Credit product intelligent recommendation method and system based on multi-source data fusion

Abstract

The invention provides an intelligent credit product recommendation method and system based on multi-source data fusion, and the method comprises the steps: carrying out the feature extraction of multi-source credit data, and obtaining a feature vector set; performing data fusion on each feature vector in the feature vector set to obtain fused feature data; performing homomorphic encryption on the fused feature data to obtain homomorphic encrypted data, and storing the homomorphic encrypted data in each security node; performing similarity calculation and clustering operation in a ciphertext space through a secret sharing protocol based on each security node to obtain an encrypted clustering result; and decrypting the encrypted clustering result, and matching a decryption result with the credit product to obtain an intelligent recommendation result of the credit product. According to the method, feature extraction is carried out on multi-source credit data, and fusion is carried out; and in combination with homomorphic encryption, the defects that the current credit product recommendation method is easy to leak privacy and cannot comprehensively and accurately evaluate the credit status of the customer are overcome.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a credit product intelligent recommendation method and system based on multi-source data fusion. Background Art

[0002] Traditional credit product recommendation methods are often based on a single data source or simple data integration, which makes it difficult to fully tap the rich information and complex relationships hidden behind multi-source data. On the one hand, the dispersion and heterogeneity of data make it a huge challenge to fully and accurately assess the credit status of customers, which can easily lead to credit decision-making errors, such as overestimating the repayment ability of customers and issuing high-value loans to high-risk customers, or underestimating the potential of high-quality customers and missing business expansion opportunities. On the other hand, in the process of data processing, the privacy protection of customer credit data is crucial. Since a large amount of sensitive personal information is involved, once a data leak occurs, it will not only damage the rights and interests of customers, but also expose financial institutions to serious legal risks and reputation losses. Summary of the invention

[0003] The main purpose of the present invention is to provide a credit product intelligent recommendation method and system based on multi-source data fusion, aiming to overcome the defects of the current credit product recommendation method that is easy to leak privacy and cannot comprehensively and accurately assess the customer's credit status.

[0004] To achieve the above object, the present invention provides a credit product intelligent recommendation method based on multi-source data fusion, comprising the following steps: Extracting features from multi-source credit data to obtain a feature vector set; fusing data on each feature vector in the feature vector set to obtain fused feature data; Performing homomorphic encryption on the fused feature data to obtain homomorphic encrypted data, and storing the data in each security node; Based on each of the security nodes performing similarity calculation and clustering operation in the ciphertext space through a secret sharing protocol, an encrypted clustering result is obtained; The encrypted clustering result is decrypted, and the decrypted result is matched with the credit product to obtain the credit product intelligent recommendation result.

[0005] Furthermore, the multi-source credit data includes basic customer information, credit history data, financial and asset status data, consumer behavior data, and occupation-related data.

[0006] Furthermore, based on each of the security nodes, similarity calculation and clustering operation are performed in the ciphertext space through a secret sharing protocol to obtain encrypted clustering results, including: Each of the security nodes accesses the ciphertext space based on a secret sharing protocol, maps the homomorphically encrypted data to feature spaces of different scales, calculates the similarity between feature vectors in the feature space of each scale, and weightedly fuses the similarity results at different scales to obtain a comprehensive similarity matrix; A clustering algorithm based on density peak is used to cluster the comprehensive similarity matrix to obtain an encrypted clustering result. In the clustering process, the local density of each data point and the encrypted distance to the high-density point are calculated using the characteristics of homomorphic encryption. The clustering center is determined according to the preset density threshold and distance threshold, and the homomorphic encrypted data is divided into corresponding clustering clusters to obtain an encrypted clustering result.

[0007] Further, performing homomorphic encryption on the fused feature data to obtain homomorphic encrypted data includes: Collect node information of each security node respectively, wherein the node information includes hardware identification information, network environment feature information, and preset random seed information; For each security node, generate a subkey based on the corresponding node information; A multi-party computing collaborative mechanism is used to merge the sub-keys to generate a master key; The fused feature data is homomorphically encrypted based on the master key to obtain homomorphically encrypted data.

[0008] Furthermore, for each security node, a subkey is generated based on the corresponding node information, including: For each security node, the hardware identification information is hash-encoded to obtain a hardware hash value; Extracting the network delay fluctuation value, bandwidth peak value and mean value from the network environment feature information, and combining them with the hardware hash value and the preset random seed information to form a node feature string; The node feature string is input into a key generator based on the principle of quantum encryption. The key generator uses the superposition and entanglement of quantum states to perform encryption transformation on the node feature string to generate an initial encryption subkey. The initial encryption subkey is embedded in the lattice space through a pre-constructed lattice structure, and multiple random walks and encryption perturbations are performed in the lattice space to generate the subkey.

[0009] Furthermore, for each security node, a subkey is generated based on the corresponding node information, including: For each security node, the hardware identification information is segmented and encoded and converted into a multi-dimensional vector; Performing spectrum analysis on the network environment characteristic information, extracting the main frequency characteristics and mapping them into encryption vectors, and generating the preset random seed information into interference vectors; The multidimensional vector, encryption vector and interference vector are used as pheromones and heuristic factors in the ant colony algorithm to perform path search and combinatorial optimization in the high-dimensional vector space. During the search process, the number of iterations and parameters of the algorithm are dynamically adjusted according to the complexity of the node information, and the sub-key of the initial form is gradually converged; The initial subkey is embedded into the lattice structure, and after multiple rounds of lattice operations and verifications, the subkey corresponding to each security node is finally determined.

[0010] The present invention also provides a credit product intelligent recommendation system based on multi-source data fusion, comprising: A fusion unit is used to extract features from multi-source credit data to obtain a feature vector set; and to fuse the feature vectors in the feature vector set to obtain fused feature data; An encryption unit, used for performing homomorphic encryption on the fused feature data to obtain homomorphic encrypted data, and storing the data in each security node; A clustering unit, used to perform similarity calculation and clustering operation in the ciphertext space based on each of the security nodes through a secret sharing protocol to obtain an encrypted clustering result; The recommendation unit is used to decrypt the encrypted clustering result, match the decrypted result with the credit product, and obtain the credit product intelligent recommendation result.

[0011] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0012] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above-mentioned methods are implemented.

