A method for dual privacy protection in social networks with improved utility

Abstract

The present invention relates to the field of privacy information protection technology, and specifically to a method for dual privacy protection of social networks with improved utility. The method combines external link privacy with an uncertain graph. The external link privacy is used to protect the out-degree of a node. The uncertain graph algorithm post-processing technology based on the external link privacy injects uncertainty into the in-degree edge of the node to achieve privacy protection of the in-degree edge, and improves data utility while meeting the differential privacy protection requirements. The uncertain graph is used as a post-processing of differential privacy, which does not consume the privacy budget, so that the overall algorithm protects the in-degree of data privacy while meeting the external link privacy, thereby achieving dual privacy protection.

Description

Technical Field

[0001] The present invention relates to the technical field of privacy information protection, and in particular to a method for dual privacy protection of a social network with improved effectiveness. Background Art

[0002] The rapid popularization of social networking platforms has made social networks an important source of data for sociological and business model research. At the same time, the issue of data privacy disclosure has brought a lot of trouble to individuals participating in social activities. If data privacy is not effectively protected, it is easy to leak. With the enhancement of privacy awareness, people pay more and more attention to the protection of data privacy.

[0003] In order to accurately describe the relationship between entities in a social network, the social network can be abstracted into a social graph structure, where each entity is represented by a node, the relationship between entities is represented by one or more types of relationships, and the degree of association between entities can be represented by weighting the edges. Therefore, the research on privacy protection of social networks is transformed into the research on privacy protection of graph structures.

[0004] The most advanced methods for privacy protection in social networks can be divided into five categories (and their variants): (1) Identifier replacement, which replaces the user's unique identifier attribute with a synthetic identifier. (2) Graph modification technology, which achieves anonymity in social graphs by modifying some nodes or edges in the graph. (3) Differential privacy, a strict privacy standard that provides strong privacy guarantees based on strictly provable mathematical definitions and quantifiable evaluations, regardless of the background knowledge possessed by the attacker. (4) Graph clustering methods, which aggregate multiple nodes or edges into a set of supervertices and superedges to protect individual privacy. (5) Probabilistic graph methods, which randomly inject uncertainty into the original social graph. According to the different protection and effects against specific types of attacks, the above methods can be divided into random perturbation methods and anonymization methods of the graph.

[0005] Differential privacy has been widely studied and applied due to the following characteristics:

[0006] 1. It does not need to consider any possible background knowledge possessed by the attacker and is highly robust to arbitrary background knowledge;

[0007] 2. Based on a solid mathematical foundation, it provides more flexible theoretical analysis;

[0008] 3. Provides a strict definition of privacy protection and a quantitative evaluation method.

[0009] As a variant of differential privacy applicable to social graphs, external link privacy protection provides quantifiable privacy guarantees similar to differential privacy, while being more flexible in protecting structural information in social graphs.

[0010] However, external link privacy only protects the out-degree of nodes in directed social graphs, but the node's information will still be leaked through its in-degree (the out-degree of its friends). Therefore, how to perform dual protection (both out-degree and in-degree) of data privacy in social networks has become an urgent problem that needs to be solved. Summary of the invention

[0011] The present invention aims to provide a method for dual privacy protection of social networks with improved utility, so as to solve the problem in the prior art that node data privacy is easily leaked through in-degree.

[0012] A method for dual privacy protection of social networks with improved utility is to combine external link privacy with uncertain graph algorithm to perform dual privacy protection on the out-degree and in-degree of directed social graphs.

[0013] Note: The out-degree mentioned in the present invention refers to the information output of the node itself to the outside; the in-degree refers to the out-degree information output of other nodes related to the node itself to the outside.

[0014] In the present invention, the out-degree of the edge in the directed social graph is protected by external link privacy, and the in-degree of the node is protected by the uncertain graph algorithm, so that the edges in the directed social graph are double-privacy protected, thereby preventing the node information from being leaked through its out-degree and further preventing the node information from being leaked through its in-degree.

[0015] The present invention has a method for improving the effectiveness of dual privacy protection in social networks, further comprising the following steps:

[0016] Step 1, directed social graph as input;

[0017] Step 2: Use external link privacy to protect the out-degree in the directed social graph and obtain a noisy graph;

[0018] Step 3: Use the uncertain graph algorithm to protect the in-degree in the noise graph to obtain an uncertain directed graph.

[0019] In the present invention, the out-degree in the directed social graph is protected by using the external link privacy, and the uncertain graph algorithm is used as the post-processing of the external link privacy, that is, the uncertain graph algorithm is used to protect the in-degree in the noise graph, thereby protecting the in-degree of the node, to achieve double protection of data privacy in the social network.

