CN113159918B - A Bank Customer Group Mining Method Based on Federal Regiment Penetration - Google Patents

Abstract

The invention relates to a bank client group mining method based on federal group penetration. The Coordinator (Coordinator) represents a data aggregator. The method comprises the following steps: privacy protection is carried out between all the bank ends based on a client network to obtain intersection so as to obtain overlapped clients; each bank terminal respectively calculates a client local similarity matrix, encrypts the matrix by using a homomorphic encryption technology and sends the matrix to a coordination terminal; the coordination terminal aggregates the local similarity matrixes to obtain an encrypted global similarity matrix, and sends the encrypted global similarity matrix to each bank terminal; each bank end finds all k groups on an overlapped client network consisting of overlapped clients to obtain a k group set; each bank end performs k-group penetration on the overlapped client network according to the decrypted global similarity matrix and the k-group set; and each bank end divides the bank client group according to the k-group penetration result and outputs the mining structure of the bank client group. The invention can combine the customer information of a plurality of banks to improve the mining accuracy of bank customer groups under the condition of protecting the data privacy of bank customers.

Description

A Bank Customer Group Mining Method Based on Federal Regiment Penetration

technical field

The invention relates to the technical field of federated learning, in particular to a method for mining bank customer groups based on the penetration of federated groups.

Background technique

Banks need to understand customer behavior and conduct value mining around customer needs. At the same time, this also means that banks need to obtain more customer information for accurate analysis. In recent years, Internet privacy leakage incidents have emerged one after another. At the same time, privacy leakage incidents have attracted more and more users' attention, and the government has paid more and more attention to network security. The European Union promulgated the General Protection Regulation (GDPR) in 2018 to protect users' data privacy, and many countries such as China and the United States have also formulated and improved a series of privacy protection regulations to punish privacy breaches. Federated learning is a decentralized, privacy-preserving distributed machine learning framework proposed by Google. It supports the distributed parallel processing of large-scale data with decentralized distributed computing, and guarantees banking through local computing and encrypted transmission. The secret data of the terminal is not leaked during the calculation process. It is of great value to study the mining method of bank customer groups based on the penetration of federation. While protecting the data privacy of bank customers, it combines the customer information of multiple banks to mine bank customer groups, which does not violate privacy protection laws, and can make full use of customer data owned by banks to help them more accurately mine bank customers. Groups, in turn, can build high-quality bank customer profiles, conduct targeted advertising, and detect financial crimes.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide a bank customer group mining method based on federal group penetration, which can more accurately divide bank customer groups while protecting the privacy of bank customers.

In order to achieve the above object, the technical solution of the present invention is: a method for mining bank customer groups based on federal group penetration, and a system is provided, which includes a bank-side overlapping customer identification module, a bank-side customer similarity calculation module, and a coordination terminal. Customer similarity aggregation module, bank-side customer network k-group discovery module, bank-side customer network K-group penetration module and bank-side customer group division module; the system performs bank customer group mining according to the following steps:

Step S1, the bank-end overlapping customer identification module reads the bank customer network G(V, E, R, A) at each bank end P _h respectively, where V represents the customer set, E represents the edge set, and R represents the feature set, A is the customer characteristic matrix; randomly select a bank to generate the RSA encryption algorithm key pair, and send the RSA public key to all other banks; the selected bank uses the RSA public key to encrypt the customer ID, and uses the RSA public key with other banks respectively. The customer ID encrypted by the key is used to calculate the intersection; the selected bank terminal seeks the public intersection of the obtained intersection points to obtain the overlapping customer set, and sends the overlapping customer set to other bank terminals, and each bank terminal P _h obtains the public overlapping customer set X _h ;