[0013] The present invention provides a method and system for intelligent recommendation of credit products based on multi-source data fusion, including: extracting features from multi-source credit data to obtain a feature vector set; fusing each feature vector in the feature vector set to obtain fused feature data; homomorphically encrypting the fused feature data to obtain homomorphically encrypted data, and storing it in each security node; performing similarity calculation and clustering operations in the ciphertext space based on each of the security nodes through a secret sharing protocol to obtain an encrypted clustering result; decrypting the encrypted clustering result, matching the decrypted result with the credit product, and obtaining an intelligent recommendation result of the credit product. In the present invention, feature extraction and fusion of multi-source credit data are performed; combined with homomorphic encryption, similarity calculation and clustering operations are performed in the ciphertext space through a secret sharing protocol. Not only is the privacy security of customer credit data guaranteed, but credit assessment can also be performed comprehensively, overcoming the defects of the current credit product recommendation method that is easy to leak privacy and cannot comprehensively and accurately assess the customer's credit status. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a schematic diagram of the steps of a credit product intelligent recommendation method based on multi-source data fusion in one embodiment of the present invention; Figure 2 It is a structural block diagram of a credit product intelligent recommendation system based on multi-source data fusion in one embodiment of the present invention; Figure 3 It is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0015] The implementation, functional features and advantages of the present invention will be further described in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0017] Reference Figure 1 In one embodiment of the present invention, a credit product intelligent recommendation method based on multi-source data fusion is provided, comprising the following steps: Step S1, extracting features from multi-source credit data to obtain a feature vector set; fusing data on each feature vector in the feature vector set to obtain fused feature data; Step S2, homomorphically encrypting the fused feature data to obtain homomorphically encrypted data, and storing the data in each security node; Step S3, performing similarity calculation and clustering operation in the ciphertext space based on each of the security nodes through a secret sharing protocol to obtain an encrypted clustering result; Step S4, decrypting the encrypted clustering result, matching the decrypted result with the credit product, and obtaining the credit product intelligent recommendation result.

[0018] In this embodiment, as described in step S1 above, the multi-source credit data comes from a wide range of sources and in various formats, such as customer account transaction flow data within the bank, credit report data obtained from credit reporting agencies, and consumption data provided by e-commerce platforms. In the feature extraction process, different extraction strategies are used for data from different sources. For transaction flow data, features such as average monthly transaction amount, transaction frequency, maximum transaction amount, and capital inflow and outflow ratio can be extracted; features such as the number of overdue payments, overdue duration, and credit limit usage ratio can be extracted from credit report data; and features such as consumer category distribution, average consumption amount, and consumption time period can be extracted from consumption data. The representative and potentially relevant features extracted from various data sources are summarized to form a feature vector set. Each feature vector represents the multi-dimensional feature information of a customer, so that subsequent comprehensive analysis and processing can be carried out to lay the foundation for accurate credit product recommendations.

[0019] Since different feature vectors may have different dimensions, data distribution ranges, and semantic meanings, appropriate normalization or standardization methods need to be used in the data fusion stage to map each feature vector to a unified numerical range or data space. For example, min-max normalization can be used to scale the data to the [0, 1] interval, or z-score normalization can be used to make the data conform to the standard normal distribution. At the same time, considering that different features may have different importance for credit assessment, a feature weight allocation mechanism can be introduced to determine the weight coefficient of each feature through data analysis or machine learning algorithms, and perform weighted summation or other fusion operations on each feature vector according to its weight to obtain fused feature data. The above fused feature data can integrate information from multiple source data to more comprehensively and accurately reflect the customer's credit-related status.

[0020] As in step S2 above, homomorphic encryption technology allows mathematical operations to be performed directly on encrypted data without decryption, which makes it possible to perform subsequent ciphertext data processing while ensuring data privacy. In this step, the fused feature data is encrypted using a homomorphic encryption algorithm. First, a key pair required for encryption is generated, including a public key and a private key. The fused feature data is encrypted using the public key to convert the original fused feature data into homomorphically encrypted data. The encryption process follows the mathematical rules of the homomorphic encryption algorithm, so that the encrypted data can perform subsequent designed operations in the ciphertext space, such as homomorphic operations of addition or multiplication, etc. The specific homomorphic properties depend on the type of homomorphic encryption algorithm selected (such as partial homomorphic encryption, full homomorphic encryption, etc.).

[0021] The generated homomorphic encrypted data is stored in each security node. The above security node can be a dedicated data storage server cluster built within a financial institution, or a secure storage service node based on a cloud service provider, and is protected by strict access control and network security mechanisms. When storing, it is allocated according to a pre-designed storage strategy, such as determining the amount of data and data blocks stored in each node based on factors such as node load balancing, geographical distribution, and data redundancy requirements. In this way, even if a node fails or is attacked, the data integrity and availability can be restored and guaranteed through data from other nodes. At the same time, decentralized storage also reduces the risk of data leakage.

[0022] As described in step S3 above, each security node interacts and collaborates through a secret sharing protocol. In similarity calculations, such as calculating the distance or similarity index (such as Euclidean distance, cosine similarity, etc.) between customer feature vectors, nodes exchange partial information of encrypted data based on the secret sharing protocol and use the operational properties of homomorphic encryption to perform calculations in the ciphertext space.

[0023] After completing the similarity calculation in the ciphertext space, clustering operations are performed based on the encrypted similarity information obtained. A clustering algorithm suitable for ciphertext data processing is used, such as a density-based clustering algorithm or an encrypted version of a hierarchical clustering algorithm. Taking the density-based clustering algorithm as an example, in the ciphertext space, the local density of the data point is determined according to the encrypted similarity, and a set of density-connected data points is identified as a clustering cluster. Since the data is in an encrypted state, the calculations in the clustering process are completed through homomorphic encryption operations and secret sharing protocols, such as calculating the encrypted density value of the data point, determining the cluster center, and other operations. Finally, the encrypted clustering results are obtained, that is, the classification of the customer group in the ciphertext space. Each cluster represents a group of customers with similar credit characteristics, but the data is always in an encrypted protection state throughout the process.

[0024] As in step S4 above, the encrypted clustering results are decrypted using the private key corresponding to the homomorphic encryption process. The decryption process converts the clustering results in the ciphertext state into plaintext data for subsequent credit product matching. The decryption operation needs to be performed in a strict security environment and authorization mechanism to prevent private key leakage from causing data security issues.

[0025] The decrypted clustering results are matched with various credit products provided by financial institutions. According to the characteristics of each cluster, such as the average credit level, repayment ability, risk preference and other factors of the customer, the application conditions, interest rate level, credit limit range, repayment period and other details of the credit product are matched. For example, for cluster customers with good credit status and strong repayment ability, high-quality credit products with higher credit limits, lower interest rates and longer terms can be recommended; while for customer groups with average credit records and relatively weak repayment ability, credit products with smaller credit limits and controllable risks are matched. Through this precise matching, the success rate of credit product recommendations and customer satisfaction are improved, while the credit risk of financial institutions is reduced, and the credit business process and resource allocation are optimized.

[0026] In one embodiment, the multi-source credit data includes basic customer information, credit history data, financial and asset status data, consumer behavior data, and career-related data.

