A financial potential risk collaborative analysis method based on fuzzy privacy set intersection

Abstract

The application relates to the field of privacy calculation, and discloses a financial potential risk collaborative analysis method based on fuzzy privacy set intersection, which comprises the following steps: firstly, performing privacy protection on deposit and loan data of each financial institution through Paillier homomorphic encryption; then, realizing fuzzy matching of multi-institution user identities by using a fuzzy privacy set intersection algorithm based on an oblivious key-value store (OKVS) and a discrete logarithm problem; under the premise of ensuring the privacy of user feature data, identifying the same user records satisfying a predetermined distance threshold; finally, performing aggregation and threshold comparison on the ciphertext deposit and loan data of the matched records, decrypting by a trusted third party, and completing financial potential risk identification, so that financial privacy security risk analysis under cross-institution data collaboration is realized.

Description

Technical Field

[0001] This invention relates to the field of privacy computing, specifically to a collaborative analysis method for potential financial risks based on the intersection of fuzzy privacy sets. Background Technology

[0002] In recent years, with the rapid development of fintech, financial institutions have generated massive amounts of customer identity, asset status, and transaction behavior data during their digital transformation. This data is not only a core asset of financial institutions but also a crucial clue for preventing financial risks and identifying potential threats. However, there is an inherent contradiction between efficient data flow and privacy protection. In the field of financial risk control, traditional plaintext sharing or centralized data models are no longer adequate to meet current stringent regulatory requirements due to serious privacy leaks and compliance risks. Therefore, how to achieve collaborative risk information analysis among multiple institutions while meeting privacy compliance has become an urgent issue to be addressed in the field of financial security.

[0003] Currently, financial institutions mainly rely on the following methods to conduct potential risk analysis, but all of them have significant drawbacks:

[0004] 1. Plaintext data sharing: Direct exchange of high-risk lists or transaction details between institutions. While this method is efficient, it seriously violates the Data Security Law and the Personal Information Protection Law. Financial data is extremely sensitive; direct leakage could lead to customer privacy breaches and even trigger systemic compliance risks in the financial system.

[0005] 2. Centralized Blacklist Platform: A unified risk information database is established by regulatory agencies. However, the centralized model carries the risk of single points of failure, and financial institutions, for reasons of business competition and data sovereignty, are often unwilling to upload all underlying data to a third-party platform, resulting in the persistent problem of data silos.

[0006] To address the aforementioned shortcomings, fuzzy privacy set intersection is a feasible technique. As a specific secure multi-party computation protocol, it allows participants to learn the intersection of elements in the sender's set that are less than a certain threshold from elements in the receiver's set, without revealing their original sets. Compared to traditional privacy set intersection, fuzzy privacy set intersection is more fault-tolerant and can identify data with minor differences but essentially pointing to the same entity. In financial risk control, fuzzy privacy set intersection can be used to compare blacklists or identify common suspicious individuals. Summary of the Invention

[0007] To address the technical problem of existing technologies failing to securely match identities and assess risks while protecting user identity and transaction privacy in cross-institutional data collaboration, this invention provides a collaborative analysis method for potential financial risks based on the intersection of fuzzy privacy sets. The technical solution is as follows:

[0008] A collaborative analysis method for potential financial risks based on the intersection of fuzzy privacy sets includes the following steps:

[0009] Step 1: There are m banking financial institutions: ,in As the main organization, each organization have The financial clients are all accounted for by the institutions, and the financial behavior characteristics, deposit and loan data of their respective financial clients are compiled.

[0010] Step 2: Generate a Paillier encrypted public-private key pair through a trusted third party, and save the private key. and public key This information is made public to all of the aforementioned institutions;

[0011] Step 3: All institutions use the public key Each user's deposit and loan data is homomorphically encrypted to generate corresponding ciphertext deposit and loan data.

[0012] Step 4, the main mechanism The main institution performs fuzzy privacy set intersection with other institutions. The system encodes the multidimensional financial behavior features of its users into an elliptic curve point-pair array using an unintentional key-value storage method and distributes it to other institutions.

