Credit product personalized recommendation method and system based on enterprise credit evaluation

Abstract

The invention relates to the technical field of product personalized recommendation, in particular to a credit product personalized recommendation method and system based on enterprise credit assessment, and the method comprises the steps: obtaining the basic data and known social relation data of a to-be-assessed enterprise and a known enterprise sent to a server side by a client; the server side performs feature extraction on the basic data to obtain basic feature data; constructing a heterogeneous knowledge graph based on the basic feature data of the to-be-assessed enterprise and the known enterprise and the known social relationship, and then inputting the heterogeneous knowledge graph into the trained social relationship reasoning model to obtain comprehensive social relationship data of the to-be-assessed enterprise; the server side inputs the basic feature data and the comprehensive social relation data of the to-be-evaluated enterprise into the trained analysis network to obtain a credit rating evaluation result of the to-be-evaluated enterprise; and determining a credit product matched with the enterprise, and feeding back the recommended credit product to the client. According to the invention, personalized credit products can be recommended for enterprises.

Description

Technical Field

[0001] The present invention relates to the technical field of personalized product recommendation, and in particular to a method and system for personalized recommendation of credit products based on enterprise credit evaluation. Background Art

[0002] Before recommending credit products to corporate user groups, banks need to understand the attributes and characteristics of each enterprise. They can do this through information collection or questionnaire surveys.

[0003] The inventors discovered that existing questionnaires for credit product recommendations can omit or erroneously enter data due to negligence on the part of business users. Furthermore, some business users are reluctant to cooperate, resulting in inaccurate categorization of business users, which in turn hinders accurate credit product recommendations. For users uninterested in a product, ineffective advertising can be distracting, diminishing their experience. For users with potential purchasing intent, inaccurate ad categorization forces them to re-sort and screen all products, requiring significant effort and strong, professional identification skills. This results in a poor match between the products offered by the service platform and customer needs. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention provides a method and system for personalized credit product recommendations based on corporate credit assessments. This method not only uses a company's own basic information as a key indicator for assessing its credit rating, but also considers important external factors such as the industry, market environment, and industry trends. Furthermore, the present invention comprehensively considers the social relationships between the company and its shareholders involved in credit activities, thereby achieving a comprehensive and accurate assessment of the company's credit rating, reflecting its financial risk, and subsequently recommending personalized credit products that are more appropriate to its development stage and financing capabilities, thereby enhancing the pertinence and effectiveness of financial services.

[0005] On the one hand, a personalized credit product recommendation method based on corporate credit assessment is provided, which is applied on the server side and includes: Obtaining basic data and known social relationship data of the enterprise to be evaluated and known enterprises sent by the client to the server, wherein the basic data includes internal data and external data; the server extracts features from the basic data to obtain basic feature data; Construct a heterogeneous knowledge graph based on the basic feature data and known social relationships of the enterprise to be evaluated and known enterprises, and then input the heterogeneous knowledge graph into the trained social relationship reasoning model to obtain the comprehensive social relationship data of the enterprise to be evaluated; The server inputs the basic feature data and comprehensive social relationship data of the enterprise to be evaluated into the trained analysis network to obtain the credit rating evaluation result of the enterprise to be evaluated; further based on the evaluation results, it determines the credit products that match the enterprise and feeds back the recommended credit products to the client.

[0006] On the other hand, a personalized recommendation system for credit products based on enterprise credit assessment is provided, including: client and server; The server side obtains the basic data and known social relationship data of the enterprise to be evaluated and the known enterprises sent by the client to the server side, wherein the basic data includes internal data and external data; the server side extracts features from the basic data to obtain basic feature data; The server constructs a heterogeneous knowledge graph based on the basic feature data and known social relationships of the enterprise to be evaluated and known enterprises, and then inputs the heterogeneous knowledge graph into the trained social relationship inference model to obtain the comprehensive social relationship data of the enterprise to be evaluated; The server inputs the basic feature data and comprehensive social relationship data of the enterprise to be evaluated into the trained analysis network to obtain the credit rating evaluation result of the enterprise to be evaluated; further based on the evaluation results, it determines the credit products that match the enterprise and feeds back the recommended credit products to the client.

[0007] The above technical solution has the following advantages or beneficial effects: (1) The present invention proposes a method and system for personalized recommendation of credit products based on enterprise credit assessment. This method and system not only uses the basic information of the enterprise itself as an important indicator for assessing its credit rating, but also considers important external factors of the enterprise, such as the industry and market environment, and industry situation. At the same time, the present invention also comprehensively considers the relevant social relationships between the enterprise and its shareholders in the credit activities, and promotes personalized and accurate matching of credit products with enterprise needs through comprehensive and accurate assessment of the enterprise's credit rating.

[0008] (2) This invention proposes a method for inferring multi-subject relationships in credit activities, which conducts in-depth inference on the complex relationships between multiple entities such as enterprises and shareholders. This method, combined with artificial intelligence technology, can effectively explore the potential social relationships between the enterprise to be evaluated and its shareholders, revealing key information that is difficult to capture in traditional data analysis. This multi-dimensional relationship inference helps to improve the accuracy of corporate credit assessments, enables credit institutions to obtain more comprehensive reference basis in the decision-making process, and promotes personalized and precise matching of credit products with corporate needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0010] Figure 1 This is a flow chart of the method of embodiment 1. DETAILED DESCRIPTION

[0011] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0012] Known enterprises in this invention refer to enterprises that have already registered and exist in the client system. Credit assessment is crucial for enterprises. It not only determines their access to key resources such as financing and credit, but also directly impacts supply chain collaboration, subsidy applications, and market trust. Existing credit assessment methods typically focus on the enterprise's internal information while ignoring the influence of external factors, limiting the accuracy of credit assessments. Furthermore, existing credit assessment methods overlook two key factors crucial to enterprise credit assessments: external factors such as the industry and market environment, and the industry landscape; and the social relationships between the enterprise and its shareholders during credit activities (social relationships between the enterprise and its shareholders and other enterprises, or between shareholders of other enterprises). For enterprises with incomplete information disclosure, existing methods cannot fully and objectively reflect their true credit status. Therefore, relying solely on internal enterprise information while ignoring the influence of relevant social relationships and external information often makes it difficult to accurately assess an enterprise's credit rating. This not only weakens the accuracy of personalized recommendations but also affects financing efficiency and the rationality of resource allocation.

