A greedy strategy-based method to resist user identity association in social networks

Abstract

The present invention relates to the technical field of social network analysis, and discloses a method for resisting social network user identity association based on a greedy strategy. In order to effectively model and solve the problem of resisting social network user identity association and balance costs and benefits, an account injection scheme based on a greedy strategy is proposed, which models the attack problem on the user identity association model as a problem of maximizing the number of wrong pairings, so that the similarity scores of these pairings must be greater than the similarity scores between existing account pairs, so that the attack problem on social network user identity association can be solved; based on the greedy strategy, a vulnerability assessment method based on account network roles is used to identify vulnerabilities, and a progressive attack from easy to difficult is realized; at the same time, an injection strategy search method based on a greedy strategy is designed to determine which existing accounts should initiate friend connections between the injected account and the friends, so as to achieve efficient attacks at low cost.

Description

Technical Field

[0001] The present invention relates to the technical field of social network analysis, and in particular to a method for resisting social network user identity association based on a greedy strategy. Background Art

[0002] The goal of social network user identity linkage (UIL) is to identify whether accounts on different social network platforms belong to the same user. In recent years, billions of users around the world use multiple social network platforms at the same time every day, and social network UIL has become a very popular research field. In addition, it is also the basis for many downstream applications such as e-commerce recommendations, information dissemination analysis, user identity authentication, and cross-network user profiling. Today, researchers in industry and academia are constantly enhancing the capabilities of UIL algorithms, models, and tools from various perspectives, including accuracy, efficiency, scalability, and so on.

[0003] Although social network UIL has brought good news to various applications, it also comes with significant network security risks. Malicious actors can use related tools to integrate user sensitive information that may be scattered on different social platforms, such as occupation, age, address, email, hobbies, friend relationships, and daily commuting methods. Many related codes and tools are now available on open source sites. If someone enters "social network UIL" or other related terms such as "user identity connection", "anchor connection prediction" and "graph matching" on well-known code sharing platforms such as GitHub, they will find a large number of related code libraries. These code libraries cover user association methods for various social network platforms, such as QQ, WeChat, Weibo, etc. Since malicious actors also have the opportunity to download these code libraries, users of these platforms are at risk of sensitive privacy leakage, and their families may become victims of fraud or related manipulation. Therefore, we urgently need to study anti-social network UIL methods to effectively protect users from these threats while ensuring the availability of social networks.

[0004] From a technical perspective, current research on anti-UIL in social networks usually follows the approach of network modification attack (NMA), which is to perturb the relationship network structure by manipulating the connections between existing accounts. Its basic operations include deleting existing connections, modifying existing connections, or adding new connections between existing accounts. For example: by adding and deleting connections of existing accounts, accounts are pushed to areas with denser connections, thereby reducing the original closeness of user associations; or finding the most important connections for correct identity associations and attacking social networks by removing them.

[0005] In order to carry out the attack, all of these studies require the manipulation of a large number of existing accounts. However, since only users representing accounts have the right to add or remove social connections, the above methods must be carried out by gaining the trust of a large number of social network users and making them perform the required deletion or addition operations, or by controlling these accounts through hacking. Obviously, it is not easy to achieve the former, and resorting to the latter will face serious legal consequences. Therefore, from the perspective of technical necessity, we urgently need to find a more practical way to disturb the original relationship network structure and protect user privacy. Summary of the invention

[0006] In view of the above problems, the purpose of the present invention is to provide a method for resisting social network user identity association based on a greedy strategy. By optimizing the injection attack strategy, it can be operated within a limited budget and maximize the interference with social network analysis; it not only improves the efficiency of the attack, but also enhances the robustness and adaptability of the existing strategy. The technical solution is as follows:

[0007] A method for resisting social network user identity association based on a greedy strategy includes the following steps:

[0008] Step 1: Modeling the anti-social network user identity association model:

[0009] On social networks Inject the account and inject the account and social network Establish friend connections between matching accounts in the Unmatched accounts in The similarity score must be greater than the unmatched account With social networks The maximum similarity score between all unmatched accounts in , thus modeling the problem of anti-social network user identity association based on account injection as a problem of maximizing the number of determined wrong pairings;

[0010] Step 2: Account vulnerability assessment:

[0011] By calculating the similarity score between accounts to evaluate the vulnerability of each account, a vulnerability assessment method based on account network roles is proposed to identify the accounts that are most vulnerable to attacks and cause false associations;

[0012] Step 3: Injection based on greedy strategy:

[0013] An account injection method based on a greedy strategy is adopted to sort the accounts in descending order according to their vulnerability, so as to give priority to attacking accounts with high vulnerability under a given budget. The injected accounts are allowed to establish friend relationships with anchor users who are neighbors of the vulnerable accounts first, and the friend adding strategy of the injected accounts is effectively obtained, thereby ensuring the effectiveness of the attack while maintaining a low cost.

