model based on homomorphic encryption nesting, which allows direct interaction between the model provider and the data owner, and utilizes homomorphic encryption and model obfuscation technology to ensure the privacy protection of the model and data while maintaining high efficiency.

[0011] In a first aspect, the present invention provides a privacy prediction method for a two-party decision tree model based on homomorphic encryption nesting; A privacy prediction method for a two-party decision tree model based on homomorphic encryption nesting, comprising: The model provider performs obfuscation and homomorphic encryption on the decision tree model, obtains the encrypted model and transmits it to the data owner; The data owner traverses the path of the encrypted model according to the prediction sample set and scrambles the path based on the comparison result of the preset threshold and the characteristic value, generates an encrypted path set and transmits it to the model provider; The model provider performs path evaluation based on the encrypted path set, converts the encrypted path into ciphertext based on the evaluation result, and sends it to the data owner; The data owner calculates the weight based on the path evaluation result of the ciphertext path, generates the ciphertext prediction result based on the confidentiality requirement of the prediction result, and sends it to the model provider. The model provider decrypts the ciphertext prediction result to obtain the real prediction result.

[0012] In some embodiments, the performing of the obfuscation operation on the decision tree model specifically includes: randomly rotating the nodes in the decision tree model and inserting virtual nodes and virtual trees into the decision tree model.

[0013] In some embodiments, the decision tree model performs homomorphic encryption specifically by using different homomorphic encryption schemes to encrypt the thresholds and leaf node weights in the decision tree model.

[0014] In some implementations, the encrypted model is traversed according to the prediction sample set and the path is disrupted based on the comparison result of the preset threshold and the characteristic value to generate the encrypted path set specifically as follows: Based on the prediction samples in the prediction sample set, the encryption model is traversed from the root node to the leaf node. During the traversal, the characteristic value corresponding to each intermediate node is compared with a preset threshold, and the intermediate node is replaced based on the comparison result.

[0015] In some implementations, performing path evaluation based on the encrypted path set and converting the encrypted path into a ciphertext state based on the evaluation result is specifically: Each encryption path in the encryption path set is evaluated, and if each node in the encryption path is correct, the node is converted to a first ciphertext state; otherwise, the node is converted to a second ciphertext state.

[0016] In some implementations, the weight calculation is performed based on the path evaluation result of the ciphertext path, and the ciphertext prediction result is generated in combination with the confidentiality requirement of the prediction result, specifically: The weight calculation is performed based on the path evaluation result and the ciphertext state corresponding to the ciphertext path to obtain the confidentiality requirement of the prediction result of the data provider. Based on the confidentiality requirement of the prediction result, it is determined whether to mask the calculation result to generate the ciphertext prediction result.

[0017] In a second aspect, the present invention provides a privacy prediction system for a two-party decision tree model based on homomorphic encryption nesting; A two-party decision tree model privacy prediction system based on homomorphic encryption nesting, including a model provider and a data owner; The model provider performs obfuscation operations and homomorphic encryption on the decision tree model, obtains the encrypted model and transmits it to the data owner; The data owner traverses the path of the encrypted model according to the prediction sample set and scrambles the path based on the comparison result of the preset threshold and the characteristic value, generates an encrypted path set and transmits it to the model provider; The model provider performs path evaluation based on the encrypted path set, converts the encrypted path into a ciphertext state based on the evaluation result, and sends the ciphertext state to the data owner; The data owner performs weight calculation based on the path evaluation result of the ciphertext path, generates a ciphertext prediction result based on the confidentiality requirement of the prediction result, and sends it to the model provider. The model provider decrypts the ciphertext prediction result to obtain the true prediction result.

[0018] In some implementations, the encrypted model is traversed according to the prediction sample set and the path is disrupted based on the comparison result of the preset threshold and the characteristic value to generate the encrypted path set specifically as follows: Based on the prediction samples in the prediction sample set, the encryption model is traversed from the root node to the leaf node. During the traversal, the characteristic value corresponding to each intermediate node is compared with a preset threshold, and the intermediate node is replaced based on the comparison result.

[0019] In some implementations, performing path evaluation based on the encrypted path set and converting the encrypted path into a ciphertext state based on the evaluation result is specifically: Each encryption path in the encryption path set is evaluated, and if each node in the encryption path is correct, the node is converted to a first ciphertext state; otherwise, the node is converted to a second ciphertext state.