Step S2, the bank-end customer similarity calculation module randomly selects a bank-end to generate a key pair of the homomorphic encryption algorithm, and sends the key pair to all other bank-ends; the bank-end P _h calculates the dimension of the customer characteristic matrix A _h |a _h |, encrypt |a _h | with the public key of the homomorphic encryption algorithm, and send it to the coordinator; among them, a _h is the characteristic vector of the customer, and the coordinator is the data aggregator, which aggregates the encrypted data sent by P _h ; The coordinator receives the encrypted customer feature matrix dimension | _ah | sent by each bank end P _h , adds the encrypted state | _ah | to obtain the global customer feature matrix dimension, and sends the global customer feature matrix dimension to each bank Terminal _Ph ; each bank terminal _Ph uses the private key of the homomorphic encryption algorithm to decrypt the global customer characteristic matrix dimension, and calculates the local similarity between overlapping customers according to the global customer characteristic matrix dimension and the customer characteristic matrix A _h , and obtains the partial similarity of overlapping customers. Similarity matrix _Sh ; each bank end _Ph uses the homomorphic encryption algorithm public key to encrypt _Sh , and sends _Sh to the coordinator;

Step S3, the coordinating end client similarity aggregation module receives the client local similarity matrix _Sh sent by each bank end P _h at the coordinating end; the coordinating end adds the encrypted state _Sh to obtain an encrypted client global similarity matrix , and send the global similarity matrix to each bank terminal _Ph ;

Step S4, the bank-side client network k-group discovery module discovers all k-groups on the overlapping customer network formed by overlapping customers at each bank-side P _h , and obtains a k-group set;

Step S5, the bank-end client network k-group penetration module uses the homomorphic encryption algorithm private key at each bank end P _h to decrypt the customer global similarity matrix sent by the coordinating end; each bank end P _h decrypts the global similarity matrix according to the decrypted global similarity matrix. Perform k-group infiltration with the k-group set to get the group map

的连通分支，每一个连通分支内的节点集合即为一个银行客户群体，连通分支集合即为所述银行客户网络G上重叠客户X_h的群体划分C；输出最终银行客户网络的群体划分结果C。Step S6, the bank customer group division module calculates the group graph of each bank terminal _Ph respectively

connected branches, the node set in each connected branch is a bank customer group, and the connected branch set is the group division C of overlapping customers X _h on the bank customer network G; output the group division result C of the final bank customer network .

In an embodiment of the present invention, the step S1 specifically includes the following steps:

Step S11, randomly select a bank end P _i(i∈h) to generate an RSA key pair, and send the RSA public key to other bank end P _{j(j∈h∩j≠i)} ;

Step S12, the bank end P _i encrypts the client V _i of the bank client network G _i with the RSA public key, seeks intersection with other bank end P _j under privacy protection respectively, and obtains X _i,j with the RSA private key decryption and decryption;

Step S13, the bank end P _i seeks a common intersection customer for the obtained intersection customers X _i,j , and obtains an overlapping customer set X _i =∪{X _i,j };

Step S14, the bank end P _i sends the overlapping client set X _i to the other bank ends P _j , and all the bank ends _{Ph obtain the overlapping client set X h =} X _i .

In an embodiment of the present invention, the step S2 specifically includes the following steps:

Step S21, the bank end P _i generates a homomorphic encryption algorithm key pair, and sends the key pair to other bank ends P _j ;

Step S22, each bank end P _h calculates the dimension | _ah | of the customer feature matrix A _h , and encrypts | _ah | with the public key of the homomorphic encryption algorithm to obtain the encrypted local customer feature matrix dimension E(| _ah |) , and send E(| _ah |) to the coordinator; where E() is the encryption function;

Step S23, the coordinating terminal receives the encrypted local customer characteristic matrix dimension E(| _ah |) sent by each bank terminal _Ph ;

Step S24, the coordinator adds E(| _ah |) to obtain the global customer feature matrix dimension

发送给各银行端P_h；Step S25, the coordinator will encrypt the encrypted

Send to each bank terminal _Ph ;

使用同态加密算法的私钥解密

得到全局客户特征矩阵维度

其中，D()是解密函数；Step S26, the bank terminal _Ph receives the data sent by the coordinating terminal.