[0027] In one embodiment, based on each of the security nodes performing similarity calculation and clustering operation in the ciphertext space through a secret sharing protocol, an encrypted clustering result is obtained, including: Each of the security nodes accesses the ciphertext space based on a secret sharing protocol, maps the homomorphically encrypted data to feature spaces of different scales, calculates the similarity between feature vectors in the feature space of each scale, and weightedly fuses the similarity results at different scales to obtain a comprehensive similarity matrix; A clustering algorithm based on density peak is used to cluster the comprehensive similarity matrix to obtain an encrypted clustering result. In the clustering process, the local density of each data point and the encrypted distance to the high-density point are calculated using the characteristics of homomorphic encryption. The clustering center is determined according to the preset density threshold and distance threshold, and the homomorphic encrypted data is divided into corresponding clustering clusters to obtain an encrypted clustering result.

[0028] In this embodiment, each security node is a key unit for storing homomorphically encrypted data. They use a secret sharing protocol to achieve collaborative operation and access to the ciphertext space. The secret sharing protocol aims to split and share sensitive information (here, homomorphically encrypted data) among multiple nodes in a distributed and secure manner, so that no single node can obtain the complete picture of the data alone, thereby ensuring the privacy and security of the data.

[0029] Homomorphic encrypted data itself contains a wealth of customer credit-related features, but in order to more comprehensively and meticulously mine the similarity information hidden in the data, it needs to be mapped to feature spaces of different scales. Different scales can be understood as observing and analyzing data features from different granularities and angles. For example, from a macroscopic scale, the focus is on large-dimensional features such as the overall asset size of the customer and the length of credit history; while at a microscopic scale, the focus is on detailed features such as the customer's recent specific consumption frequency and the regularity of small transactions. Through specific mathematical transformations or feature extraction methods (such as wavelet transforms, multi-resolution analysis and other technologies), homomorphic encrypted data is re-represented at different scales, so that it can present a unique feature structure in the feature space of each scale, which is convenient for subsequent targeted calculation of the similarity between feature vectors and mining the correlation between customer data at different levels.

[0030] In each mapped scale feature space, the similarity between the feature vectors corresponding to different customers should be measured. Commonly used similarity metrics include Euclidean distance, cosine similarity, Manhattan distance, etc. The appropriate method can be selected according to the characteristics of credit data and the feasibility of homomorphic encryption. Taking cosine similarity as an example, for the feature vectors of two customers in a certain scale feature space, the cosine value of their angle is calculated through the operation method supported by homomorphic encryption to characterize the similarity (the closer to 1, the more similar, and the closer to 0, the greater the difference). Since the operation is performed in the ciphertext space, the entire calculation process is directly carried out based on homomorphic encrypted data. The characteristic of homomorphic encryption that allows specific mathematical operations on encrypted data is used to ensure that data privacy is not leaked. Each security node exchanges the necessary data shares according to the secret sharing protocol to jointly complete these similarity calculations. A set of similarity results between feature vectors can be obtained at each scale.

[0031] The similarity information reflected by feature spaces of different scales has different contribution weights to the final judgment of the overall similarity between customers. For example, the similarity calculated by the feature space at the macro scale is more influential in judging the overall risk category of customers, while the similarity at the micro scale is more valuable in analyzing the similarity of customers' recent behavior patterns. Therefore, it is necessary to assign corresponding weights to the similarity results at different scales, which can be determined by historical data verification or machine learning algorithms. Then, through fusion methods such as weighted summation, the similarity results at each scale are integrated together to form a comprehensive similarity matrix. The above comprehensive similarity matrix comprehensively considers the similarity relationship between customers from multiple dimensions, providing a comprehensive basis for subsequent clustering operations.

[0032] The clustering algorithm based on density peak is a method for identifying clusters based on the density distribution of data points. Its core idea is that the density around the data points at the center of the cluster is relatively high, and there is a relatively large distance interval with other high-density areas. In this embodiment, for the encrypted data represented by the comprehensive similarity matrix, each data point (corresponding to the encrypted feature vector of a customer) has its "position" in the ciphertext space (determined by the comprehensive similarity relationship). First, the local density of each data point must be calculated. The local density reflects the density of the data points around the data point. In the ciphertext space, the characteristics of homomorphic encryption are used to measure the local density by counting the number of data points similar to the data point within a certain range (the neighborhood range set by the algorithm) or the cumulative value of similarity through specific encryption operations. Similarly, the encryption distance between each data point and other high-density points (data points with relatively high density that have been preliminarily identified or determined through encryption comparison) is calculated.

[0033] The cluster center is determined based on the preset density threshold and distance threshold. The density threshold is used to screen out data points with sufficiently high local density. Only data points with local density exceeding the threshold have the possibility of becoming cluster centers; the distance threshold is used to measure the interval between data points. If the encrypted distance between a high-density point and other high-density points is greater than the distance threshold, then it is more likely to be an independent cluster center. Once the cluster center is determined, the homomorphically encrypted data can be divided into corresponding cluster clusters based on the encrypted similarity relationship between the data point and the cluster center and the encrypted connectivity between the data points. For example, if a data point has a high encrypted similarity with a cluster center and it is judged by the encrypted distance that it is more inclined to belong to the group represented by the cluster center, then it is divided into the corresponding cluster cluster, and finally the encrypted clustering result is obtained, that is, the classification of customer groups based on the similarity of credit characteristics is completed in the ciphertext space, and further decryption and credit product matching operations can be performed based on this.

[0034] In one embodiment, performing homomorphic encryption on the fused feature data to obtain homomorphic encrypted data includes: Collect node information of each security node respectively, wherein the node information includes hardware identification information, network environment feature information, and preset random seed information; For each security node, generate a subkey based on the corresponding node information; A multi-party computing collaborative mechanism is used to merge the sub-keys to generate a master key; The fused feature data is homomorphically encrypted based on the master key to obtain homomorphically encrypted data.

[0035] In this embodiment, the hardware identification information is a physical feature identification used to uniquely distinguish each security node. It may include content such as the server's network card MAC address, hard disk serial number, motherboard unique identification code, etc. The above hardware-related identification is like the "identity card" of each node, which is unique and relatively fixed, and provides a basic basis for distinguishing operations such as generating keys based on individual node characteristics.

[0036] The network environment feature information reflects the relevant conditions of the network environment in which each security node is located. For example, it includes the IP address segment of the network where the node is located, the size of the network bandwidth, the network delay, the network topology information, and the type of network equipment connected. The above network-related characteristics have an important impact on subsequent encryption and data interaction operations. Different network bandwidths and delays may affect the efficiency and stability of data transmission, while factors such as network topology and the type of connected network equipment are related to the security of the network and the path and accessibility of data transmission between nodes.