[0014] Furthermore, the step 1 is specifically as follows:

[0015] Step 1.1: For any social network (Can be or Any one), while maintaining the adjacency matrix representing the friend relationship In the case where the original friend relationship remains unchanged, new accounts are injected to generate disturbances, and a social network is constructed within the budget Δ according to the account injection attack , the expression is:

[0016] (1);

[0017] In the formula, represents the adjacency matrix of the entire social network after the account is injected. is the adjacency matrix of the friend relationship between the injected account and the existing accounts, is the adjacency matrix of the friend relationship of the injected account, To inject new accounts into the social network;

[0018] The constraint expression of the account injection attack is:

[0019] (2);

[0020] Where Δ is the budget, which is the number of friend add requests that the injected account can initiate; Indicates the size of the network, that is, the total number of edges in the network;

[0021] Step 1.2: Get the objective function of the attack, that is, within a given budget, minimize the number of associated anchor users by injecting accounts. The expression is:

[0022] (3);

[0023] In the formula, is the set of unobserved anchor user pairs, For social networks The adjacency matrix representing the friend relationship after the account is injected, For social networks The adjacency matrix representing the friend relationship after the account is injected, For social networks No matching account in Unmatched account Anchor user, is a sign function, if the value in the brackets is true, the result is 1, otherwise it is 0; For a given social network UIL model, and Social networks and social networks Implement the injected adjacency matrix representing the friendship relationship, and The budgets allocated to the two social networks respectively;

[0024] Step 1.3: Inject your account and social network Unmatched accounts to be protected The similarity score is greater than the unmatched account With social networks The maximum similarity score among all unmatched accounts in , the objective function is further transformed into:

[0025] (4);

[0026] In the formula, For social networks The collection of accounts injected into is the similarity score of the two accounts, To inject an account, For social networks Any unmatched account in For social networks The set of all unmatched accounts in , For social networks The collection of all unmatched accounts in .

[0027] Furthermore, the step 2 is specifically as follows:

[0028] Step 2.1: Define parameters:

[0029] Define matching neighbors: Given two accounts and , both belong to social networks And they are friends, and the account On social networks If the anchor user in is known, then it is called an account It is an account The matching neighbors of

[0030] Defining common matching neighbors: Given an anchor user pair , a social network Accounts in and one in social network Accounts in ;

[0031] If the account and account There is a friend relationship between accounts and account There is also a friend relationship between them, then the anchor user It is an account and account Common matching neighbors; at the same time, given a social network An account set in , if each account in the set is related to account Have a friend relationship and account On social networks If the anchor user in is known, then it is called an account Is an account set Common matching neighbors of all accounts in;

[0032] Step 2.2: Use To indicate the unmatched account to be protected The matching neighbor set of Representative account pair The number of common matching neighbors is calculated as follows: Vulnerability , the expression is:

[0033] (5);

[0034] In the formula, is a hyperparameter used to distinguish the importance of different influencing factors; “:” indicates conditional restrictions.

[0035] Furthermore, the step 3 is specifically as follows:

[0036] Step 3.1: Search all social networks in descending order The unmatched accounts are sorted in reverse order of vulnerability, the most vulnerable account is attacked first, and then other unmatched accounts are attacked in turn;

[0037] Step 3.2: Calculate the budget allocated to the injection node for precision attack, which is the sum of two factors; factor one is the selected account and social network The maximum number of common matching neighbors among all accounts in the network; the second factor is a factor related to the difference, which ensures that the similarity score of the injected account is higher, thereby achieving the effectiveness of the interference; among them, the number of common matching neighbors assigned to the social network is Used to attack unmatched accounts The budget is determined by the following formula:

[0038] (6);

[0039] Where sign (•) is the sign function;