[0020] The present invention has a method for improving the effectiveness of dual privacy protection in social networks, and steps 2 and 3 specifically include the following contents:

[0021] Step 201: Based on the set privacy budget ò and the sensitivity Δf of the query function f out-link, calculate the parameter λ of the Laplace distribution = Δf out-link / ò, use the Laplace mechanism to protect the out-degree of the node and obtain the noise graph G';

[0022] Step 301: The uncertain graph algorithm is used as a post-processing of the external link privacy to protect the in-degree in the noise graph G'. Specifically, the noise value y generated by the Laplace mechanism is used to calculate the corresponding probability The probabilities are then added to the edges of the noise graph G' to obtain an uncertain directed graph.

[0023] In the present invention, external link privacy is used as a variant of differential privacy (edge ​​differential privacy and node differential privacy) to meet the requirements of differential privacy protection. The uncertain graph algorithm post-processing technology based on external link privacy injects uncertainty into the in-degree edges of the node to achieve privacy protection of the in-degree edges.

[0024] Furthermore, the present invention has a method for improving the effectiveness of dual privacy protection in social networks, and step 301 specifically includes the following contents:

[0025] Input noise graph G', privacy budget ò and sensitivity Δf of query function f out-link ,

[0026] According to the number of edges |E| in the directed social graph G(V,E), |E| noise values ​​Y=(y1,y2,···,y |E| ),

[0027] Noise optimization, let x i =|y i |, and then convert the optimized noise value into a probability value p i ←F(x i ),

[0028] The probability value p i Add edge e to the noise graph G' i , we get an uncertain directed graph

[0029] Uncertain graph is a new privacy protection method. Its main principle is to inject uncertainty into the edges of social network graphs and publish obfuscated uncertain graphs to achieve the purpose of privacy protection. This method achieves privacy protection by assigning probability values ​​to the edges in the graph, while making little change to the original graph data. Compared with completely removing or adding edges, the utility of the original data is maintained to a certain extent.

[0030] However, the existing uncertain graph algorithms are for undirected graphs. Although the utility of the original data is maintained to a certain extent, the utility of the perturbed data is still relatively low. In the present invention, the uncertain graph algorithm technology is applied to directed social graphs, and noise optimization technology is used to improve the utility of the data.

[0031] Furthermore, the present invention has a method for improving the effectiveness of dual privacy protection in social networks. For any external link neighborhood pair (G1, G2), the global sensitivity of the query function f is expressed as:

[0032]

[0033] The outer link neighborhood is defined as: for two given directed graphs G1(V1,E1) and G2(V2,E2), if V1=V2, Then G1 and G2 are said to be adjacent in external links.

[0034] In the present invention, by modifying the definition of the neighborhood data set, both the privacy protection effect and the data utility of the output data set can meet actual needs.

[0035] Furthermore, the present invention has a method for improving the effectiveness of dual privacy protection in social networks, and the external link privacy is:

[0036] Suppose there is a random algorithm M, Range(M) is the set of all possible outputs of M, for any two external link neighborhood graphs G1 and G2 and any subset S of Range(M), if the algorithm M satisfies:

[0037] Pr(M(G1)∈S)≤e ò ×Pr(M(G2)∈S),

[0038] The random algorithm M is said to provide ò-external link privacy protection, where the parameter ò is called the privacy protection budget.

[0039] Furthermore, the present invention has a method for improving the effectiveness of dual privacy protection in social networks, and the uncertainty graph is defined as:

[0040] Given an undirected graph G(V,E), if the mapping P:V P →[0,1] is the probability function of the existence of each edge in the edge set, then G'(V,P) is an uncertain graph about the undirected graph G(V,E), where V p Represents all possible vertex pairs in the vertex set V. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 A flowchart of an embodiment of a method for dual privacy protection in a social network with improved effectiveness according to the present invention;

[0042] Figure 2 The degree distribution of the noise-optimized PEP algorithm in an embodiment of a method for dual privacy protection in a social network with improved utility of the present invention;

[0043] Figure 3 This is the degree distribution of the PEP algorithm without noise optimization in an embodiment of a method for dual privacy protection in a social network with improved utility of the present invention. DETAILED DESCRIPTION

[0044] The following is further described in detail through specific implementation methods:

[0045] In an embodiment of the present invention, in order to accurately describe the relationship between entities in a social network, the social network is abstracted into a social graph structure, each entity is represented by a node, and the relationship between entities is represented by one or more types of relationships.

[0046] We define privacy leakage as an attacker being able to determine with high probability whether an individual has participated in the generation of a social network. The direction of an edge indicates the direction of information flow or data contribution. A→B means that node A can contribute information about node B, while B→A means that node B can contribute information about node A. Therefore, in order to protect the evidence of individual participation, both the out-degree and in-degree of nodes must be protected. We combine external link privacy with an uncertain graph algorithm to protect the out-degree and in-degree of directed social graphs. This method has a dual privacy guarantee and uses noise optimization to improve the utility of output data in the process of generating uncertain graphs.