Private key decryption using homomorphic encryption algorithm

Get the global customer feature matrix dimension

Among them, D() is the decryption function;

和客户特征矩阵A_h计算重叠客户X_h的局部相似度矩阵S_h；其中，重叠客户间相似度计算如公式(4)所示；Step S27, the bank terminal _Ph according to

Calculate the local similarity matrix S _h of overlapping customers X _h with the customer feature matrix A _h ; wherein, the similarity calculation between overlapping customers is as shown in formula (4);

是异或运算操作，|a_i|是特征向量a_i的长度，s(a_i，a_j)代表客户v_i和v_j的相似度；where a _i and a _j are the attribute vectors of overlapping customers v _i and v _j ,

is the XOR operation, |a _i | is the length of the feature vector a _i , and s(a _i , a _j ) represents the similarity between customers v _i and v _j ;

Step S28, the bank terminal Ph encrypts _Sh using the public key of the homomorphic encryption algorithm to obtain E(S _h ), and sends E(S _{h )} to the coordinator.

In an embodiment of the present invention, the step S3 specifically includes the following steps:

Step S31, the coordinating end receives the encrypted local similarity matrix E(S _h ) sent by each bank end P _h ;

Step S32, the coordinating end adds the local similarity matrix E(S _h ) to obtain the global similarity matrix

发送给各银行端P_h。Step S33, the global similarity matrix that the coordinator will encrypt

Sent to each bank terminal _Ph .

In an embodiment of the present invention, the step S4 specifically includes the following steps:

Step S41, the bank end P _h calculates the overlapping customer network composed of the overlapping customer set X _h on G _h

中的k团，得到k团集合；其中，k团是一个由k个重叠客户组成的子客户网络，子客户网络中每个重叠客户与其他所有重叠客户都存在关联关系。Step S42, the bank terminal _Ph uses the k-group discovery algorithm to find overlapping customer networks

The k group in , obtains the k group set; among them, the k group is a sub-customer network composed of k overlapping customers, and each overlapping customer in the sub-customer network has an association relationship with all other overlapping customers.

In an embodiment of the present invention, the step S5 specifically includes the following steps:

得到

Step S51, the bank terminal _Ph uses the private key of the homomorphic encryption algorithm to decrypt the global similarity matrix

get

Step S52, the bank end _Ph uses each k group in the k group set as a node to construct a group graph

计算两个k团之间的相似度，如果相似度大于所设定的阈值α，则为这两个k团添加一条边到

中。其中，阈值α取0.8，两个k团之间的相似度计算如公式(5)所示；Step S53, the bank end _Ph according to the decrypted global similarity matrix

Calculate the similarity between two k-clusters, if the similarity is greater than the set threshold α, add an edge to the two k-clusters to

middle. Among them, the threshold α is taken as 0.8, and the similarity calculation between the two k groups is shown in formula (5);

中的客户，C_p和C_q是k团，s(v_i，v_j)是客户v_i和v_j的相似度，s(C_p，C_q)是k团C_p和C_q的相似度，Ind(v_i，v_j)函数表示如果

中客户v_i和v_j存在关联关系，则返回1，否则返回0。where v _i and v _j are overlapping client networks

Customers in , C _p and C _q are k-clusters, s(vi, v _j ) is the similarity of customers vi and v _j , _s (C _p , C _q ) is the similarity of _k -clusters C _p and C _q degree, the Ind(v _i , v _j ) function indicates that if

If there is an association relationship between customers v _i and v _j , it returns 1, otherwise it returns 0.

In an embodiment of the present invention, the step S6 specifically includes the following steps:

的连通分支；Step S61, bank-side _Ph calculation group graph

connected branch of

Step S62, the bank end _Ph merges all the customers of each connected branch into one customer group, and obtains the bank customer group set C;

Step S63, the bank terminal Ph writes the customers v _{i ,j} in each group C _i in the bank customer group set C into a row vector form R _i =(vi _,j );

Step S64, the output vector set {R _i }, 0<i<m, m is the number of customer groups, and each row represents a customer group.