[0040] According to the above formula, the injected account only needs to match the unmatched account Add the corresponding account of the matching neighbor Friends connect;

[0041] Step 3.3: Select different account combinations, give the selected account combinations a certain budget and launch attacks; for each selected account, The injected account initiates a friend connection request to the anchor users of the selected account's matching neighbors. When the number of anchor users who can initiate the add request exceeds the budget, the anchor users of the selected account's matching neighbors are sorted in reverse order according to the size of their node degrees, and accounts with larger node degrees are added in sequence within the given budget, thereby achieving the attack purpose while influencing other accounts to the greatest extent in a greedy manner.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] 1. The present invention proposes an account injection scheme based on a greedy strategy, which models the attack problem on the user identity association model as a problem of maximizing the number of incorrect pairings, so that the similarity scores of these pairings must be greater than the similarity scores between existing account pairs, thereby making the attack problem on social network user identity association solvable, achieving effective modeling and solving of the anti-social network user identity association problem, and balancing costs and benefits.

[0044] 2. The account injection scheme based on the greedy strategy of the present invention adopts a vulnerability assessment method based on the account network role to identify vulnerabilities and implement a progressive attack from easy to difficult.

[0045] 3. The present invention designs an injection strategy search method based on a greedy strategy to determine which existing accounts should initiate friend connections between the injection account, thereby achieving efficient attacks at low cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 The flowchart of the method for anti-social network user identity association based on the greedy strategy of the present invention.

[0047] Figure 2 for Figure 1 A magnified view of the injection node schematic in the lower right corner of the middle right box. DETAILED DESCRIPTION

[0048] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0049] In order to effectively model and solve the problem of resisting social network user identity association and balance costs and benefits, the present invention proposes a greedy strategy-based account injection solution (GSAIS). GSAIS models the attack problem on the user identity association model as a problem of maximizing the number of incorrect pairings, so that the similarity scores of these pairings must be greater than the similarity scores between existing account pairs, so that the attack problem on social network user identity association can be solved. Specifically, GSAIS uses a vulnerability assessment method based on account network roles to identify vulnerabilities and implement progressive attacks from easy to difficult. At the same time, a greedy strategy-based injection strategy search method is designed to determine which existing accounts should initiate friend connections between the injected account and the friends, so as to achieve efficient attacks at low cost.

[0050] Before introducing the specific steps, let’s first define the relevant concepts as follows:

[0051] First, use To represent a social network. Given two social networks, the first and second social networks are represented as and . On social networks middle, Is the account set, is the set of edges, Using the adjacency matrix Indicates that Indicates friendship and exist.

[0052] Friendship: exists in Friends connection in .

[0053] Anchor users and anchor links: Accounts from two different social networks but belonging to the same entity are called anchor users, and the connection between them is called anchor link. ) and on WeChat ( ) belong to the same user. For ease of analysis, a virtual connection can usually be added between these accounts to represent this relationship. and social networks , we can call them anchor links. Called Account and vice versa. We label this relationship as or .

[0054] Matching neighbors: Given two accounts and , they are both part of the social network , if they are friends and the account On social networks The anchor user in is known, which can be called account It is an account The matching neighbors of .

[0055] Common matching neighbors: Given an anchor user pair , one in the network Accounts in and one in the network Accounts in ; If the account and There is a friend relationship between accounts and There is also a friend relationship between them, then the anchor user It's an account and Common matching neighbors; at the same time, given a social network An account set in , if each account in the set is related to account Have a friend relationship and account On social networks If the anchor user in is known, then it is called an account Is an account set Common matching neighbors of all accounts in .

[0056] The anti-social network UIL method based on greedy strategy designed by the present invention is as follows Figure 1 and Figure 2 As shown in Figure 1, SMN (Social Media Network) represents a real social network in reality. Ⅰ and SMN Ⅱ are two real social networks; accounts with the same subscript number represent each other as anchor users, such as and , red dot Express to social network The blue and purple dots connected in the two social networks are matched accounts, and the rest are unmatched accounts.