[0047] Uncertain graphs have become a new privacy protection method. Its main principle is to inject uncertainty into the edges of social network graphs and publish obfuscated uncertain graphs to achieve the purpose of privacy protection. This method achieves privacy protection by assigning probability values ​​to directed edges in the graph, while making little change to the original graph data. Compared with completely removing or adding edges, the utility of the original data is maintained to a certain extent.

[0048] This embodiment has a method for improving the effectiveness of dual privacy protection in social networks, and first defines the following:

[0049] The outer link neighborhood is defined as: for two given directed graphs G1(V1,E1) and G2(V2,E2), if V1=V2, Then G1 and G2 are said to be adjacent in external links.

[0050] Sensitivity is defined as: For any outlink neighborhood pair (G1, G2), the global sensitivity of the query function f is expressed as:

[0051]

[0052] External link privacy is defined as:

[0053] Suppose there is a random algorithm M, Range(M) is the set of all possible outputs of M, for any two external link neighborhood graphs G1 and G2 and any subset S of Range(M), if the algorithm M satisfies:

[0054] Pr(M(G1)∈S)≤e ò ×Pr(M(G2)∈S),

[0055] The random algorithm M is said to provide ò-external link privacy protection, where the parameter ò is called the privacy protection budget.

[0056] The Laplace mechanism implements ò-differential privacy protection by adding random noise that follows the Laplace distribution to the exact query results. Let the Laplace distribution with mean 0 and scale parameter λ be Lap(λ), then its probability density function is

[0057]

[0058] Definition of Laplace mechanism:

[0059] Given a dataset D, query function f:D→R n , its sensitivity is Δf, then the random algorithm M(G)=f(G)+<lap(Δf / ?)> n Provides ò-differential privacy protection.

[0060] The uncertainty graph is defined as:

[0061] Given an undirected graph G(V,E), if the mapping P:V P →[0,1] is the probability function of the existence of each edge in the edge set, then G'(V,P) is an uncertain graph about the undirected graph G(V,E), where V p Represents all possible vertex pairs in the vertex set V.

[0062] As attached Figure 1 As shown, this embodiment has a method for improving the effectiveness of dual privacy protection in social networks, and adopts the PEP algorithm, which is a personal evidence protection algorithm.

[0063] Here are the steps:

[0064] Step 1, directed social graph as input;

[0065] Step 2: Use external link privacy to protect the out-degree in the directed social graph and obtain a noisy graph;

[0066] Step 3: Use the uncertain graph algorithm to protect the in-degree in the noise graph to obtain an uncertain directed graph.

[0067] In this embodiment, the uncertain graph algorithm is also called a perturbation algorithm.

[0068] Among them, step 2 and step 3 specifically include the following contents:

[0069] Step 201: Based on the set privacy budget ò and the sensitivity Δf of the query function f out-link , calculate the parameter λ of the Laplace distribution = Δf out-link / ò, use the Laplace mechanism to protect the out-degree of the node and obtain the noise graph G';

[0070] Step 301: The uncertain graph algorithm is used as a post-processing of the external link privacy to protect the in-degree in the noise graph G'. Specifically, the Laplace mechanism is used to generate the noise value y and calculate its corresponding probability. Then the probabilities are assigned to the edges of the noise graph G' to obtain an uncertain directed graph.

[0071] More specifically, step 301 includes the following contents:

[0072] Input noise graph G', privacy budget ò and sensitivity Δf of query function f out-link ,

[0073] According to the number of edges |E| in the directed social graph G(V,E), |E| noise values ​​Y=(y1,y2,···,y |E| ),

[0074] Noise optimization, let x i =|y i |, and then convert the optimized noise value into a probability value p i ←F(x i ),

[0075] The probability value p i Add edge e to the noise graph G' i , we get an uncertain directed graph

[0076] This paper combines external link privacy with the uncertain graph algorithm and proposes an optimization algorithm. While meeting the requirements of differential privacy protection, the data utility is improved by optimizing the noise graph. As a variant of differential privacy (edge ​​differential privacy and node differential privacy), external link privacy modifies the definition of the neighborhood data set so that the privacy protection effect and the data utility of the output data set meet the actual needs. As a post-processing of differential privacy, the uncertain graph algorithm does not consume the privacy budget, so that the overall algorithm provides dual privacy protection while meeting the requirements of external link privacy.

[0077] The following explains the superiority of the social network dual privacy protection method (i.e., PEP algorithm) with improved utility from two aspects: privacy analysis and utility analysis.

[0078] The PEP algorithm proposed in this invention satisfies external link privacy.