[0037] The preset random seed information is a set of data elements that are pre-set and random. It plays a key role in introducing additional randomness in the entire encryption key generation process. Due to the randomness of the random seed information itself, it can add uncertainty and unpredictability to the subsequently generated keys, thereby improving the security of the keys. For example, it can be a sequence of random numbers generated according to a specific algorithm or a random byte stream obtained based on a reliable random number generator. Combining it with hardware identification information and network environment feature information can enable the final generated key to incorporate more uncontrollable and difficult-to-crack random factors based on the characteristics of the node itself, greatly enhancing the confidentiality of the encryption system.

[0038] For the hardware identification information, network environment feature information and preset random seed information of each security node collected, it is first necessary to perform a series of processing and conversion operations on this information so that it can adapt to the subsequent key generation algorithm requirements.

[0039] After completing the information processing and conversion, the key generation algorithm is used to generate sub-keys based on the integrated node information. The above key generation algorithm is usually constructed based on the security principles in cryptography. For example, some derivative algorithms in symmetric encryption algorithms or related principles of asymmetric encryption algorithms based on mathematical problems (such as discrete logarithm problems, elliptic curve discrete logarithm problems, etc.) can be used. Taking the algorithm based on elliptic curves as an example, the processed node information is used as the input parameter of the algorithm. After a series of complex mathematical operations such as elliptic curve point multiplication and key expansion, a sub-key that meets the security requirements of cryptography is finally generated. Each security node will generate a unique sub-key corresponding to its own unique node information. The sub-key has a certain encryption capability and is closely related to the hardware and network environment of the corresponding node, laying the foundation for the subsequent multi-party computing to collaboratively generate the master key.

[0040] The multi-party computing collaborative mechanism aims to allow multiple participants (here, each security node) to jointly complete a computing task, that is, to generate a master key, through specific interactive protocols and computing rules without disclosing their own private information (that is, the sub-keys generated by each party). In this process, each security node participates in the interactive process by encrypting or hiding key information in accordance with the predetermined multi-party computing protocol. For example, in the multi-party computing method using the secret sharing protocol, each node will split its sub-key into multiple shares, and then share these shares with other nodes. No single node can obtain the complete sub-key information of other nodes from the interactive process, but through the combination and calculation of multiple nodes according to the protocol rules, the key information of the sub-keys of all parties can be finally gathered to achieve fusion processing.

[0041] Under the multi-party computing collaborative mechanism, the sub-keys generated by each security node are gradually integrated through interaction and calculation steps. For example, the nodes first exchange the sub-key shares processed by homomorphic encryption, and then use the characteristics of homomorphic encryption to perform addition or multiplication in the ciphertext state to integrate the key encryption information contained in each sub-key share. After multiple rounds of such interactions and calculations, a unified master key is finally generated. The above master key integrates the sub-key characteristics of each security node and has stronger encryption capabilities and higher security. It is no longer limited to the attributes of a single node, but guarantees the reliability of encrypting the fused feature data from the entire system level. At the same time, it also ensures that no single node can control the master key alone, avoiding the risk of key leakage due to the security problems of a single node.

[0042] After obtaining the master key, you need to select a suitable homomorphic encryption algorithm to encrypt the fused feature data. There are many types of homomorphic encryption algorithms, such as partial homomorphic encryption (for example, supporting a limited number of addition or multiplication homomorphic operations), hierarchical fully homomorphic encryption, and complete fully homomorphic encryption. You need to weigh and select based on specific business needs and multiple factors such as computing resources and encryption efficiency.

[0043] The fused feature data is encrypted using the configured homomorphic encryption algorithm with the master key as the core encryption element. Specifically, the elements of the fused feature data are subjected to corresponding mathematical operations with the master key according to the operation rules of the homomorphic encryption algorithm. For example, in some homomorphic encryption algorithms, the master key may be used to perform multiple exponential operations, multiplication operations on the feature data and combine noise addition operations to convert the original fused feature data into homomorphic encrypted data.

[0044] In one embodiment, for each security node, a subkey is generated based on the corresponding node information, including: For each security node, the hardware identification information is hash-encoded to obtain a hardware hash value; Extracting the network delay fluctuation value, bandwidth peak value and mean value from the network environment feature information, and combining them with the hardware hash value and the preset random seed information to form a node feature string; The node feature string is input into a key generator based on the principle of quantum encryption. The key generator uses the superposition and entanglement of quantum states to perform encryption transformation on the node feature string to generate an initial encryption subkey. The initial encryption subkey is embedded in the lattice space through a pre-constructed lattice structure, and multiple random walks and encryption perturbations are performed in the lattice space to generate the subkey.

[0045] In this embodiment, hash coding plays an important role in the field of cryptography and data processing. Its main purpose is to convert data of any length (here, hardware identification information) into a fixed-length, unique (ideally, different inputs produce different outputs) and irreversible (it is difficult to infer the original data from the hash value) hash value. Hardware identification information, such as the server's network card MAC address, hard disk serial number, etc., is processed through a hash function (such as common hash algorithms such as SHA-256 and MD5). The hash function performs a series of complex mathematical operations on the binary data of the hardware identification information. These operations are based on specific algorithm logic, compressing and obfuscating the input data, and finally outputting a fixed-length hash value. For example, the SHA-256 algorithm outputs a 256-bit hash value. On the one hand, it simplifies the data representation form, which is convenient for subsequent processing in a unified format; on the other hand, it greatly enhances the confidentiality of the hardware identification information. Even if the hardware identification information itself may be obtained externally to a certain extent, after hash coding, it is difficult for others to restore the original hardware identification from the hash value, thereby ensuring the security of the key generation link based on hardware characteristics.

[0046] The network environment characteristic information covers many aspects, but in this scheme, the network delay fluctuation value, bandwidth peak and mean are specially selected as the key indicators because they have a direct and important impact on the transmission, interaction and subsequent encryption operations of data between security nodes. The network delay fluctuation value reflects the stability of the network. Large fluctuations may mean that the network environment is unstable, which will affect the timeliness and accuracy of the encrypted data interaction, calculation and other operations between nodes; the bandwidth peak and mean are related to the rate and capacity of data transmission. For processing data such as credit data, which may be large in volume, the bandwidth situation determines whether the data can flow between nodes quickly and efficiently and whether it can meet the data transmission requirements during encryption and decryption operations. Extracting these characteristic values ​​can accurately grasp the key performance status of the network environment of each security node, and provide an important reference for generating subkeys that fit the actual network conditions.

[0047] The extracted network delay fluctuation value, bandwidth peak and average are combined with the obtained hardware hash value and the preset random seed information in order to integrate information of different dimensions and reflecting different characteristics to form a node feature string that comprehensively reflects the overall characteristics of the security node.