[0057] The present invention is mainly divided into three steps: anti-social network UIL model modeling, account vulnerability assessment and injection based on greedy strategy. The specific implementation method is as follows:

[0058] 1. Modeling of anti-social network UIL model:

[0059] The anti-social network UIL method based on account injection is modeled as an optimization problem. The optimization goal is to maximize the number of determined wrong pairings, so that the similarity scores of these pairings must be greater than the similarity scores between existing account pairs, so that the attack problem on the social network user identity association method can be solved. The specific steps are as follows:

[0060] Step 1.1: The research goal is to solve the UIL problem in social networks based on a greedy strategy. Inject the account and inject the account and social network By establishing friend connections between matching accounts in the network, the structure of the relationship network can be changed, thereby misleading the UIL model or algorithm to match accounts of different users.

[0061] Step 1.2: Known social networks and To effectively interfere with the discrimination process of the UIL algorithm or model, we can make the similarity score of the injected account greater than that of the unmatched account. With social networks This is achieved by maximizing the similarity score among all unmatched accounts in the network. In this way, the anti-social network UIL based on account injection is transformed into an optimization problem that can be solved by an algorithm.

[0062] Currently, many studies have explored the possibility of modifying the original social network The network modification attack can be divided into three types: adding, deleting and modifying the set. L Friendship within.

[0063] This paper proposes a new method to attack social network related tools or models, namely, account injection attack NIA (Node Injection Attack). L In the case that the original friendship relationship remains unchanged, Inject new accounts to create disturbances.

[0064] First, according to the account injection attack NIA is built within the budget Δ The expression is:

[0065] (1);

[0066] In the formula, represents the adjacency matrix of the entire social network after the account is injected. To inject new accounts into social networks, is the adjacency matrix of the friend relationship between the injected account and the existing accounts, is the adjacency matrix of the friend relationship of the injected account, and the attack targets and Expand.

[0067] Since the cost of adding a friend relationship between two injected accounts is much lower than adding a friend relationship between an injected account and an existing account, the constraint expression for the account injection attack is:

[0068] (2);

[0069] There are two main aspects to focus on: the number of injected accounts and the number of friend relationships added by the injected accounts.

[0070] The purpose of this constraint is to ensure the availability of the social networking platform. While unrestricted injection may be effective, it may cause false data to overwhelm real information. At the same time, the cost and defender detection issues also need to be considered.

[0071] From this, we can get the objective function of the attack, that is, within a given budget, the expression for minimizing the number of anchor users that may be associated by injecting accounts is:

[0072] (3);

[0073] In the formula, is the set of unobserved anchor user pairs, For social networks The adjacency matrix representing the friend relationship after the account is injected, For social networks The adjacency matrix representing the friend relationship after the account is injected, for No matching account in Unmatched account Anchor user, is a sign function, if the value in the brackets is true, the result is 1, otherwise it is 0; For a given social network UIL model, and Social networks and Implement the injected adjacency matrix representing the friendship relationship, and Budget allocated to the two social networks respectively.

[0074] In order to better model the algorithm, it is important to simulate the attack process of UIL. Assume that given a social network and , when processing an unmatched account When the anti-social network UIL algorithm or model usually calculates the unmatched account and The similarity score of each unmatched account in Then, these similarity scores are sorted in reverse order, and the account with the highest score will be selected as The corresponding account.

[0075] In order to carry out the attack, the unmatched account With its anchor user The similarity score between them is less than the unmatched account and The similarity score between one or more other accounts in . Assume , then the objective function is transformed from formula (3) to:

[0076] (4);

[0077] In the formula, is the similarity score of the two accounts, To inject an account, For social networks Any unmatched account in For social networks The set of all unmatched accounts in , For social networks The collection of all unmatched accounts in .

[0078] This is for The same objective function applies to the injected account.

[0079] Because the anchor user of the account is not matched It is unknown and cannot be determined Which account is The real An alternative is to allow the injection account to and The similarity score is greater than and The maximum similarity score among all unmatched accounts in , then the goal can be further transformed into:

[0080] (5);

[0081] In the formula, yes A collection of accounts to be injected into.

[0082] In the above manner, the anti-social network UIL problem is modeled as a problem of maximizing the number of determined wrong pairings, the principle of which is to make the similarity score of the injected account pair exceed the score of the original account pair. This method effectively disrupts the discrimination of the UIL algorithm or model, making the method of anti-social network UIL attack proposed by the present invention effective.

[0083] 2. Vulnerability analysis based on account network roles (account vulnerability assessment):

[0084] By analyzing the existing UIL method to calculate the similarity score between accounts to evaluate the vulnerability of each account, a vulnerability assessment method based on account network role is proposed to identify the accounts that are most vulnerable to attacks and cause false associations for UIL.