[0079] Experimental analysis: In order to evaluate the privacy effect of the dual privacy protection method for social networks of the present invention, the edge probability information of the uncertain graph is used to introduce edge entropy to measure the privacy protection degree of the PEP algorithm, and this is used as the evaluation basis to measure the change in the privacy degree before and after the transformation. At the same time, in order to verify the degree of influence of the PEP algorithm on the utility of data, some graph statistical indicators are defined to illustrate the data utility of the PEP algorithm.

[0080] Privacy analysis: In this invention, edge entropy is used to measure the privacy protection effect of the PEP algorithm. The definition of edge entropy is as follows:

[0081]

[0082] In the experiment, the PEP algorithm is compared with or without noise optimization. Tables 1 and 2 show that the edge entropy value changes accordingly with the change of privacy budget. We can conclude that in comparison, noise-free optimization has a larger edge entropy value and better privacy protection effect, but the data utility will be reduced.

[0083]

[0084] Table 1 Edge entropy changes of PEP algorithm without noise optimization

[0085]

[0086] Table 2 Edge entropy changes of PEP algorithm with noise optimization

[0087] Data utility analysis: In this paper, we measure the data utility through degree distribution and compare the degree distribution of PEP algorithm with noise optimization (PEP) and PEP algorithm without noise optimization (PEP-wno) under different privacy budgets. Figure 2 and Figure 3 It can be concluded that the data utility of the PEP algorithm with noise optimization is higher than that of the PEP without noise optimization. At the same time, the larger the privacy budget, the worse the privacy protection, and the better the data utility. Therefore, the present invention improves the utility of data through noise optimization, so that the privacy and utility of data are better balanced.

[0088] The above is only an embodiment of the present invention. A person of ordinary skill in the art can improve and implement the present solution in accordance with his / her own ability under the enlightenment given in the present application. Some typical well-known structures or well-known methods should not become an obstacle for a person of ordinary skill in the art to implement the present application. It should be pointed out that for a person of ordinary skill in the art, several deformations and improvements can be made without departing from the structure of the present invention, which should also be regarded as the protection scope of the present invention, and these will not affect the effect of the implementation of the present invention and the practicality of the patent. The protection scope claimed by the present application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. A method for dual privacy protection of social networks with improved utility, characterized in that: The external link privacy is combined with the uncertain graph algorithm to provide dual privacy protection for the out-degree and in-degree of the directed social graph. in, The steps include: Step 1, directed social graph as input; Step 2: Use external link privacy to protect the out-degree in the directed social graph and obtain a noisy graph; Step 3, using the uncertain graph algorithm to protect the in-degree in the noise graph to obtain an uncertain directed graph; in, Steps 2 and 3 specifically include the following: Step 201: Based on the set privacy budget ò and the sensitivity Δf of the query function f out-link , calculate the parameter λ of the Laplace distribution = Δf out-link / ò, use the Laplace mechanism to protect the out-degree of the node and obtain the noise graph G'; Step 301: The uncertain graph algorithm is used as a post-processing of the external link privacy to protect the in-degree in the noise graph G'. Specifically, the noise value y generated by the Laplace mechanism is used to calculate the corresponding probability Then the probability is added to the edges of the noise graph G' to obtain an uncertain directed graph; The probability density function is Wherein, step 301 specifically includes the following contents: Input noise graph G', privacy budget ò and sensitivity Δf of query function f out-link , According to the number of edges |E| in the directed social graph G(V, E), |E| noise values ​​Y=(y1, y2, ..., y |E| ), Noise optimization, let x i =|y i |, and then convert the optimized noise value into a probability value p i ←F(x i ), The probability value p i Add edge e to the noise graph G' i , we get an uncertain directed graph in, For any out-link neighborhood pair (G1, G2), the global sensitivity of the query function f is expressed as: The outer link neighborhood is defined as: for two given directed graphs G1(V1, E1) and G2(V2, E2), if V1 = V2, E2=E1-{(v1,v2)v1=x}, then G1 and G2 are said to be adjacent in external links; in, External link privacy is: Suppose there is a random algorithm M, Range(M) is the set of all possible outputs of M, for any two external link neighborhood graphs G1 and G2 and any subset S of Range(M), if algorithm M satisfies: Pr(M(G1)∈S)≤e ò ×Pr(M(G2)∈S), Then the random algorithm M is said to provide ò-external link privacy protection, where the parameter ò is called the privacy protection budget; in, The uncertainty graph is defined as: Given an undirected graph G(V, E), if the mapping P:V P →[0, 1] is the probability function of the existence of each edge in the edge set, then G'(V, P) is an uncertain graph about the undirected graph G(V, E), where V p Represents all possible vertex pairs in the vertex set V.

Images