[0013] Step 5: The remaining institutions perform unintentional key-value storage decoding on the local features using the elliptic curve point-pair array, generating DDH-secure tuples based on the discrete logarithm problem, and return them to the main institution along with encrypted deposit and loan data. ;

[0014] Step 6, Main Organization Verify the DDH security tuple to confirm that the user features of both parties match within a set distance range, identify the records of the same user in different institutions, and complete the intersection operation of the fuzzy privacy set;

[0015] Step 7, Main Organization The encrypted deposit and loan data of the matching records are homomorphically summed, and an encrypted risk discrimination identifier is generated according to a preset loan-to-deposit ratio threshold and sent to the trusted third party.

[0016] Step 8: After decryption by a trusted third party, risk assessment is completed, and users with potential financial risks are reported to the main institution. To complete risk identification and analysis under multi-agency data security collaboration.

[0017] Further, step 1 specifically includes:

[0018] Step 1.1: There are m banking financial institutions: ,in As the main organization, each organization have Financial clients;

[0019] Step 1.2, each institution Financial behavior records are used to record various financial behaviors of financial customers and to compile financial behavior characteristic data, which includes an index of financial institutions. User index And the proportion of d different behavioral characteristics, for each institution The number of financial behavior characteristic data possessed is strip;

[0020] Step 1.3, Each Institution The system compiles user deposit and loan data, which includes an index of financial institutions. User index Deposit amount and loan amount ;

[0021] Step 1.4: All financial institutions jointly determine the data precision t in advance, and set thresholds for all monetary data, proportions of financial behavior characteristics, and loan-to-deposit ratios. Expand in advance times.

[0022] Furthermore, step 3 specifically includes:

[0023] Step 3.1, each institution Expanding the deposit and loan data of their respective users times the deposit amount and loan amount Through public key Homomorphic encryption yields the ciphertext deposit amount. And the amount of the encrypted loan ,in Select random numbers respectively Calculated:

[0024] ,

[0025] ;

[0026] Step 3.2, Each Institution Other content related to retaining deposit and loan data, and , Encapsulate deposit and loan data into encrypted form.

[0027] Furthermore, step 4 specifically includes:

[0028] Step 4.1: All financial institutions jointly set a distance threshold for the intersection of fuzzy privacy sets. and related parameters , Main organization Generate a random value s and select the base point of an elliptic curve. And calculate the points on the elliptic curve: ;

[0029] Step 4.2, Main Organization For each d+2-dimensional financial behavior feature data, in the last d dimensions representing the specific financial behavior feature, each dimension is multiplied by... Obtain a d-dimensional integer vector w;

[0030] Step 4.3, Main Organization It performs the first round of interaction in the two-point fuzzy matching process by performing intersection of the fuzzy privacy set with the other institutions.

[0031] Furthermore, the first round of interaction specifically includes:

[0032] Step 4.3.1, Main Organization Extract the value of each dimension of the integer vector w being fuzzy matched, and generate a list of key-value pairs for inadvertent key-value storage. That is, for each dimension ,as well as Calculate the hash value:

[0033] ,

[0034] Randomly select elliptic curve points ,calculate :

[0035] ;

[0036] Step 4.3.2, Main Organization For each dimension Each operation performs an unintentional key-value storage encoding operation, and the input list of key-value pairs is the one obtained in the previous step. , Length is ,set up The generated length is Furthermore, an array D containing all elements at infinity is obtained by using a striped matrix and Gaussian elimination, which constitutes the data structure. and form a set ;

[0037] Step 4.3.3, Main Organization The generated g, h, and set Encapsulated In the middle, sent to the main organization The remaining organizations that conducted the first round of interaction.

[0038] Furthermore, step 5 specifically includes:

[0039] Step 5.1: For the remaining institutions, in the d+2 dimension elements of the financial behavior characteristic data, in the last d dimensions representing the specific financial behavior characteristics, each dimension is multiplied by... Obtain a d-dimensional integer vector q;

[0040] Step 5.2: The remaining organizations perform the second round of interaction in the two-point fuzzy matching;

[0041] Step 5.3: The remaining institutions find the index of their corresponding financial institution based on the currently matching integer vector q. and user index And retrieve the corresponding encrypted deposit and loan data and send it to the main institution that is currently engaged in a second round of interaction with other institutions. .