The present invention also provides a computer-readable storage medium on which computer program instructions that can be executed by a processor are stored. When the processor executes the computer program instructions, the above-mentioned method steps can be implemented.

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

(1) The present invention proposes a bank customer group mining method based on federal group penetration, which can more accurately mine the relationship between bank customers while protecting the privacy of bank customers.

(2) The present invention proposes a new group similarity measure index, which takes into account the network topology and characteristic information at the same time, and improves the accuracy of dividing bank customer groups by group penetration.

Description of drawings

FIG. 1 is a flowchart of an embodiment of the present invention.

Detailed ways

The technical solutions of the present invention will be described in detail below with reference to the accompanying drawings.

It should be noted that the following detailed description is exemplary and intended to provide further explanation of the present application. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

It should be noted that the terminology used herein is for the purpose of describing specific embodiments only, and is not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly dictates otherwise, the singular is intended to include the plural as well, furthermore, it is to be understood that when the terms "comprising" and/or "including" are used in this specification, it indicates that There are features, steps, operations, devices, components, and/or combinations thereof.

As shown in FIG. 1 , this embodiment provides a method for mining bank customer groups based on federal group penetration, and provides a system that includes a bank-side overlapping customer identification module, a bank-side customer similarity calculation module, and a coordination-end customer Similarity aggregation module, bank-end customer network k-group discovery module, bank-end customer network K-group penetration module and bank-end customer group division module; the system performs bank customer group mining according to the following steps:

Step S1, each bank end _Ph of the bank-end overlapping customer identification module reads the bank customer network G(V, E, R, A) respectively, where V represents the customer set, E represents the edge set, R represents the feature set, and A is the customer characteristic matrix. Randomly select a bank to generate the RSA encryption algorithm key pair, and send the RSA public key to all other banks. The selected bank uses the RSA public key to encrypt the customer ID, and calculates the intersection with the customer ID encrypted by other banks using the RSA public key. The selected bank terminal seeks a common intersection of the obtained intersection points to obtain an overlapping customer set, and sends the overlapping customer set to other bank terminals, and each bank terminal P _h obtains a common overlapping customer set X _h ;

Step S2, the bank client similarity calculation module randomly selects a bank end to generate a key pair of the homomorphic encryption algorithm, and sends the key pair to all other bank ends. The bank side _Ph calculates the dimension | _{ah | of the customer characteristic matrix A h ,} encrypts | _ah | with the public key of the homomorphic encryption algorithm, and sends it to the coordinator. Among them, a _h is the feature vector of the customer, and the coordinator is the data aggregator, which aggregates the encrypted data sent by P _h . The coordinating end receives the encrypted customer characteristic matrix dimension | _ah | sent by each bank end P _h , adds the encrypted state |a _h | to obtain the global customer characteristic matrix dimension, and sends the global customer characteristic matrix dimension to each bank end _Ph . Each bank end P _h uses the private key of the homomorphic encryption algorithm to decrypt the global customer characteristic matrix dimension, calculates the local similarity between overlapping customers according to the global customer characteristic matrix dimension and the customer characteristic matrix A _h , and obtains the local similarity matrix S of overlapping customers _h . Each bank end _{Ph uses the homomorphic encryption algorithm public key to encrypt Sh , and sends Sh to the} coordinator;

Step S3, the coordinating end client similarity aggregation module receives the client local similarity matrix _Sh sent by each bank end _Ph at the coordinating end. The coordinating terminal adds the encrypted state _Sh to obtain the encrypted global similarity matrix of customers, and sends the global similarity matrix to each bank terminal _Ph ;

Step S4, the bank-side client network k-group discovery module discovers all k-groups on the overlapping customer network formed by overlapping customers at each bank-side P _h , and obtains a k-group set;

Step S5, each bank end P _h of the bank end client network k group penetration module decrypts the customer global similarity matrix sent by the coordinating end using the private key of the homomorphic encryption algorithm; each bank end P _h according to the decrypted global similarity matrix and The k-group set performs k-group infiltration to obtain a group map

的连通分支，每一个连通分支内的节点集合即为一个银行客户群体，连通分支集合即为所述银行客户网络G上重叠客户X_h的群体划分C。输出最终银行客户网络的群体划分结果C。Step S6, the bank customer group division module calculates the group graph of each bank terminal _Ph respectively

The connected branch of each connected branch is a bank customer group, and the connected branch set is the group division C of overlapping customers X _h on the bank customer network G. Output the group division result C of the final bank customer network.