[0085] The existing UIL methods for calculating similarity scores between accounts are analyzed and summarized. Existing account similarity calculation methods include neighborhood-based social network UIL methods and embedding-based social network UIL methods. According to the characteristics and rules of these methods, the influence of various factors on account vulnerability is summarized. The two aspects are mainly the number of common matching neighbors and the number of matching neighbors. It is precisely because they include the difference between the number of matching neighbors and the number of common matching neighbors, unmatched accounts and their matching neighbors, possible equivalent accounts in other networks, matching neighbors of possible equivalent accounts, and their common matching neighbors, which affect the account network positioning in many aspects, so all objects that affect the account network identity, including them, are collectively referred to as the account's network role.

[0086] Based on UIL, a method for determining the network role of an account is summarized, and an account vulnerability calculation method that covers all factors affecting the network role of an account is proposed to identify the accounts that are most vulnerable to attacks and cause incorrect associations for UIL.

[0087] In order to achieve the goal modeled by formula (5), if the budget is limited, accounts that are more susceptible to interference in the network can be attacked first. The concept of "vulnerability" is used to describe the degree of vulnerability of an account. Accounts with higher vulnerability are more susceptible to UIL attacks, so the cost required is lower. Accounts with lower vulnerability are more resistant to account-associated UIL attacks and require higher costs. Therefore, a method is needed to distinguish the vulnerabilities of different accounts in order to find the optimal attack strategy.

[0088] In order to identify the vulnerability of accounts, the existing account similarity calculation methods are analyzed, including the neighborhood-based social network UIL method and the embedding-based social network UIL method. The results show that as the number of common matching neighbors increases, the similarity score tends to become larger. ,if and The number of common matching neighbors between Therefore, interference is used to increase Exceed It may require higher costs, and it may also be found that the difference between the number of matching neighbors and the number of common matching neighbors will also affect the vulnerability of existing accounts. The network role of the account proposed in the present invention covers but is not limited to all of the above factors that affect the social network positioning of the account.

[0089] use express The set of matching neighbors of Representative account pair The number of common matching neighbors is used to calculate the unmatched accounts Vulnerability , the expression is:

[0090] (6);

[0091] In the formula, is a hyperparameter used to distinguish the importance of different influencing factors.

[0092] This formula ensures that the more matching neighbors an account has among all neighbors, the more susceptible it is to precise attacks and therefore the more vulnerable it is. Conversely, the more common matching neighbors an account has, the higher the cost of precise attacks on it, making it less vulnerable. In addition, when the difference between the number of matching neighbors and the number of common matching neighbors is large, it is easier to make the similarity score of other account pairs exceed the similarity score of the correct pair. In general, this formula also satisfies all the above rules for determining account vulnerability through account similarity calculation methods.

[0093] 3. Optimal injection strategy retrieval (injection based on greedy strategy):

[0094] A greedy strategy-based account injection method is adopted to sort the accounts in descending order according to their vulnerability, so as to give priority to attacking accounts with high vulnerability under a given budget. The injected accounts are allowed to establish friend relationships with peer accounts of the neighbors of the vulnerable accounts first, and the friend adding strategy of the injected accounts is effectively obtained, thereby ensuring the effectiveness of the attack while maintaining low cost.

[0095] After obtaining the vulnerabilities of different unmatched accounts, we consider how to inject accounts into the network. Based on the greedy strategy, we develop an optimal injection strategy retrieval method to determine which friend relationships should be added between the injected account and the existing accounts in order to achieve a high impact of the attack effect at a low cost.

[0096] First, after all unmatched accounts are sorted according to the account vulnerability calculation method proposed by the present invention, given one or more injection accounts and their budgets, the accounts that are most vulnerable to attack and produce incorrect associations for UIL are attacked first, and then the attacks are carried out in turn according to the greedy strategy. The budget is the number of friend add requests that an account can initiate.

[0097] Secondly, a budget allocation scheme for each existing account is proposed to determine the minimum budget required for each account to be successfully attacked. The budget allocated to the injected node for precise attack is calculated as the sum of the maximum number of common matching neighbors between the selected account and all accounts in another network and a factor related to the difference, so as to ensure that the similarity score of the injected account is higher, thereby achieving the effectiveness of the interference.