[0042] Furthermore, the second round of interaction specifically includes:

[0043] Step 5.2.1: The remaining organizations receive the main organization's... Sent Extract g, h and set ;

[0044] Step 5.2.2, the remaining organizations for each dimension ,calculate and use respectively and Perform an unintentional key-value storage decoding operation, using striped matrix and elliptic curve operations to obtain... corresponding ;

[0045] Step 5.2.3: The remaining mechanisms generate random coefficients a, b, and c, and calculate the elliptic curve points. , :

[0046] ,

[0047] ,

[0048] right Calculate the verification items and store in The formula is:

[0049] ,

[0050] in It is to take the former A truncated hash function of 1 bit;

[0051] Step 5.2.4, the remaining institutions will , and Encapsulated middle.

[0052] Furthermore, step 6 specifically includes:

[0053] Step 6.1, Main Organization In the final round of interaction during the two-point fuzzy matching process, the system first receives data from the remaining organizations. And the accompanying encrypted deposit and loan data;

[0054] Step 6.2, Main Organization calculate:

[0055] ,

[0056] right calculate and verify Is it If the structure is correct, then w and q are successfully fuzzy matched, and the main mechanism... The receiver will receive q and assume that the users corresponding to w and q have a very high probability of having the same identity. Otherwise, if the fuzzy match between w and q fails, the receiver will only receive a random string.

[0057] Step 6.3, Main Organization After performing two-point fuzzy matching on all point pairs with the other institutions, the intersection of the fuzzy privacy set is completed.

[0058] Furthermore, step 7 specifically includes:

[0059] Step 7.1, Main Organization Divide the validated w and q by each dimension. The raw values ​​of the financial behavior characteristic proportions are obtained, and then the indices of the financial institutions corresponding to w and q are found through encrypted deposit and loan data. and user index Join to obtain the index set ;

[0060] Step 7.2, Main Organization Using public key index set Chinese correspondence The total encrypted deposit amount and the total encrypted loan amount are obtained by homomorphically summing all encrypted deposit amounts and the total encrypted loan amount:

[0061] ,

[0062] ,

[0063] Based on the loan-to-deposit ratio threshold Obtain the encrypted risk identification identifier (CFF):

[0064] ,

[0065] The corresponding plaintext operation is to determine whether the total loan-to-deposit ratio, calculated by summing all deposits and all loans, exceeds the threshold. .

[0066] Furthermore, step 8 specifically includes:

[0067] Step 8.1, Main Organization Send all encrypted risk flags (CFF) to a trusted third party;

[0068] Step 8.2: A trusted third party decrypts all encrypted risk identification identifiers CFF using Paillier's public-private key pair to obtain the risk identification identifier FF.

[0069] ,

[0070] A risk assessment is performed to obtain a result flag, where if the following conditions are met... This indicates that the corresponding user faces potential financial risks. ,on the contrary ;

[0071] Step 8.3: The trusted third party sends the results back to the main organization in the same sending order. Main organization Identify the users with potential financial risks based on the order of the results; that is, the index set corresponding to each user. Complete the risk analysis.

[0072] Beneficial effects

[0073] This invention solves the problem of multiple banks and financial institutions being unable to securely match user identities and collaboratively complete encrypted loan-to-deposit ratio risk analysis while protecting the privacy of users' original data by combining fuzzy privacy set intersection and homomorphic encryption technology. This invention supports secure fuzzy matching of user identities across institutions and encrypted computation of risk data, enabling financial institutions to collaboratively analyze potential financial risks without disclosing users' original data throughout the process. This invention supports adjustments to parameters such as plaintext amount precision, distance threshold, and loan-to-deposit ratio threshold during multi-institutional collaboration. This invention supports encrypted protection and storage of sensitive user data such as deposit and loan amounts. Attached Figure Description