[0020] In some implementations, the weight calculation is performed based on the path evaluation result of the ciphertext path, and the ciphertext prediction result is generated in combination with the confidentiality requirement of the prediction result, specifically: The weight calculation is performed based on the path evaluation result and the ciphertext state corresponding to the ciphertext path to obtain the confidentiality requirement of the prediction result of the data provider. Based on the confidentiality requirement of the prediction result, it is determined whether to mask the calculation result to generate the ciphertext prediction result.

[0021] Compared with the prior art, the present invention has the following beneficial effects: 1. The technical solution provided by the present invention, through innovative data encryption and distributed collaboration technology, can ensure efficient reasoning performance while protecting data privacy, while removing dependence on third-party trust and adapting to more practical application scenarios.

[0022] 2. The technical solution provided by the present invention implements a simplified interaction strategy, allows direct interaction between the model provider and the data owner, performs preliminary encryption through the data provision method, effectively shares the computational burden, and is suitable for real-time application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0024] Figure 1 A schematic diagram of a privacy prediction method for a two-party decision tree model based on homomorphic encryption nesting provided in an embodiment of the present invention; Figure 2 This is an example diagram of a right-handed node provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0025] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used in the present invention have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0026] It should be noted that the terms used herein are only for describing specific embodiments, and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that the terms "include" and "have" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0027] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0028] Embodiment 1 The computational and communication overheads of the privacy protection methods in existing decision tree reasoning are too large, which affects the real-time performance of their actual application in prediction. Therefore, the present invention provides a two-party decision tree model privacy prediction method based on homomorphic encryption nesting, which utilizes the direct interaction between the model provider and the data owner to ensure the privacy security of the data and the model through homomorphic encryption and model obfuscation technology.

[0029] Next, combine Figure 1-Figure 2 , a privacy prediction method for a two-party decision tree model based on homomorphic encryption nesting disclosed in this embodiment is described in detail. The privacy prediction method for a two-party decision tree model based on homomorphic encryption nesting includes the following steps: S1. The model provider and the data owner confirm their willingness to cooperate and negotiate global security parameters.

[0030] Among them, global security parameters include homomorphic encryption algorithm combinations, key lengths, etc., which are used for subsequent homomorphic encryption of decision tree models; negotiating global security parameters means that the model provider and the data owner determine the homomorphic encryption algorithm combinations, key lengths and other parameters for subsequent applications.

[0031] S2. The model provider performs obfuscation and homomorphic encryption on the decision tree model, obtains the encrypted model and transmits it to the data owner.

[0032] As an implementation mode, S2 specifically includes: S201. The model provider randomly rotates the nodes in the decision tree model through a rotation algorithm, and inserts virtual nodes and virtual trees into the randomly rotated decision tree model.

[0033] For example, represents the intermediate node, Represents its left child node, represents its right child node, , It is also an intermediate node; the specific process of performing right rotation operation on the decision tree model through the rotation algorithm is described in detail: (1) Disconnect and connection, Indicates that at this time The subtree with root express The left subtree of express The right subtree of The two child nodes of , namely its corresponding left subtree and right subtree, are replaced by .

[0034] (2) The left child nodes of the neutron nodes are replaced by and .

[0035] Similarly, the specific process of performing a left rotation operation on the decision tree model through the rotation algorithm is as follows: (a) Disconnect and connection, Indicates that at this time The subtree with root express The left subtree of express The right subtree of The two child nodes of , namely its corresponding left subtree and right subtree, are replaced by .

[0036] (b) The right child nodes of the neutron nodes are replaced by and .

[0037] Based on this, by randomly rotating the nodes in the decision tree, the tree structure of the decision tree model is confused and the order of feature comparison is disrupted while the decision logic behavior of the decision tree model remains unchanged, which helps to improve the privacy and security of the prediction.

[0038] Subsequently, virtual nodes are inserted into the randomly rotated decision tree model; the steps for inserting virtual nodes are as follows: Step 1: Find two thresholds in the sample space and ,satisfy , here, Represents any eigenvalue that may appear in the sample space; It can be selected as the minimum value in the sample space; It can be selected as the maximum value in the sample space.