[0048] Quantum encryption is based on the basic principles of quantum mechanics, such as the superposition and entanglement of quantum states, which provides unprecedented high security for information encryption. The superposition of quantum states means that a quantum bit (qubit) can be in a superposition of multiple states at the same time. Compared with classical bits that can only represent two states, 0 or 1, the amount of information it carries is richer and more uncertain; quantum entanglement is reflected in the existence of a special correlation between multiple quantum bits. The measurement of one quantum bit will instantly affect the state of other quantum bits entangled with it. This feature enables quantum encryption to detect any eavesdropping or interference during information transmission and processing, because once there is external intervention to try to obtain information, the quantum state will undergo an irreversible change, thereby ensuring the security of encrypted information.

[0049] In this scheme, the node feature string containing hardware, network and random factors is input into the key generator based on the principle of quantum encryption, and the key generator will use these unique properties of quantum states to perform encryption transformations on it. For example, the data information in the node feature string may be encoded into the state of the quantum bit first, and then the superposition of the quantum state is used to perform complex transformations on the encoded quantum bits with the help of quantum gate operations (similar to the logic gates in classical circuits, but specific quantum state transformation operations are performed on quantum bits), so that the information it carries is fully obfuscated and encrypted. At the same time, by preparing entangled quantum bit pairs and entangled with the quantum bits related to the node feature string, the security and complexity of encryption are further enhanced. After this series of encryption transformation operations based on quantum state superposition and entanglement, the initial encryption subkey is finally generated. This initial encryption subkey already has the high security characteristics given by quantum encryption, making it more difficult to crack and steal than the keys generated by traditional methods.

[0050] Lattice structure is an important mathematical construction in modern cryptography. It consists of a set of discrete points generated by a set of linearly independent vectors (called lattice basis), and has unique advantages in cryptographic applications. Lattice-based cryptographic systems often have the characteristics of being resistant to quantum attacks and security that can be reduced to classical mathematical problems (such as the shortest vector problem and the nearest vector problem), making them a powerful tool for building high-security encryption schemes. In this scheme, the lattice structure is pre-built to further enhance the security of the subkeys, and the special properties of the lattice space are used to deeply process the initial encryption subkeys that already have quantum encryption characteristics.

[0051] Embedding the initial encryption subkey into the lattice space is to map and place the information contained in the initial encryption subkey in a way that conforms to the lattice structure rules. For example, the initial encryption subkey can be represented by a corresponding position in the lattice space through a specific lattice basis and coordinate transformation. Afterwards, multiple random walks and encryption perturbation operations are performed in the lattice space. Random walk means moving the lattice points representing the initial encryption sub-key along a certain random path in the lattice space. This random movement will change their positional relationship in the lattice space, and thus change the form of information they represent. Encryption perturbation is based on random walks and is further encrypted through specific encryption operations based on the mathematical rules of the lattice space (such as adding lattice vectors, transforming lattice point coordinates, etc., and combining elements such as cryptographically secure random numbers) to make its information more complex and chaotic. After multiple such random walks and encryption perturbations, the sub-key finally generated inherits the high security of quantum encryption and incorporates the security characteristics of lattice cryptography. It has extremely high confidentiality, anti-attack and unpredictability, and can provide solid security guarantees for the subsequent multi-party computing to collaboratively generate master keys and the entire credit data encryption processing process.

[0052] In one embodiment, for each security node, a subkey is generated based on the corresponding node information, including: For each security node, the hardware identification information is segmented and encoded and converted into a multi-dimensional vector; Performing spectrum analysis on the network environment characteristic information, extracting the main frequency characteristics and mapping them into encryption vectors, and generating the preset random seed information into interference vectors; The multidimensional vector, encryption vector and interference vector are used as pheromones and heuristic factors in the ant colony algorithm to perform path search and combinatorial optimization in the high-dimensional vector space. During the search process, the number of iterations and parameters of the algorithm are dynamically adjusted according to the complexity of the node information, and the sub-key of the initial form is gradually converged; The initial subkey is embedded into the lattice structure, and after multiple rounds of lattice operations and verifications, the subkey corresponding to each security node is finally determined.

[0053] In this embodiment, the hardware identification information usually includes data content with a certain format and length, such as the server network card MAC address, hard disk serial number, etc. The reason for segmenting and encoding it is that it may be difficult to fully explore its inherent characteristics if the information is processed as a whole, and it is not conducive to subsequent fusion and unified operation with other types of information. By segmenting according to specific rules, for example, according to the semantics of different components of the hardware identification information, byte length and other factors, the entire hardware identification information is divided into multiple small segments. Then, for each small segment, a suitable encoding method (such as binary encoding, decimal encoding or custom mapping encoding rules, etc.) is used to convert it into a digital form, so that the digital forms corresponding to multiple small segments are combined to form a multidimensional vector. For example, if the hardware identification information is evenly divided into 5 segments, each segment is represented by a number after encoding, and finally a five-dimensional vector can be obtained. The values ​​on different dimensions reflect the characteristic information of different parts of the hardware identification, thereby presenting the hardware identification information in a structured and easy-to-process manner.

[0054] Converting hardware identification information into multi-dimensional vectors enables hardware-related features to be quantified in vector space, which is consistent with the overall idea of ​​subsequent operations and optimizations in vector space based on algorithms. Multi-dimensional vectors can more comprehensively reflect the detailed features of hardware identification, and are convenient for mathematical operations and combination operations with other network environment feature information and random seed information that are also represented in vector form, making it easier to implement the sub-key generation process under a unified mathematical framework. At the same time, the vector form also helps to use some mature vector space-related algorithms and theories to mine potential relationships in data, enhance the scientificity and rationality of the sub-key generation process, and improve the security and reliability of the key.

[0055] Spectrum analysis is a commonly used signal processing technology. It converts time domain signals into frequency domain representations based on mathematical methods such as Fourier transform, thereby revealing the distribution of different frequency components in the signal. For network environment feature information, although it is not a time domain signal in the traditional sense, it can be regarded as a data sequence that changes over time or other related dimensions. For example, the fluctuation of network delay over time and the change of bandwidth in different time periods can be regarded as a "signal". By performing spectrum analysis on these network environment feature information, the dominant frequency features can be extracted, that is, those frequency components that have a key impact on network performance, data transmission, etc. These main frequency features often contain deep information such as network environment stability and periodic changes, reflecting the essential characteristics of the network environment.

[0056] After extracting the main frequency features, they are mapped to encryption vectors according to the rules. This process involves quantization, normalization, and encoding of frequency features to convert them into vector forms suitable for subsequent algorithm processing. In this process, some encryption-related processing can be integrated, such as transforming vector elements through specific encryption functions, so that the final encryption vector contains both key feature information of the network environment and has a certain degree of confidentiality. At the same time, for the preset random seed information, it is generated as an interference vector. For example, the random seed information can be used to generate a set of random numerical sequences through a pseudo-random number generation algorithm, and organized in the form of a vector. The role of the interference vector is to introduce additional random factors for subsequent algorithm operations, disrupt and confuse the information carried by other vectors, and prevent the key information from being inferred by analyzing the regular relationship between vectors, thereby enhancing the unpredictability and security of the sub-key generation process.