[0074] Figure 1 The flowchart shows a collaborative analysis method for potential financial risks based on intersection of fuzzy privacy sets. Detailed Implementation

[0075] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0076] like Figure 1 As shown, the collaborative analysis method for potential financial risks based on the intersection of fuzzy privacy sets of the present invention includes the following steps:

[0077] Step 1: There are m banking financial institutions: ,in As the main organization, each organization have The financial clients are all accounted for by the institutions, and the financial behavior characteristics, deposit and loan data of their respective financial clients are compiled.

[0078] Step 2: Generate a Paillier encrypted public-private key pair through a trusted third party, save the private key, and make the public key public to all the aforementioned organizations;

[0079] Step 3: All institutions use the public key to homomorphically encrypt their respective users' deposit and loan data, generating corresponding ciphertext deposit and loan data;

[0080] Step 4, the main mechanism The main institution performs fuzzy privacy set intersection with other institutions. The system encodes the multidimensional financial behavior features of its users into an elliptic curve point-pair array using an unintentional key-value storage method and distributes it to other institutions.

[0081] Step 5: The remaining institutions perform unintentional key-value storage decoding on the local features using the elliptic curve point-pair array, generating DDH-secure tuples based on the discrete logarithm problem, and return them to the main institution along with encrypted deposit and loan data. ;

[0082] Step 6, Main Organization Verify the DDH security tuple to confirm that the user features of both parties match within a set distance range, identify the records of the same user in different institutions, and complete the intersection operation of the fuzzy privacy set;

[0083] Step 7, Main Organization The encrypted deposit and loan data of the matching records are homomorphically summed, and an encrypted risk discrimination identifier is generated according to a preset loan-to-deposit ratio threshold and sent to the trusted third party.

[0084] Step 8: After decryption by a trusted third party, risk assessment is completed, and users with potential financial risks are reported to the main institution. To complete risk identification and analysis under multi-agency data security collaboration.

[0085] The following example illustrates a method for analyzing potential financial risks based on the intersection of fuzzy privacy sets. Due to limitations in data size, only essential data from financial transaction records and financial behavior characteristics within financial institutions are retained.

[0086] 1) There are a total of 3 financial institutions , , The financial institution IDs are 1001, 1002, and 1003 respectively; The user IDs are 1, 170, 383, 250, 145, 102; The user ID is 2,114,337,391,331,267; The user IDs are 3,495,215,185,446,208. When analyzing financial behavior characteristics, only records with the transaction type "expense" will be analyzed. When calculating encrypted deposit amounts, only records with the transaction type "income" will be counted, and when calculating encrypted loan amounts, only records with the transaction type "loan" will be counted. Also, in this example, all data precision is assumed to be [value missing]. That is, retain two decimal places; financial behavior characteristics are used to calculate proportions in the analysis of financial behavior characteristics, and the corresponding relationships are: (Catering), (Business Travel), (Luxury), (Supermarket), (Gaming), (Medical), (Transport), (Education), (Home Decor), (Invest). All financial behavior records. As shown in Table 1:

[0087] Table 1 Financial Behavior Records

[0088]

[0089]

[0090]

[0091] 2) Various institutions Financial behavior characteristic data obtained after analyzing financial behavior records As shown in Table 2:

[0092] Table 2 Financial Behavior Characteristics Data

[0093]

[0094] Table 2 shows the proportions of different behavioral characteristics for catering, business travel, ..., and investment. This refers to the ratio of the number of times a certain financial behavior feature appears in the total financial behavior records of users with the corresponding financial institution ID and payer ID to the total number of financial behavior records. .