[0039] Here, the sample space refers to the data sample space supported by the decision tree model.

[0040] Step 2: Randomly select a node in the tree and let the subtree with the node as the root be , and replace it with a dummy node , the virtual node has a random feature index and threshold or .

[0041] Step 3: If the virtual node The threshold value is , then set its right child node to , the left child node is set to a randomly generated subtree; if the virtual node The threshold value is , then perform the opposite operation.

[0042] After that, insert a virtual tree into the decision tree model where the virtual node is inserted and disrupt the order of the tree. The steps for inserting the virtual tree are as follows: Step (1) Assume that you want to insert A virtual tree, building an array ,satisfy .

[0043] Step (2) Randomly generate trees, each tree contains only one correct path, and the The weight of each tree is set to , other parameters are set randomly.

[0044] Here, by setting the threshold to or , each tree can contain only one correct path by setting the threshold of all edges extending to the left on only one path to , the threshold of the edge extending to the right is , only the path that meets this condition is the correct path, and other paths have at least one edge that violates this principle.

[0045] Step (3) inserts the generated virtual tree into the original tree.

[0046] S202. The model provider performs homomorphic encryption on the decision tree model inserted into the virtual tree through two sets of homomorphic encryption schemes.

[0047] Specifically, first, the model provider initializes two sets of homomorphic encryption schemes, denoted as and , here, represents the encryption algorithm corresponding to the first set of homomorphic encryption schemes, represents the decryption algorithm corresponding to the first homomorphic encryption scheme, represents the encryption algorithm corresponding to the second set of homomorphic encryption schemes, represents the decryption algorithm corresponding to the second homomorphic compression scheme, represents the public key corresponding to the second set of homomorphic encryption scheme, Represents the private key corresponding to the second homomorphic encryption scheme.

[0048] Here, due to The encryption key is not shared with any other party, so its encryption key is omitted here.

[0049] In this embodiment, the encryption and decryption algorithm used in the first homomorphic encryption scheme may be Paillier, and the homomorphic encryption and decryption algorithm used in the second homomorphic encryption scheme may be RLWE.

[0050] Next, the model provider passes and The thresholds and leaf node weights in the decision tree model are encrypted sequentially.

[0051] Finally, the model provider publishes the encrypted tree (denoted as ) and encryption parameters and The encrypted model is then transmitted to the data owner. This step implements the following transformations:

[0052] ; in, represents the threshold of the jth node in the i-th tree after encryption, represents the threshold of the jth node in the i-th tree, represents the weight of the jth node in the encrypted i-th tree, represents the weight of the jth node in the i-th tree, , Respectively represent the threshold and weight of node storage.

[0053] S3. The data owner traverses the encrypted model based on the prediction sample set and scrambles the path based on the comparison result of the preset threshold and the characteristic value, generates an encrypted path set and transmits it to the model provider. Specifically, it includes: S301: The data owner selects samples that need to be predicted from the sample set owned by the data owner to obtain a prediction sample set.

[0054] Exemplarily, if the power grid is the model provider and the power plant is the data owner, the sample set is data samples including characteristics such as wind power, water power, and time period. The purpose of the data owner is to predict the power generation at a specific time point through the model of the model provider.

[0055] S302: For each prediction sample in the prediction sample set, the data owner inputs the prediction sample into the encryption model, traverses all possible paths from the root node to the leaf node, and obtains a set of paths. ; During the traversal process, the data owner replaces the intermediate nodes with the comparison results of the feature value corresponding to each intermediate node and the threshold.

[0056] Specifically, for the prediction sample , starting from the root node, the path traversal, the data provider calculates the difference between the corresponding threshold of the intermediate node and the corresponding feature value, expressed as: ; in, represents a random positive value, Represents the prediction sample In Features The eigenvalue at Representation Node Storage characteristics, Representation Node The storage threshold.

[0057] Obviously, The positive and negative signs of the decryption results indicate whether the feature value is greater than or less than the threshold. For each intermediate node in the path , if the node is the left child of the parent node, the data provider uses Replace it; otherwise, replace Replace it, i.e. .

[0058] By adjusting the order of subtraction, the data owner ensures that all comparisons along the correct path are non-positive. After removing its root node, each path is represented as: ; In the formula, It indicates the weight stored in the leaf node corresponding to the path, and d indicates the path depth.