[0057] The ant colony algorithm is an optimization algorithm that simulates the foraging behavior of ants. Its core mechanism is that ants leave pheromones on the path they pass through when searching for food. The ants will then choose the path to walk based on the concentration of pheromones on the path and some heuristic information. After multiple iterations, the entire ant colony will find an optimal path from the ant nest to the food source. In this scheme, multidimensional vectors, encrypted vectors, and interference vectors are used as pheromones and heuristic factors in the ant colony algorithm respectively. The multidimensional vector represents the characteristic information of the hardware identification, which can be regarded as a relatively stable and guiding information based on the hardware itself, similar to the heuristic information in the ant colony algorithm, which helps guide the search direction; the encrypted vector carries the key characteristics of the network environment and is encrypted. Its element value and other information can be compared to pheromones, reflecting the "path quality" related to the network environment; the interference vector continuously introduces randomness, affecting the ants (reflected as the search process in the algorithm) in choosing the path, making the search more diverse and unpredictable.

[0058] Path search and combination optimization in high-dimensional vector space means that the virtual "ants" in the algorithm constantly try different vector combination paths based on pheromone concentration and heuristic information in this high-dimensional space composed of multi-dimensional vectors, encryption vectors and interference vectors, and arrange, calculate, and fuse each vector element to find an optimal combination method. This process is to explore how to generate the safest and most suitable sub-key form. Dynamically adjusting the number of iterations and parameters of the algorithm according to the complexity of the node information is a key adaptive mechanism. Because the hardware identification information, network environment feature information, etc. of different security nodes are different, their complexity is also different. Simple node information may find a better initial form of sub-keys with fewer iterations and conventional parameters, while complex node information requires more iterations and fine-tuning parameters to fully explore the relationship between each vector and optimize the combination method, ensuring that the sub-key in the initial form obtained by the final convergence can fully consider the specific characteristics of the node and achieve higher security and adaptability requirements.

[0059] Lattice structures have unique security advantages in the field of cryptography. Lattice-based cryptographic schemes can often resist a variety of attack methods, especially in dealing with potential threats brought by new computing capabilities such as quantum computing. To embed the initial form of the subkey into the lattice structure, it is first necessary to determine the appropriate lattice basis and the corresponding embedding rules. For example, through coordinate transformation, vector mapping, etc., the vector corresponding to the initial form of the subkey is accurately placed in the lattice space, making it a part of the lattice structure, so that it can use the characteristics of the lattice structure to further enhance security.

[0060] Perform multiple rounds of lattice operations, such as vector addition and multiplication in the lattice space, which comply with the lattice operation rules, and continuously transform and adjust the embedded initial form subkey, making its information more complex and difficult to crack. At the same time, verification is performed after each round of operation. The verification process can be based on the mathematical properties related to the lattice structure, such as checking whether the vector after the operation meets the specific lattice space constraints and whether it is consistent with the preset security standards (such as the shortest vector length, the distance between vectors, etc.). Through multiple rounds of such lattice operations and verifications, the potential security risks of the subkey are gradually eliminated, so that it is continuously optimized and finally meets the requirements of high security, and a reliable subkey corresponding to each security node that can be used for subsequent multi-party computing to collaboratively generate the master key and other operations is determined, so as to comprehensively ensure the security of the entire credit data encryption processing process.

[0061] In one embodiment, for each security node, a subkey is generated based on the corresponding node information, including: The hardware identification information of each security node is converted into a binary sequence, and the distribution ratio of 0 and 1 is counted and recorded as the identification ratio value.

[0062] For the network environment characteristic information, a network characteristic matrix is ​​constructed, its eigenvalues ​​and eigenvectors are calculated, and the eigenvector corresponding to the maximum eigenvalue is selected as the main eigenvector of the network.

[0063] The preset random seed information is combined with the identification ratio value and the network main eigenvector to construct a three-dimensional data point set.

[0064] Based on the three-dimensional data point set, a space curve is generated by using the cubic spline interpolation method. The space curve is subjected to Fourier series expansion, and the first N coefficients are extracted to construct a coefficient vector.

[0065] The coefficient vector is input into an encryption function based on chaotic mapping. The chaotic mapping function encrypts the coefficient vector multiple times with preset initial parameters and iteration rules. During each iteration, the encryption parameters are fine-tuned according to the system timestamp of the current node as a dynamic adjustment factor, and a sub-key is finally generated. The length of the sub-key is associated with the dimension of the coefficient vector. The encrypted coefficient vector is converted into a sub-key format through encoding rules.

[0066] In this embodiment, the hardware identification information of each security node is converted into a binary sequence in order to convert the original information of the hardware identification with a certain format and semantics into a data form that can be processed by the computer in basic operation units (binary bits). Hardware identification information, such as the MAC address of the server's network card and the hard disk serial number, can be subsequently mined based on binary 0 and 1 after conversion. The distribution ratio of 0 and 1 is counted and recorded as the identification ratio value. This ratio value can summarize the inherent characteristics of the hardware identification information from a macro perspective. It reflects a digital feature law of the binary sequence in the overall composition. Different hardware identifications often have different distributions of 0 and 1 in the binary sequence after conversion.

[0067] The network environment characteristic information covers many aspects of data content such as network bandwidth, network delay, packet loss rate, etc. These data information of different dimensions are sorted and organized to construct a network characteristic matrix. The purpose is to integrate the scattered network environment data into a unified mathematical structure so as to use the matrix-related mathematical analysis methods to explore its deep-level characteristics. By calculating the eigenvalues ​​and eigenvectors of the network characteristic matrix, the eigenvalue represents the degree of expansion and contraction of the matrix in the direction of the corresponding eigenvector, while the eigenvector describes the specific directional characteristics of the matrix. The eigenvector corresponding to the maximum eigenvalue is selected as the main eigenvector of the network because the eigenvector corresponding to the maximum eigenvalue often reflects the direction of the network environment characteristic information that is most dominant in the overall structure and best reflects its core characteristics. This vector carries the key characteristic information of the network environment. When combined with other information later, it can highlight the important influencing factors of the network environment and help generate sub-keys that fit the actual network situation.

[0068] The preset random seed information itself is a data element with randomness set by humans. Its introduction is to add additional unpredictability factors to the entire subkey generation process and improve the security of the key. Combining it with the identification ratio value based on hardware identification and the main network feature vector based on network environment characteristics to construct a three-dimensional data point set is to integrate the information with key characteristics from three different dimensions of hardware, network and random factors to form a data set that can comprehensively reflect the characteristics of all aspects of the security node. These data points in the three-dimensional space, each point contains comprehensive information of hardware, network and randomness, which lays the foundation for the subsequent generation of space curves through mathematical methods and further mining the hidden features that can be used to generate subkeys, so that the generated subkeys can fully consider the multivariate characteristics of the nodes.