[0095] 3) Various institutions Deposit and loan data obtained after analyzing financial behavior records As shown in Table 3:

[0096] Table 3 Deposit and Loan Data

[0097]

[0098] 4) Trusted third-party TTP settings for Paillier encryption parameters, including... and All are 128-bit:

[0099] =8232962335600616915770699275483536437,

[0100] =326409207837143500258410212373407070817,

[0101] : = = 2687314714116436143307856336978360973300933168309734181688986126248458859029,

[0102] = = 2687314714116436143307856336978360973300933168309734181688986126248458859030,

[0103] : = = 74647630947678781758551564916065582582396972170471946236522367072210229216,

[0104] = 1015770149942764062476780496146980271950363127204174187467287196265276000743.

[0105] In practice, larger safety parameters should be used.

[0106] 5) Various institutions Ciphertext deposit and loan data obtained after encrypting deposit and loan data As shown in Table 4:

[0107] Table 4. Encrypted Deposit and Loan Data

[0108]

[0109]

[0110]

[0111] 6) All financial institutions jointly set the parameters for the intersection of fuzzy privacy sets. , , .

[0112] 7) Main organization In calculation The elliptic curve selected was ed25519, with the following parameters:

[0113] Base point = (15112221349535400772501151409588531511454012693041857206046113283949847762202,46316835694926478169428394003475163141307993866256225615783033603165251855960),

[0114] Step ,

[0115] = ,

[0116] Randomly selected = 2834491037478281003922479298409057567851309021268773670755933920496475843129,

[0117]

[0118] (37025010829379346131399554185013529619670772130707846676863822936938498758408,45652383322953107450331226070380633020787587438837141701905370082325176661299).

[0119] 8) Main organization For each of the 12 dimensions of financial behavioral characteristics data In the last 10 dimensions representing specific financial behavior characteristics, each dimension is multiplied by 100 to obtain a 10-dimensional integer vector w. When j=1, the obtained w=[10, 10, 10, 10, 10, 10, 10, 10, 10, 10].

[0120] 9) Main organization With the other institutions The main organization begins execution of fuzzy privacy set intersection based on two-point fuzzy matching. The list of key-value pairs in the unintentional key-value store (OKVS) obtained after the first round of interaction in the two-point fuzzy matching process. As shown in Table 5, due to data space limitations, only the catering proportion dimension generated by the financial behavior characteristics of financial institution ID 1001 and payer ID 1 is displayed here. 10 data points:

[0121] Table 5. The proportion of catering services generated from the financial behavior characteristics of financial institutions with ID 1001 and payer ID 1. )

[0122]

[0123]

[0124] In this example, the main organization All obtained length .

[0125] 10) Main organization The data structure is obtained by performing OKVS encoding operations. As shown in Table 6, due to space limitations, only the data obtained in the previous step is displayed. After inputting 10 data points:

[0126] Table 6 The data structure is obtained after OKVS encoding.

[0127]

[0128] In this example, the main organization All obtained length .

[0129] 11) Other institutions 12-dimensional elements of financial behavior characteristic data In the last 10 dimensions representing specific financial behavior characteristics, each dimension is multiplied by 100 to obtain a 10-dimensional integer vector q. When i=2 and j=1, the obtained q=[10, 20, 0, 10, 10, 10, 10, 10, 10, 10].

[0130] 12) Other institutions For each dimension ,calculate and use and Perform OKVS decoding. Obtain the results using striped matrix and elliptic curve operations. corresponding When i=2 and k=1, the remaining mechanisms Through OKVS decoding operation Received =((11944517047224790616606920353295856112789423599816634821060250302276189524454,33636729312112958042265219298293047631833659425546567382532954847821716451766),(39556747) The corresponding records for 47493463845046217332173286322701216243102516647014057179757201599181 and 45180418834381529367257044328554930220832300532008251962909039620966094815899 can be found in the last row of Table 5.

[0131] 13) Other institutions The second round of interaction in the two-point fuzzy matching begins, where when i=2, the remaining mechanisms... The result obtained after the second round of interaction in the two-point fuzzy matching , As shown below:

[0132] = (37420440063345167612100431060781475122765475850832753425111285944047363555914, 54053196426769011803548365952672251838751571196658229752459289192981572418782),

[0133] = (16963208381208629907265275952252112301722233358640901765859691069348640999332,5285331227679415431202852318647334910565164046657151692654578042630921109075).