Further, the step S1 specifically includes the following steps:

Step S11, randomly select a bank end P _i(i∈h) to generate an RSA key pair, and send the RSA public key to other bank end P _{j(j∈h∩j≠i)} ;

Step S12, the bank end P _i encrypts the client V _i of the bank client network G _i with the RSA public key, seeks intersection with other bank end P _j under privacy protection respectively, and obtains X _i,j with the RSA private key decryption and decryption;

Step S13, the bank end P _i seeks a common intersection customer for the obtained intersection customers X _i,j , and obtains an overlapping customer set X _i =∪{X _i,j };

Step S14, the bank end P _i sends the overlapping client set X _i to the other bank ends P _j , and all the bank ends _{Ph obtain the overlapping client set X h =} X _i .

Further, the step S2 specifically includes the following steps:

Step S21, the bank end P _i generates a homomorphic encryption algorithm key pair, and sends the key pair to other bank ends P _j ;

Step S22, each bank end P _h calculates the dimension | _ah | of the customer feature matrix A _h , and encrypts | _ah | with the public key of the homomorphic encryption algorithm to obtain the encrypted local customer feature matrix dimension E(| _ah |) , and send E(| _ah |) to the coordinator; where E() is the encryption function;

Step S23, the coordinating terminal receives the encrypted local customer characteristic matrix dimension E(| _ah |) sent by each bank terminal _Ph ;

Step S24, the coordinator adds E(| _ah |) to obtain the global customer feature matrix dimension

发送给各银行端P_h；Step S25, the coordinator will encrypt the encrypted

Send to each bank terminal _Ph ;

使用同态加密算法的私钥解密

得到全局客户特征矩阵维度

其中，D()是解密函数；Step S26, the bank terminal _Ph receives the data sent by the coordinating terminal.

Private key decryption using homomorphic encryption algorithm

Get the global customer feature matrix dimension

Among them, D() is the decryption function;

和客户特征矩阵A_h计算重叠客户X_h的局部相似度矩阵S_h；其中，重叠客户间相似度计算如公式(7)所示；Step S27, the bank terminal _Ph according to

Calculate the local similarity matrix S _h of overlapping customers X _h with the customer feature matrix A _h ; wherein, the similarity calculation between overlapping customers is as shown in formula (7);

是异或运算操作，|a_i|是特征向量a_i的长度，s(a_i，a_j)代表客户v_i和v_j的相似度；where a _i and a _j are the attribute vectors of overlapping customers v _i and v _j ,

is the XOR operation, |a _i | is the length of the feature vector a _i , and s(a _i , a _j ) represents the similarity between customers v _i and v _j ;

Step S28, the bank terminal Ph encrypts _Sh using the public key of the homomorphic encryption algorithm to obtain E(S _h ), and sends E(S _{h )} to the coordinator.

Further, the step S3 specifically includes the following steps:

Step S31, the coordinating end receives the encrypted local similarity matrix E(S _h ) sent by each bank end P _h ;

Step S32, the coordinating end adds the local similarity matrix E(S _h ) to obtain the global similarity matrix

发送给各银行端P_h。Step S33, the global similarity matrix that the coordinator will encrypt

Sent to each bank terminal _Ph .