[0069] Based on the three-dimensional data point set, a spatial curve is generated using the cubic spline interpolation method. The cubic spline interpolation method can construct a smooth and continuous spatial curve based on the given discrete three-dimensional data points, so that it passes through all the data points or is as close to these data points as possible. Such a spatial curve can not only fit the previously synthesized node information data points well, but also present the changing trend and internal connection of the node information in the three-dimensional space in a more continuous and smooth way compared to discrete data points, which is convenient for subsequent more in-depth mathematical analysis and feature extraction operations, and provides a more favorable curve morphology basis for mining more representative and unique key generation features.

[0070] The generated space curve is expanded by Fourier series. Fourier series expansion is a method to decompose a periodic function (the space curve here can be regarded as a function representation with similar periodic characteristics within a certain interval) into a series of linear combinations of sine and cosine functions. The coefficient vector is constructed by extracting the first N coefficients. These coefficients actually reflect the main characteristic information of the space curve in the frequency domain. Different coefficient values ​​represent the weight and contribution of the curve on different frequency components. The extracted coefficient vector condenses the node information characteristics of the space curve from the perspective of frequency domain, which are composed of hardware, network and random factors, and provides key characteristic input data for subsequent input into the encryption function to generate sub-keys.

[0071] Chaotic mapping is extremely sensitive to initial conditions, and its long-term behavior is highly random and unpredictable. The encryption function based on chaotic mapping can make full use of these characteristics to encrypt the input data (i.e., coefficient vector). The coefficient vector is encrypted multiple times with preset initial parameters and iteration rules. In each iteration, the encryption parameters are fine-tuned according to the system timestamp of the current node as a dynamic adjustment factor. The system timestamp is a unique data identifier that changes dynamically over time. Introducing it as a dynamic adjustment factor makes the encryption parameters change according to the current time situation at each iteration, further enhancing the randomness and unpredictability of the encryption process, avoiding the regularity of the encryption results, and making the generated subkeys extremely difficult to be cracked by external attackers through analysis of the rules, thereby improving the security of the subkeys.

[0072] The length of the subkey is associated with the dimension of the coefficient vector, which means that the dimension of the coefficient vector directly determines the length of the subkey. The above association method ensures that the subkey can make full use of the characteristic information carried by the coefficient vector, and because the coefficient vector dimension may be different at different nodes, the length of the generated subkey will also be different, which increases the diversity and complexity of the key. Finally, the encrypted coefficient vector is converted into a subkey format through specific encoding rules. The encoding rules will be designed according to factors such as cryptographic requirements and compatibility of the actual application of the system to ensure that the generated subkey can meet the subsequent use specifications in the multi-party computing collaborative generation of the master key and the entire credit data encryption processing process, and participate in the subsequent encryption operation in a reliable format to ensure data security.

[0073] In yet another embodiment, for each security node, generating a subkey based on the corresponding node information respectively includes: The hardware identification information of the security node is converted into a binary sequence, the network environment feature information is sampled and quantized to construct a multidimensional vector, and the binary sequence and the multidimensional vector are subjected to bitwise XOR operation and vector translation operation to obtain a composite data matrix.

[0074] Perform singular value decomposition on the composite data matrix to obtain the singular value sequence, and construct a scatter plot with the singular value sequence as the horizontal axis and the corresponding serial number as the vertical axis.

[0075] The scatter plot was curve fitted and a smooth fitting curve was obtained using piecewise cubic Hermite interpolation method.

[0076] The curvature function of the fitting curve is extracted, and the integral value of the curvature function in a specific interval is calculated. The integral value is normalized and used as the parameter input of the key generation function. The key generation function is constructed based on the chaotic mapping theory, and the subkey sequence is generated through iterative calculation. In the generation process, the key check bit is introduced. The check bit is determined by the hash value of the subkey sequence and the encrypted summary of the node information, so as to ensure the accuracy and integrity of the subkey.

[0077] In this embodiment, first, the hardware identification information of the security node is converted into a binary sequence, which enables the hardware identification to be presented in a binary form that is easy for the computer to process, and facilitates subsequent operations with other data. For network environment feature information, since it often has data of various types and magnitudes (such as network bandwidth, delay, etc.), it is regularized by sampling quantization and constructed into a multidimensional vector so that it can be analyzed in a unified vector space. Then, the binary sequence and the multidimensional vector are subjected to bitwise XOR operation and vector shift operation. The bitwise XOR operation can fuse and confuse the information of the two based on the logical operation of the binary bit, and the vector shift operation further changes the positional relationship of the vector. After these two operations, a composite data matrix integrating the information of the two is obtained.

[0078] Singular value decomposition (SVD) is an important matrix decomposition method. Singular value decomposition of a composite data matrix can be decomposed into three matrix products with specific mathematical properties. The extracted singular value sequence contains the main characteristic information of the composite data matrix and reflects the "importance" of the data in different dimensions. Afterwards, a scatter plot is constructed with the singular value sequence as the horizontal axis and its corresponding serial number as the vertical axis. Such a scatter plot can intuitively show the distribution of singular values, help observe the distribution law of data features and the correlation between different singular values, and provide a visual basis for subsequent operations such as curve fitting.

[0079] The piecewise cubic Hermite interpolation method is used to fit the scatter plot. This method can construct a smooth curve based on the given scattered data, making it as close as possible to the distribution trend of the scattered points. The fitting curve obtained in this way can not only better reflect the overall characteristic change trend contained in the singular value sequence, but also, compared with discrete scattered points, the smooth curve is easier to operate and process in subsequent mathematical analysis and calculations, and can more smoothly and continuously reflect the internal connection between data features, preparing for further extraction of key curve features.

[0080] The curvature function is extracted from the fitting curve. The curvature function describes the important geometric feature of the curvature degree of the curve at each point. The integral value of the curvature function in a specific interval is calculated. This integral value integrates the overall curvature degree of the curve in the interval, and reflects the key feature quantity contained in the original node information after a series of processing from another perspective. It is used as the parameter input of the subsequent key generation function, and the node information features mined through complex data processing and analysis are integrated into the key link of key generation.

[0081] The key generation function is constructed based on the theory of chaotic mapping. Chaotic mapping has the characteristics of being extremely sensitive to initial conditions and unpredictable in long-term behavior. It is used to iteratively calculate and generate a subkey sequence, so that the generated subkey has high randomness and unpredictability, which greatly enhances the security of the subkey. At the same time, a key check bit is introduced in the generation process. The check bit is determined by the hash value of the subkey sequence and the encrypted summary of the node information. The hash value can be used to verify the integrity and accuracy of the subkey itself to prevent errors or tampering during the generation or transmission process. The encrypted summary of the node information further associates the subkey with the original information of the corresponding node to ensure that the correspondence between the subkey and the node is accurate. In this way, the accuracy and integrity of the subkey are fully guaranteed, so that it can be reliably used in subsequent operations such as generating the master key.