[0134] As shown in Table 7, due to space limitations, only a few data points are presented. 10 data points:

[0135] Table 7 DDH security tuples

[0136]

[0137] 14) Main organization To begin the final round of interaction in the two-point fuzzy matching, the first step is to receive... DDH security tuple sent And accompanying encrypted deposit and loan data When i=2 and j=1, the main organization The calculation yielded:

[0138] 31398117185212647386189104067225307343867224760652094038944169612863207292104882481255192966497192991559729111345,

[0139]

[0140] 31398117543571806218074771335420608979542598718311269266106831455542022190452897195505227550984232021114062201137, This refers to the last row of data in Table 7.

[0141]

[0142] 358359158831885667299278003911287039092791814903111272810455757173185044425682439491786641790952957542400,

[0143] First The corresponding structure after conversion to byte stream Extract points The corresponding byte stream, after conversion, is the point. Points for performing two-point fuzzy matching =[10, 20, 0, 10, 10, 10, 10, 10, 10, 10].

[0144] 15) Main organization With other institutions After performing repeated two-point fuzzy matching, the intersection of the fuzzy privacy set is completed.

[0145] 16) Main organization The final set of indexes The corresponding encrypted deposit amounts (TCDA) are: [(1001, 1), (1002, 2), (1003, 3)] and [(1001, 383), (1002, 114), (1003, 495)], with total TCDAs of 279126175296172924698755539847603959283062688318343220589875694013713548002172186181876778742 1413395097707079089742828636320333986358320823582170093988 and 714757979785309540928772842064931557175866481683752044418597825107824079077955467058409537824535556929366284968885893077 235645324759395816465665650188, the total encrypted loan amount TCLA is 6154562316557069342271282485481096811994019887423954271795115995163250982860054533069657222630897068866896988657804717276816858502225 970599294535273766 and 732376123152644380117310310108918902214440649116238608890535897822447792347030065867644052457404224393099541186099947070849624203392068001699141274022, then based on the loan-to-deposit ratio threshold. =0.5, the resulting ciphertext risk discrimination identifiers (CFFs) are 73237612315264438011731031010891890221444064911623860889053589782244779234703006586764405245740422439309954118609994707084962420339206800169 9141274022 and 4089369878025710614448489277199261705721636483655895593776384269026043341948035873404027403386113161736476900705525565365075266285068148731022317671158.

[0146] 17) The FFs decrypted from the CFF obtained by the trusted third party TTP are 6003763150 and 71981050, respectively. Since n=2687314714116436143307856336978360973300933168309734181688986126248458859029, Therefore, the final set of Flags is {True, True}.

[0147] 18) The trusted third-party TTP sends the results back to the host organization in the same sending order. Main organization Based on the order of the results, identify the users with potential financial risks. In this example, the two users with potential financial risks have their respective index sets. The indices are {(1001, 1), (1002, 2), (1003, 3)} and {(1001, 383), (1002, 114), (1003, 495)}. Furthermore, the actual deposit and loan data corresponding to the above index sets are statistically analyzed, as shown in Table 8.

[0148] Table 8. Statistics of deposit and loan data corresponding to the matched index sets.

[0149]

[0150] Further calculations show that the loan-to-deposit ratios for these two users are 789174.9 / 377597.24 = 2.08 and 7448.5 / 500.79 = 14.87, respectively, both exceeding the set threshold for loan-to-deposit ratios. =0.5. If the two users found through fuzzy matching from these three institutions are indeed the same person, then both of them do indeed pose financial risks, thus completing the risk analysis.