Further, the step S4 specifically includes the following steps:

Step S41, the bank end P _h calculates the overlapping customer network composed of the overlapping customer set X _h on G _h

中的k团，得到k团集合。其中，k团是一个由k个重叠客户组成的子客户网络，子客户网络中每个重叠客户与其他所有重叠客户都存在关联关系。Step S42, the bank terminal _Ph uses the k-group discovery algorithm to find overlapping customer networks

The k groups in , get the set of k groups. Among them, the k group is a sub-customer network composed of k overlapping customers, and each overlapping customer in the sub-customer network is associated with all other overlapping customers.

Further, the step S5 specifically includes the following steps:

得到

Step S51, the bank terminal _Ph uses the private key of the homomorphic encryption algorithm to decrypt the global similarity matrix

get

Step S52, the bank end _Ph uses each k group in the k group set as a node to construct a group graph

计算两个k团之间的相似度，如果相似度大于所设定的阈值α，则为这两个k团添加一条边到

中。其中，阈值α取0.8，两个k团之间的相似度计算如公式(8)所示；Step S53, the bank end _Ph according to the decrypted global similarity matrix

Calculate the similarity between two k-clusters, if the similarity is greater than the set threshold α, add an edge to the two k-clusters to

middle. Among them, the threshold α is taken as 0.8, and the similarity calculation between the two k groups is shown in formula (8);

中的客户，C_p和C_q是k团，s(v_i，v_j)是客户v_i和v_j的相似度，s(C_p，C_q)是k团C_p和C_q的相似度，Ind(v_i，v_j)函数表示如果

中客户v_i和v_j存在关联关系，则返回1，否则返回0。where v _i and v _j are overlapping client networks

Customers in , C _p and C _q are k-clusters, s(vi, v _j ) is the similarity of customers vi and v _j , _s (C _p , C _q ) is the similarity of _k -clusters C _p and C _q degree, the Ind(v _i , v _j ) function indicates that if

If there is an association relationship between customers v _i and v _j , it returns 1, otherwise it returns 0.

Further, the step S6 specifically includes the following steps:

的连通分支；Step S61, bank-side _Ph calculation group graph

connected branch of

Step S62, the bank end _Ph merges all the customers of each connected branch into one customer group, and obtains the bank customer group set C;

Step S63, the bank terminal Ph writes the customers v _{i ,j} in each group C _i in the bank customer group set C into a row vector form R _i =(vi _,j );

Step S64, the output vector set {R _i }, 0<i<m, m is the number of customer groups, and each row represents a customer group.

As will be appreciated by those skilled in the art, the embodiments of the present application may be provided as a method, a system, or a computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present application. It will be understood that each flow and/or block in the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to the processor of a general purpose computer, special purpose computer, embedded processor or other programmable data processing device to produce a machine such that the instructions executed by the processor of the computer or other programmable data processing device produce Means for implementing the functions specified in a flow or flow of a flowchart and/or a block or blocks of a block diagram.

These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory result in an article of manufacture comprising instruction means, the instructions The apparatus implements the functions specified in the flow or flows of the flowcharts and/or the block or blocks of the block diagrams.

These computer program instructions can also be loaded on a computer or other programmable data processing device to cause a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process such that The instructions provide steps for implementing the functions specified in the flow or blocks of the flowcharts and/or the block or blocks of the block diagrams.

The above are only preferred embodiments of the present invention, and are not intended to limit the present invention in other forms. Any person skilled in the art may use the technical content disclosed above to make changes or modifications to equivalent changes. Example. However, any simple modifications, equivalent changes and modifications made to the above embodiments according to the technical essence of the present invention without departing from the content of the technical solutions of the present invention still belong to the protection scope of the technical solutions of the present invention.