[0098] Finally, a budget-based account injection strategy is proposed. Different account combinations are selected, and then a certain budget is given to the selected account combination and the attack is carried out. The account combination is a combination of accounts that can accept friend addition requests under a given budget to achieve the attack. For each selected account, on another network, the injected account is used to initiate a friend connection request to the peer account of the selected account's matching neighbor. When the number of peer accounts that can initiate an addition request exceeds the budget, the peer accounts of the selected account's matching neighbor are sorted in reverse order according to the size of the node degree, and the accounts with larger degrees are added in turn within the given budget, so as to achieve the purpose of the attack while affecting other accounts to the greatest extent in a greedy manner. The node degree is the number of edges connected to the account.

[0099] Specifically: all social networks in descending order The unmatched accounts are ranked by their vulnerabilities, and the most vulnerable accounts are attacked first. Then these unmatched accounts are attacked in turn to ensure that these accounts are not The similarity score of the corresponding account in Africa is greater. The attack did not match the account The budget is determined by the following formula:

[0100] (7);

[0101] In the formula, sign (•) is the sign function, that is, the factor 2 in step 3.2 is the dynamic factor of the difference. According to formula (7), the injected account only needs to be Add the corresponding account of the matching neighbor This method ensures that the injected account is consistent with The similarity score must be greater than , while minimizing the required budget.

[0102] In the specific implementation strategy, for a given budget, the accounts to be attacked are sorted in reverse order of vulnerability according to formula (6) and the attacks are carried out in sequence; for the attacking account, the budget is allocated according to formula (7), and on another network, the injected account is used to initiate a friend connection request to the peer account of the selected account's matching neighbor to execute the attack.

[0103] When the number of peer accounts that can initiate add requests exceeds the budget, the peer accounts of the selected account matching neighbors are sorted in reverse order according to the size of the degree, and the accounts with the largest degree are added in sequence within the given budget. Because the larger the degree of the account, the more friends it has, and the easier it is to add strangers as friends - this reverse connection addition method not only meets the budget limit, but also increases the complexity and uncertainty of the network, making the interference behavior more covert and safe. In the above way, it is possible to achieve the purpose of the attack while affecting other accounts to the greatest extent. Thus, the problem is solved.

[0104] The advantage of the strategy proposed in this invention is that it can effectively disturb or enhance the structure of the network even when resources are limited. By adopting the above strategy, the injection attack strategy can be optimized to operate within a limited budget while maximizing the interference with social network analysis. This not only improves the efficiency of the attack, but also enhances the robustness and adaptability of existing strategies.

[0105] The method proposed in the present invention attacks the social network UIL, perturbing the structure of the social network in a practical and feasible injection attack method in real scenarios, thereby countering the algorithm or model that associates the anchor user's account and protecting user privacy. The present invention proposes this account injection attack framework based on a greedy strategy to achieve this goal, striving to ensure that the cost is balanced while achieving the maximum interference effect. This framework has the potential to prevent social network UIL tools from being abused by malicious actors, thereby avoiding user privacy leakage, economic losses and reputation damage. In addition, it can also be used to evaluate the robustness of UIL algorithms and models, thereby helping to promote the beneficial application of related tools.

Claims

1. A method for anti-social network user identity association based on a greedy strategy, characterized in that: The following steps are involved: Step 1: Modeling the anti-social network user identity association model: On social networks Inject the account and inject the account and social network Establish friend connections between matching accounts in the Unmatched accounts in The similarity score must be greater than the unmatched account With social networks The maximum similarity score between all unmatched accounts in , thus modeling the problem of anti-social network user identity association based on account injection as a problem of maximizing the number of determined wrong pairings; Step 2: Account vulnerability assessment: By calculating the similarity scores between all unmatched accounts on two social networks, the vulnerability of each unmatched account is evaluated. A vulnerability assessment method based on account network roles is proposed to identify the most vulnerable accounts among the unmatched accounts to be incorrectly associated. Step 3: Injection based on greedy strategy: An account injection method based on a greedy strategy is adopted to sort the unmatched accounts in descending order according to their vulnerability, so that the vulnerable accounts ranked first are attacked first under a given budget. The injected account is allowed to establish friendships with the anchor users who are neighbors of the vulnerable accounts ranked first, and the friend adding strategy of the injected account is effectively obtained, thereby ensuring the effectiveness of the attack while maintaining a low cost.