[0151] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A collaborative analysis method for potential financial risks based on the intersection of fuzzy privacy sets, characterized in that, Includes the following steps: Step 1: There are m banking financial institutions: ,in As the main organization, each organization have The financial clients are all accounted for by the institutions, and the financial behavior characteristics, deposit and loan data of their respective financial clients are compiled. Step 2: Generate a Paillier encrypted public-private key pair through a trusted third party, and save the private key. and public key This information is made public to all of the aforementioned institutions; Step 3: All institutions use the public key Each user's deposit and loan data is homomorphically encrypted to generate corresponding ciphertext deposit and loan data. Step 4, the main mechanism The main institution performs fuzzy privacy set intersection with other institutions. The system encodes the multidimensional financial behavior features of its users into an elliptic curve point-pair array using an unintentional key-value storage method and distributes it to other institutions. Step 5: The remaining institutions perform unintentional key-value storage decoding on the local features using the elliptic curve point-pair array, generating DDH-secure tuples based on the discrete logarithm problem, and return them to the main institution along with encrypted deposit and loan data. ; Step 6, Main Organization Verify the DDH security tuple to confirm that the user features of both parties match within a set distance range, identify the records of the same user in different institutions, and complete the intersection operation of the fuzzy privacy set; Step 7, Main Organization The encrypted deposit and loan data of the matching records are homomorphically summed, and an encrypted risk discrimination identifier is generated according to a preset loan-to-deposit ratio threshold and sent to the trusted third party. Step 8: After decryption by a trusted third party, risk assessment is completed, and users with potential financial risks are reported to the main institution. To complete risk identification and analysis under multi-agency data security collaboration.

2. The collaborative analysis method for potential financial risks based on the intersection of fuzzy privacy sets as described in claim 1, characterized in that, Step 1 specifically involves: Step 1.1: There are m banking financial institutions: ,in As the main organization, each organization have Name of financial clients; Step 1.2, each institution Financial behavior records are used to record various financial behaviors of financial customers and to compile financial behavior characteristic data, which includes an index of financial institutions. User index And the proportion of d different behavioral characteristics, for each institution The number of financial behavior characteristic data possessed is strip; Step 1.3, Each Institution The system compiles user deposit and loan data, which includes an index of financial institutions. User index Deposit amount and loan amount ; Step 1.4: All financial institutions jointly determine the data precision t in advance, and set thresholds for all monetary data, proportions of financial behavior characteristics, and loan-to-deposit ratios. Expand in advance times.

3. The collaborative analysis method for potential financial risks based on the intersection of fuzzy privacy sets as described in claim 2, characterized in that, Step 3 specifically involves: Step 3.1, each institution Expanding the deposit and loan data of their respective users times the deposit amount and loan amount Through public key Homomorphic encryption yields the ciphertext deposit amount. And the amount of the encrypted loan ,in Select random numbers respectively Calculated: ， ； Step 3.2, Each Institution Other content related to retaining deposit and loan data, and , Encapsulate deposit and loan data into encrypted form.

4. The collaborative analysis method for potential financial risks based on the intersection of fuzzy privacy sets as described in claim 3, characterized in that, Step 4 specifically involves: Step 4.1: All financial institutions jointly set a distance threshold for the intersection of fuzzy privacy sets. and related parameters , Main organization Generate a random value s and select the base point of an elliptic curve. And calculate the points on the elliptic curve: ; Step 4.2, Main Organization For each d+2-dimensional financial behavior feature data, in the last d dimensions representing the specific financial behavior feature, each dimension is multiplied by... Obtain a d-dimensional integer vector w; Step 4.3, Main Organization It performs the first round of interaction in the two-point fuzzy matching process by performing intersection of the fuzzy privacy set with the other institutions.

5. The collaborative analysis method for potential financial risks based on the intersection of fuzzy privacy sets as described in claim 4, characterized in that, The first round of interaction is as follows: Step 4.3.1, Main Organization Extract the value of each dimension of the integer vector w being fuzzy matched, and generate a list of key-value pairs for inadvertent key-value storage. That is, for each dimension ,as well as Calculate the hash value: ， Randomly select elliptic curve points ,calculate : ； Step 4.3.2, Main Organization For each dimension Each operation performs an unintentional key-value storage encoding operation, and the input list of key-value pairs is the one obtained in the previous step. , Length is ,set up The generated length is Furthermore, an array D containing all elements at infinity is obtained by using a striped matrix and Gaussian elimination, which constitutes the data structure. and form a set ; Step 4.3.3, Main Organization The generated g, h, and set Encapsulated In the middle, sent to the main organization The remaining organizations that conducted the first round of interaction.