Claims (6)

1. A bank client group mining method based on federal group penetration is characterized in that a bank client group mining system based on federal group penetration is provided, and the system comprises a bank end overlapping client identification module, a bank end client similarity calculation module, a coordination end client similarity aggregation module, a bank end client network k group discovery module, a bank end client network k group penetration module and a bank end client group division module; the system carries out bank customer group mining according to the following steps:

step S1, the bank end overlapping customer identification module is at each bank end P_hRespectively reading a bank client network G (V, E, R, A), wherein V represents a client set, E represents an edge set, R represents a feature set, and A is a client feature matrix; randomly selecting one bank end to generate an RSA encryption algorithm key pair, and sending an RSA public key to other bank ends; the selected bank end encrypts the client ID by using the RSA public key, and respectively calculates the intersection with the client IDs encrypted by using the RSA public key of other bank ends; the selected bank end calculates the public intersection of the obtained intersections to obtain an overlapped client set, and sends the overlapped client set to other bank ends, wherein each bank end P_hObtaining a common overlapping customer set X_h；

Step S2, the bank client similarity calculation module randomly selects a bank end to generate a key pair of a homomorphic encryption algorithm, and sends the key pair to other bank ends; each bank terminal P_hCalculating a customer feature matrix A_hDimension | a of_h| a, using a homomorphic cryptographic algorithm public key pair_hThe | is encrypted and sent to a coordination end; wherein, a_hIs the characteristic vector of the client, the coordinating end is the data aggregation party, and the aggregation of all bank ends P_hThe transmitted encrypted data; the coordination terminal receives each bank terminal P_hTransmitted encrypted client feature matrix dimension | a_h| a for encryption status_hL is added to obtain the dimension of the global customer feature matrixAnd sends the global customer feature matrix dimension to each bank terminal P_h(ii) a Each bank terminal P_hDecrypting the global customer feature matrix dimension using a homomorphic encryption algorithm private key based on the global customer feature matrix dimension and the customer feature matrix A_hCalculating the local similarity between the overlapped clients to obtain a local similarity matrix S of the overlapped clients_h(ii) a Each bank terminal P_hPublic key encryption S using homomorphic encryption algorithm_hAnd send S_hTo the coordinating end;

step S3, the coordination terminal client similarity aggregation module receives each bank terminal P at the coordination terminal_hTransmitted local similarity matrix S_h(ii) a Coordinating S for port pair encryption status_hAdding the obtained data to obtain an encrypted client global similarity matrix, and sending the global similarity matrix to each bank terminal P_h；

Step S4, the bank client network k group discovery module is at each bank end P_hFinding all k groups on an overlapped client network consisting of overlapped clients to obtain a k group set; wherein, the k groups are a sub-client network consisting of k overlapped clients, and each overlapped client in the sub-client network has an association relation with all other overlapped clients;

step S5, the k-group infiltration module of the bank client network is arranged at each bank end P_hDecrypting the client global similarity matrix sent by the coordination terminal by using a homomorphic encryption algorithm private key; each bank terminal P_hPerforming k-group infiltration according to the decrypted global similarity matrix and the k-group set to obtain a group diagram

Step S6, the bank customer group division module calculates each bank end P_hGroup chart of

The node set in each connected branch is a bank customer group, and the connected branch set is an overlapped customer set X on the bank customer network G_hA set of customer groups C; outputting a client group set C of the final bank client network;

the step S1 specifically includes the following steps:

step S11, randomly selecting a bank end P_iGenerating RSA encryption algorithm key pair and sending RSA public key to other bank end P_j；

Step S12, bank end P_iEncrypting bank client network G with RSA public key_iClient V of_iRespectively with other bank terminals P_jSolving intersection and obtaining X by RSA private key decryption_i，j；

Step S13, bank end P_iFor the obtained intersection X_i，jObtaining public intersection to obtain public overlapping client set X_i＝∪{X_i，j}；

Step S14, bank end P_iSending a common overlapping customer set X_iTo other bank end P_jEach bank end P_hObtaining a common overlapping customer set X_h＝X_i；

The step S2 specifically includes the following steps:

step S21, bank end P_iGenerating a homomorphic encryption algorithm key pair, and sending the key pair to other bank terminals P_j；