[0082] Reference Figure 2 In one embodiment of the present invention, a credit product intelligent recommendation system based on multi-source data fusion is provided, including: A fusion unit is used to extract features from multi-source credit data to obtain a feature vector set; and to fuse the feature vectors in the feature vector set to obtain fused feature data; An encryption unit, used for performing homomorphic encryption on the fused feature data to obtain homomorphic encrypted data, and storing the data in each security node; A clustering unit, used to perform similarity calculation and clustering operation in the ciphertext space based on each of the security nodes through a secret sharing protocol to obtain an encrypted clustering result; The recommendation unit is used to decrypt the encrypted clustering result, match the decrypted result with the credit product, and obtain the credit product intelligent recommendation result.

[0083] In this embodiment, for the specific implementation of each unit in the above system embodiment, please refer to the above method embodiment, which will not be repeated here.

[0084] Reference Figure 3 In an embodiment of the present invention, a computer device is also provided. The computer device may be a server, and its internal structure may be as follows: Figure 3 As shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0085] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0086] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0087] In summary, the credit product intelligent recommendation method and system based on multi-source data fusion provided in the embodiment of the present invention include: extracting features from multi-source credit data to obtain a feature vector set; fusing each feature vector in the feature vector set to obtain fused feature data; homomorphically encrypting the fused feature data to obtain homomorphically encrypted data, and storing it in each security node; performing similarity calculation and clustering operations in the ciphertext space through a secret sharing protocol based on each of the security nodes to obtain encrypted clustering results; decrypting the encrypted clustering results, matching the decrypted results with the credit products, and obtaining credit product intelligent recommendation results. In the present invention, feature extraction and fusion of multi-source credit data are performed; combined with homomorphic encryption, similarity calculation and clustering operations are performed in the ciphertext space through a secret sharing protocol. Not only is the privacy security of customer credit data guaranteed, but also a comprehensive credit assessment can be performed, overcoming the defects of the current credit product recommendation method that is easy to leak privacy and cannot comprehensively and accurately assess the customer's credit status.

[0088] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided by the present invention and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.

[0089] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0090] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A credit product intelligent recommendation method based on multi-source data fusion, characterized in that: The following steps are involved: Extract features from multi-source credit data to obtain a feature vector set; Performing data fusion on each feature vector in the feature vector set to obtain fused feature data; Performing homomorphic encryption on the fused feature data to obtain homomorphic encrypted data, and storing the data in each security node; Based on each of the security nodes performing similarity calculation and clustering operation in the ciphertext space through a secret sharing protocol, an encrypted clustering result is obtained; The encrypted clustering result is decrypted, and the decrypted result is matched with the credit product to obtain the credit product intelligent recommendation result.

2. The credit product intelligent recommendation method based on multi-source data fusion according to claim 1 is characterized in that: The multi-source credit data includes basic customer information, credit history data, financial and asset status data, consumer behavior data, and career-related data.

3. The credit product intelligent recommendation method based on multi-source data fusion according to claim 1 is characterized in that: Based on each of the security nodes performing similarity calculation and clustering operations in the ciphertext space through a secret sharing protocol, an encrypted clustering result is obtained, including: Each of the security nodes accesses the ciphertext space based on a secret sharing protocol, maps the homomorphically encrypted data to feature spaces of different scales, calculates the similarity between feature vectors in the feature space of each scale, and weightedly fuses the similarity results at different scales to obtain a comprehensive similarity matrix; A clustering algorithm based on density peak is used to cluster the comprehensive similarity matrix to obtain an encrypted clustering result. In the clustering process, the local density of each data point and the encrypted distance to the high-density point are calculated using the characteristics of homomorphic encryption. The clustering center is determined according to the preset density threshold and distance threshold, and the homomorphic encrypted data is divided into corresponding clustering clusters to obtain an encrypted clustering result.

4. The credit product intelligent recommendation method based on multi-source data fusion according to claim 1 is characterized in that: Performing homomorphic encryption on the fused feature data to obtain homomorphic encrypted data includes: Collect node information of each security node respectively, wherein the node information includes hardware identification information, network environment feature information, and preset random seed information; For each security node, generate a subkey based on the corresponding node information; A multi-party computing collaborative mechanism is used to merge the sub-keys to generate a master key; The fused feature data is homomorphically encrypted based on the master key to obtain homomorphically encrypted data.

5. The credit product intelligent recommendation method based on multi-source data fusion according to claim 4 is characterized in that: For each security node, a subkey is generated based on the corresponding node information, including: For each security node, the hardware identification information is hash-encoded to obtain a hardware hash value; Extracting the network delay fluctuation value, bandwidth peak value and mean value from the network environment feature information, and combining them with the hardware hash value and the preset random seed information to form a node feature string; The node feature string is input into a key generator based on the principle of quantum encryption. The key generator uses the superposition and entanglement of quantum states to perform encryption transformation on the node feature string to generate an initial encryption subkey. The initial encryption subkey is embedded in the lattice space through a pre-constructed lattice structure, and multiple random walks and encryption perturbations are performed in the lattice space to generate the subkey.

6. The credit product intelligent recommendation method based on multi-source data fusion according to claim 4 is characterized in that: For each security node, a subkey is generated based on the corresponding node information, including: For each security node, the hardware identification information is segmented and encoded and converted into a multi-dimensional vector; Performing spectrum analysis on the network environment characteristic information, extracting the main frequency characteristics and mapping them into encryption vectors, and generating the preset random seed information into interference vectors; The multidimensional vector, encryption vector and interference vector are used as pheromones and heuristic factors in the ant colony algorithm to perform path search and combinatorial optimization in the high-dimensional vector space; during the search process, the number of iterations and parameters of the algorithm are dynamically adjusted according to the complexity of the node information, and the subkey of the initial form is gradually converged; The initial subkey is embedded into the lattice structure, and after multiple rounds of lattice operations and verifications, the subkey corresponding to each security node is finally determined.

7. An intelligent credit product recommendation system based on multi-source data fusion, characterized in that: include: A fusion unit is used to extract features from multi-source credit data to obtain a feature vector set; Performing data fusion on each feature vector in the feature vector set to obtain fused feature data; An encryption unit, used for performing homomorphic encryption on the fused feature data to obtain homomorphic encrypted data, and storing the data in each security node; A clustering unit, used to perform similarity calculation and clustering operation in the ciphertext space based on each of the security nodes through a secret sharing protocol to obtain an encrypted clustering result; The recommendation unit is used to decrypt the encrypted clustering result, match the decrypted result with the credit product, and obtain the credit product intelligent recommendation result.

8. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images