[0059] S4. The model provider performs path evaluation based on the encrypted path set, converts the encrypted path into ciphertext based on the evaluation result, and sends it to the data owner.

[0060] Specifically, after receiving the shuffled path set After that, the model provider evaluates each path ; For each path, if all are all non-positive values, which means that the corresponding comparison result is "branch to the left child node or the right child node", which is correct, then replace it with ; otherwise, replace it with .

[0061] Thanks to the characteristics of the Paillier encryption system, the encryption of each value is randomized. Sent to the data owner.

[0062] S5. The data owner performs weight calculation based on the path evaluation result of the ciphertext path, generates a ciphertext prediction result based on the confidentiality requirements of the prediction result, and sends it to the model provider.

[0063] Specifically, upon receiving After that, the data owner has Perform multiplication operation, expressed as: ; In this multiplication operation, is considered as a scalar, and Indicates the path after replacement , the intermediate nodes in the path have been replaced by or , that is, this operation can be regarded as a ciphertext-scalar multiplication performed under the first set of homomorphic encryption schemes.

[0064] Then, the data owner sums all the results and obtains: .

[0065] Furthermore, according to the characteristics and design of the decision tree model, only The correct path in is replaced by , and the error path is replaced by Therefore, the calculation result is expressed as: ; in, represents the weight in the error path, represents the weight in the correct path, Represents the expected prediction result.

[0066] Furthermore, if the task requires the prediction results to be made public to the model provider, Sent to the model provider; otherwise, if the prediction result should be kept confidential to the data owner, the data owner randomly selects a mask , and calculate ,because is known to the data owner. Then, the data owner uses the following method to join in: ; in, Express The ciphertext-scalar addition is performed, and the noise generation comes from the linear transformation.

[0067] S6. The model provider decrypts the ciphertext prediction result to obtain the real prediction result.

[0068] Specifically, upon receiving Afterwards, the model provider calculates the prediction results in plain text format: .

[0069] Furthermore, if the prediction results need to be made public to the data owner, this means has been masked, the model provider will Sent to the data owner, who will remove the previously added mask and get the true prediction result.

[0070] Embodiment 2 Based on the privacy prediction method of the two-party decision tree model based on homomorphic encryption nesting described in Example 1, this embodiment discloses a privacy prediction system of the two-party decision tree model based on homomorphic encryption nesting, including a model provider and a data owner; The model provider performs obfuscation operations and homomorphic encryption on the decision tree model, obtains the encrypted model and transmits it to the data owner; The data owner traverses the path of the encrypted model according to the prediction sample set and scrambles the path based on the comparison result of the preset threshold and the characteristic value, generates an encrypted path set and transmits it to the model provider; The model provider performs path evaluation based on the encrypted path set, converts the encrypted path into a ciphertext state based on the evaluation result, and sends the ciphertext state to the data owner; The data owner performs weight calculation based on the path evaluation result of the ciphertext path, generates a ciphertext prediction result based on the confidentiality requirement of the prediction result, and sends it to the model provider. The model provider decrypts the ciphertext prediction result to obtain the true prediction result.

[0071] The description of each embodiment in the above embodiments has different emphases. For parts not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0072] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A privacy prediction method based on a two-party decision tree model nested with homomorphic encryption, characterized in that: include: The model provider performs obfuscation and homomorphic encryption on the decision tree model, obtains the encrypted model and transmits it to the data owner; The data owner traverses the path of the encrypted model according to the prediction sample set and scrambles the path based on the comparison result of the preset threshold and the characteristic value, generates an encrypted path set and transmits it to the model provider; The model provider performs path evaluation based on the encrypted path set, converts the encrypted path into ciphertext based on the evaluation result, and sends it to the data owner; The data owner calculates the weight based on the path evaluation result of the ciphertext path, generates the ciphertext prediction result based on the confidentiality requirement of the prediction result, and sends it to the model provider. The model provider decrypts the ciphertext prediction result to obtain the real prediction result.

2. The privacy prediction method of the two-party decision tree model based on homomorphic encryption nesting as claimed in claim 1 is characterized in that: The performing of the obfuscation operation on the decision tree model specifically includes: randomly rotating the nodes in the decision tree model and inserting virtual nodes and virtual trees into the decision tree model.