Step S22, each bank end P_hCalculating a customer feature matrix A_hDimension | a of_hPublic key encryption using homomorphic encryption algorithm | a_hObtaining the dimension E (| a) of the encrypted local customer feature matrix_hAnd E (| a) and_h|) to the coordinating peer; wherein E () is an encryption function;

step S23, the coordination end receives each bank end P_hThe sent encrypted local client characteristic matrix dimension E (| a)_h|)；

Step S24, coordinating the end pair E (| a)_h|) are added to obtain the global customer feature matrix dimension

Step S25, the coordinating end encrypts the wholeLocal customer feature matrix dimension

To the bank terminals P_h；

Step S26, each bank end P_hTransmitted from the receiving and coordinating end

Private key decryption using homomorphic encryption algorithm

To obtain

Wherein D () is a decryption function;

step S27, each bank end P_hAccording to

And a customer feature matrix A_hCalculating a local similarity matrix S of overlapping customers_h(ii) a Wherein, the similarity calculation between the overlapped clients is shown in formula (1);

wherein, a_iIs overlapping clients v_iAttribute vector of a_jIs overlapping clients v_jIs determined by the attribute vector of (a),

is an XOR operation, | a_iIs the feature vector a_iDimension of, s (a)_i，a_j) Representative client v_iAnd v_jThe similarity of (2);

step S28, bank end P_hPublic key pair S using homomorphic encryption algorithm_hPerforming encryption to obtain E (S)_h) And transmitting E (S)_h) To the coordinator side.

2. The bank customer group mining method based on federal group infiltration as claimed in claim 1, wherein the step S3 specifically comprises the following steps:

step S31, the coordination end receives each bank end P_hThe encrypted local similarity matrix E (S) is sent_h)；

Step S32, harmonize end pair E (S)_h) Adding to obtain a global similarity matrix

Step S33, the coordination terminal converts the global similarity matrix into a global similarity matrix

To the bank terminals P_h。

3. The bank customer group mining method based on federal group infiltration as claimed in claim 2, wherein the step S4 specifically comprises the following steps:

step S41, bank end P_hAt G_hUpper computing overlapping client network consisting of overlapping clients

Step S42, bank end P_hFinding overlapping client networks using k-clique discovery algorithm

And obtaining a k-group set.

4. The bank customer group mining method based on federal group infiltration as claimed in claim 3, wherein the step S5 specifically comprises the following steps:

step S51, each bank end P_hPrivate key decryption using homomorphic encryption algorithm

To obtain

Step S52, each bank terminal P_hConstructing a clique graph by taking each k clique in the k clique set as a node

Step S53, each bank end P_hAccording to decryption

Calculating the similarity between the two k groups, and if the similarity is greater than a preset threshold value alpha, adding an edge to the two k groups

The preparation method comprises the following steps of (1) performing; wherein the threshold value alpha is 0.8, and the similarity between two k groups is calculated as shown in the following formula;

wherein v is_iAnd v_jIs an overlapping customer network

Customer in (1), C_pAnd C_qIs a group k, s (v)_i，v_j) Is client v_iAnd v_jSimilarity of (C), s (C)_p，C_q) Is a k group C_pAnd C_qSimilarity of (d), Ind (v)_i，v_j) Function is expressed if

Chinese client v_iAnd v_jIf the association relationship exists, 1 is returned, otherwise, 0 is returned.

5. The bank customer group mining method based on federal group infiltration as claimed in claim 4, wherein the step S6 specifically comprises the following steps:

step S61, each bank end P_hCalculating a clique chart

A connected branch of (a);

step S62, each bank end P_hAll the clients of each connected branch are merged into a client group to obtain a bank client group set C;

step S63, each bank end P_hCollecting each group C in bank customer group C_iClient v in (1)_i，jWritten in the form of row vectors R_i＝(v_i，j)；

Step S64, output vector set { R }_i}，0<i<m, m is the number of customer groups, and each row represents one customer group.

6. A computer-readable storage medium, having stored thereon computer program instructions executable by a processor, the computer program instructions being capable of, when executed by the processor, implementing the method steps of any of claims 1-5.

Images