2. According to claim 1, a greedy strategy-based anti-social network user identity association method is characterized in that: The step 1 is specifically as follows: Step 1.1: For any social network , while maintaining the adjacency matrix representing the friend relationship In the case where the original friend relationship remains unchanged, new accounts are injected to generate disturbances, and a social network is constructed within the budget Δ according to the account injection attack , the expression is: （1）； In the formula, represents the adjacency matrix of the entire social network after the account is injected; is the adjacency matrix of the friend relationship between the injected account and the existing accounts, The adjacency matrix of the friend relationship of the injected account; To inject new accounts into the social network; The constraint expression of the account injection attack is: （2）； Where Δ is the budget, which is the number of friend add requests that the injected account can initiate; Indicates the size of the network, that is, the total number of edges in the network; Step 1.2: Get the objective function of the attack, that is, within a given budget, minimize the number of associated anchor users by injecting accounts: （3）； In the formula, is the set of unobserved anchor user pairs, For social networks The adjacency matrix representing the friend relationship after the account is injected, For social networks The adjacency matrix representing the friend relationship after the account is injected, For social networks No matching account in Unmatched account Anchor users; is a sign function, if the value in the brackets is true, the result is 1, otherwise it is 0; For a given social network UIL model, and Social Networks and social networks Implement the injected adjacency matrix representing the friendship relationship, and The budgets allocated to the two social networks respectively; Step 1.3: Inject your account and social network Unmatched accounts to be protected The similarity score is greater than the unmatched account With social networks The maximum similarity score among all unmatched accounts in , the objective function is further transformed into: （4）； In the formula, For social networks The collection of accounts injected into is the similarity score of the two accounts, To inject an account, For social networks Any unmatched account in For social networks The set of all unmatched accounts in , For social networks The collection of all unmatched accounts in .

3. According to claim 2, a greedy strategy-based anti-social network user identity association method is characterized in that: The step 2 is specifically as follows: Step 2.1: Define parameters: Define matching neighbors: Given two accounts and , both belong to social networks And they are friends, and the account On social networks If the anchor user in is known, then it is called an account It is an account The matching neighbors of Defining common matching neighbors: Given an anchor user pair , a social network Accounts in and one in social network Accounts in ; If the account and account There is a friend relationship between accounts and account There is also a friend relationship between them, then the anchor user It is an account and account Common matching neighbors; at the same time, given a social network An account set in , if each account in the set is related to account Have a friend relationship and account On social networks If the anchor user in is known, then it is called an account Is an account set Common matching neighbors of all accounts in; Step 2.2: Use To indicate the unmatched account to be protected The set of matching neighbors of Representative account pair The number of common matching neighbors is calculated as follows: Vulnerability , the expression is: （5）； In the formula, is a hyperparameter used to distinguish the importance of different influencing factors; ":" indicates conditional restrictions.

4. According to claim 3, a greedy strategy-based anti-social network user identity association method is characterized in that: The step 3 is specifically as follows: Step 3.1: Search all social networks in descending order Sort the unmatched accounts in reverse order of vulnerability, attack the most vulnerable account first, and then attack other unmatched accounts in turn; Step 3.2: Calculate the budget allocated to the injection node for precision attack, which is the sum of two factors; factor one is the selected account and social network The maximum number of common matching neighbors among all accounts in the network; the second factor is a factor related to the difference, which ensures that the similarity score of the injected account is higher, thereby achieving the effectiveness of the interference; among them, the number of common matching neighbors assigned to the social network is Used to attack unmatched accounts The budget is determined by the following formula: （6）； Where sign (•) is the sign function; According to the above formula, the injected account only needs to match the unmatched account Add the corresponding account of the matching neighbor Friends connect; Step 3.3: Select different account combinations, give the selected account combinations a certain budget and launch attacks; for each selected account, The injected account initiates a friend connection request to the anchor users of the selected account's matching neighbors. When the number of anchor users who can initiate the add request exceeds the budget, the anchor users of the selected account's matching neighbors are sorted in reverse order according to the size of their node degrees, and accounts with larger node degrees are added in sequence within the given budget, thereby achieving the attack purpose while influencing other accounts to the greatest extent in a greedy manner.

Images