3. The privacy prediction method of the two-party decision tree model based on homomorphic encryption nesting as claimed in claim 1 is characterized in that: The specific implementation of homomorphic encryption on the decision tree model is as follows: different homomorphic encryption schemes are used to encrypt the thresholds and leaf node weights in the decision tree model.

4. The privacy prediction method of the two-party decision tree model based on homomorphic encryption nesting as claimed in claim 1 is characterized in that: The method of performing path traversal on the encrypted model according to the prediction sample set and disrupting the path based on the comparison result between the preset threshold and the characteristic value to generate the encrypted path set is specifically as follows: Based on the prediction samples in the prediction sample set, the encryption model is traversed from the root node to the leaf node. During the traversal, the characteristic value corresponding to each intermediate node is compared with a preset threshold, and the intermediate node is replaced based on the comparison result.

5. The privacy prediction method of the two-party decision tree model based on homomorphic encryption nesting as claimed in claim 1 is characterized in that: The performing of path evaluation based on the encrypted path set and converting the encrypted path into a ciphertext state based on the evaluation result is specifically as follows: Each encryption path in the encryption path set is evaluated, and if each node in the encryption path is correct, the node is converted to a first ciphertext state; otherwise, the node is converted to a second ciphertext state.

6. The privacy prediction method of the two-party decision tree model based on homomorphic encryption nesting as claimed in claim 1, characterized in that: The weight calculation is performed based on the path evaluation result of the ciphertext path, and the ciphertext prediction result is generated in combination with the confidentiality requirement of the prediction result, specifically: The weight calculation is performed based on the path evaluation result and the ciphertext state corresponding to the ciphertext path to obtain the confidentiality requirement of the prediction result of the data provider. Based on the confidentiality requirement of the prediction result, it is determined whether to mask the calculation result to generate the ciphertext prediction result.

7. A privacy prediction system based on a two-party decision tree model nested with homomorphic encryption, characterized in that: Including model providers and data owners; The model provider performs obfuscation operations and homomorphic encryption on the decision tree model, obtains the encrypted model and transmits it to the data owner; The data owner traverses the path of the encrypted model according to the prediction sample set and scrambles the path based on the comparison result of the preset threshold and the characteristic value, generates an encrypted path set and transmits it to the model provider; The model provider performs path evaluation based on the encrypted path set, converts the encrypted path into a ciphertext state based on the evaluation result, and sends the ciphertext state to the data owner; The data owner performs weight calculation based on the path evaluation result of the ciphertext path, generates a ciphertext prediction result based on the confidentiality requirement of the prediction result, and sends it to the model provider. The model provider decrypts the ciphertext prediction result to obtain the true prediction result.

8. The privacy prediction system of the two-party decision tree model based on homomorphic encryption nesting as claimed in claim 7, characterized in that: The method of performing path traversal on the encrypted model according to the prediction sample set and disrupting the path based on the comparison result between the preset threshold and the characteristic value to generate the encrypted path set is specifically as follows: Based on the prediction samples in the prediction sample set, the encryption model is traversed from the root node to the leaf node. During the traversal, the characteristic value corresponding to each intermediate node is compared with a preset threshold, and the intermediate node is replaced based on the comparison result.

9. The privacy prediction system of the two-party decision tree model based on homomorphic encryption nesting as claimed in claim 7, characterized in that: The performing of path evaluation based on the encrypted path set and converting the encrypted path into a ciphertext state based on the evaluation result is specifically as follows: Each encryption path in the encryption path set is evaluated, and if each node in the encryption path is correct, the node is converted to a first ciphertext state; otherwise, the node is converted to a second ciphertext state.

10. The privacy prediction system of the two-party decision tree model based on homomorphic encryption nesting according to claim 7, characterized in that: The weight calculation is performed based on the path evaluation result of the ciphertext path, and the ciphertext prediction result is generated in combination with the confidentiality requirement of the prediction result, specifically: The weight calculation is performed based on the path evaluation result and the ciphertext state corresponding to the ciphertext path to obtain the confidentiality requirement of the prediction result of the data provider. Based on the confidentiality requirement of the prediction result, it is determined whether to mask the calculation result to generate the ciphertext prediction result.

Images