6. The collaborative analysis method for potential financial risks based on the intersection of fuzzy privacy sets as described in claim 5, characterized in that, Step 5 specifically involves: Step 5.1: For the remaining institutions, for the d+2 dimension elements of the financial behavior characteristic data, in the last d dimensions representing the specific financial behavior characteristics, each dimension is multiplied by... Obtain a d-dimensional integer vector q; Step 5.2: The remaining organizations perform the second round of interaction in the two-point fuzzy matching; Step 5.3: The remaining institutions find the index of their corresponding financial institution based on the currently matching integer vector q. and user index And retrieve the corresponding encrypted deposit and loan data and send it to the main institution that is currently engaged in a second round of interaction with other institutions. .

7. The collaborative analysis method for potential financial risks based on the intersection of fuzzy privacy sets as described in claim 6, characterized in that, The second round of interaction is as follows: Step 5.2.1: The remaining organizations receive the main organization's... Sent Extract g, h and set ; Step 5.2.2, the remaining organizations for each dimension ,calculate and use respectively and Perform an unintentional key-value storage decoding operation, using striped matrix and elliptic curve operations to obtain... corresponding ; Step 5.2.3: The remaining mechanisms generate random coefficients a, b, and c, and calculate the elliptic curve points. , : ， ， right Calculate the verification items and store in The formula is: ， in It is to take the former A truncated hash function of 1 bit; Step 5.2.4, the remaining institutions will , and Encapsulated middle.

8. The collaborative analysis method for potential financial risks based on the intersection of fuzzy privacy sets as described in claim 7, characterized in that, Step 6 specifically involves: Step 6.1, Main Organization In the final round of interaction during the two-point fuzzy matching process, the system first receives data from the remaining organizations. And the accompanying encrypted deposit and loan data; Step 6.2, Main Organization calculate: ， right calculate and verify Is it If the structure is correct, then the fuzzy match between w and q is successful, and the main mechanism... The receiver will receive q and assume that the users corresponding to w and q have a very high probability of having the same identity. Otherwise, if the fuzzy match between w and q fails, the receiver will only receive a random string. Step 6.3, Main Organization After performing two-point fuzzy matching on all point pairs with the other institutions, the intersection of the fuzzy privacy set is completed.

9. The collaborative analysis method for potential financial risks based on the intersection of fuzzy privacy sets as described in claim 8, characterized in that, Step 7 specifically involves: Step 7.1, Main Organization Divide the validated w and q by each dimension. The raw values ​​of the financial behavior characteristic proportions are obtained, and then the indices of the financial institutions corresponding to w and q are found through encrypted deposit and loan data. and user index Join to obtain the index set ; Step 7.2, Main Organization Using public key index set Chinese correspondence The total encrypted deposit amount and the total encrypted loan amount are obtained by homomorphically summing all encrypted deposit amounts and the total encrypted loan amount: ， ， Based on the loan-to-deposit ratio threshold Obtain the encrypted risk identification identifier (CFF): ， The corresponding plaintext operation is to determine whether the total loan-to-deposit ratio, calculated by summing all deposits and all loans, exceeds the threshold. .

10. The collaborative analysis method for potential financial risks based on the intersection of fuzzy privacy sets as described in claim 9, characterized in that, Step 8 specifically involves: Step 8.1, Main Organization Send all encrypted risk flags (CFF) to a trusted third party; Step 8.2: A trusted third party decrypts all encrypted risk identification identifiers CFF using Paillier's public-private key pair to obtain the risk identification identifier FF. ， A risk assessment is performed to obtain a result flag, where if the following conditions are met... This indicates that the corresponding user faces potential financial risks. ,on the contrary ; Step 8.3: The trusted third party sends the results back to the main organization in the same sending order. Main organization Identify the users with potential financial risks based on the order of the results; that is, the index set corresponding to each user. Complete the risk analysis.

Images