[0013] Example 1 This embodiment provides a personalized recommendation method for credit products based on enterprise credit assessment; like Figure 1 As shown in FIG, a personalized credit product recommendation method based on enterprise credit assessment is applied on the server side and includes: S101: Obtaining basic data and known social relationship data of the enterprise to be evaluated and known enterprises sent by the client to the server, wherein the basic data includes internal data and external data; the server extracts features from the basic data to obtain basic feature data; S102: Constructing a heterogeneous knowledge graph based on the basic feature data and known social relationships of the enterprise to be evaluated and known enterprises, and then inputting the heterogeneous knowledge graph into the trained social relationship reasoning model to obtain comprehensive social relationship data of the enterprise to be evaluated; S103: The server inputs the basic feature data and comprehensive social relationship data of the enterprise to be evaluated into the trained analysis network to obtain the credit rating evaluation result of the enterprise to be evaluated; further based on the evaluation result, the server determines the credit product that matches the enterprise and feeds back the recommended credit product to the client.

[0014] Furthermore, the internal data includes: corporate shareholder structure, professional personnel configuration, number of authorized patents, corporate cash flow, corporate debt-to-asset ratio and corporate innovation capabilities, etc.

[0015] Furthermore, the external data includes: industry competition pattern and industry situation, etc.

[0016] Furthermore, social relationship data includes: investment, shareholding, executives, controlling shares, relatives, shared addresses, shared telephone numbers, overlapping positions and common legal persons, etc.

[0017] For example, internal data, under the premise of ensuring that data processing is legal and compliant, authorization is clear, and privacy is controllable, uses artificial intelligence technologies such as intelligent crawling and data fusion to collect internal corporate information for assessing the company's credit rating, such as the company's shareholder structure, professional personnel configuration, the number of authorized related patents, corporate cash flow, debt-to-asset ratio, innovation capabilities and other information data; among which, the innovation capability can be reflected through information such as professional personnel configuration and the number of authorized patents.

[0018] For example, external data, under the premise of clear authorization, controllable privacy and other legal and compliant conditions, is collected using artificial intelligence technologies such as intelligent crawling and data fusion to assess the company's credit rating, such as industry competition landscape, industry situation and other data.

[0019] The industry competition landscape can be evaluated through indicator data that can reflect the competitiveness of the industry. Under the premise of legality and compliance, artificial intelligence technologies such as smart crawling can be used to automatically extract key information data such as the number of patents, patent types, technical fields, etc. of inventions related to the industry from intellectual property websites to reflect the industry competition landscape.

[0020] The aforementioned industry situation can be obtained from news websites using AI technologies such as intelligent crawling, under the premise of legality and compliance. Subsequently, the relevant information is summarized and extracted using a large language model to reflect the industry situation.

[0021] Obtaining this external data is crucial. A merchant's credit rating is not only influenced by its own operating performance but also closely linked to external confidence and expectations. Even if a company's financials are robust and its operations are performing well, sudden changes in external factors can rapidly alter the market's assessment of its creditworthiness. Therefore, this paper prioritizes key external factors that influence a company's credit assessment, such as the industry's competitive landscape and industry trends. Furthermore, this paper also considers the relevant social relationships between the company and its shareholders.

[0022] Furthermore, the server extracts features from the basic data to obtain basic feature data, including: Standardize the structured data in the internal data to obtain the first type of feature data; Extract semantic vectors from unstructured data in internal data to obtain the second type of feature data; Extract semantic vectors from external data to obtain the third type of feature data; The first, second and third types of feature data are fused to obtain basic feature data.

[0023] It should be understood that structured internal data (such as financial and operating indicators like cash flow, debt-to-asset ratio, and number of authorized patents) undergoes standardization and feature engineering. Simultaneously, technologies such as OpenAI Embeddings in the field of natural language processing (NLP) are used to extract semantic vectors from unstructured internal data (such as shareholder structure, professional staffing, and innovation capabilities), as well as external data. Ultimately, all features are fused to construct a unified high-dimensional feature vector, forming the company's fundamental feature data.

[0024] The beneficial effect of this technical solution is that it integrates structured and unstructured data, automatically constructing a unified high-dimensional feature vector through normalization and semantic representation techniques such as OpenAI Embeddings. Compared to traditional methods that rely on manual feature design, this process significantly reduces manual intervention and reliance on expert experience, improving the efficiency and consistency of feature extraction. It also significantly enhances the comprehensiveness and depth of information representation, reduces reliance on manual feature construction, and improves the efficiency and consistency of feature extraction, providing a high-quality, automated, and scalable input foundation for subsequent analysis and decision-making.

[0025] Further, S102: construct a heterogeneous knowledge graph based on the basic feature data and known social relations of the enterprise to be evaluated and the known enterprises, and then input the heterogeneous knowledge graph into the trained social relation reasoning model to obtain the comprehensive social relation data of the enterprise to be evaluated, wherein the comprehensive social relation data includes: known social relation data and potential social relation data obtained based on the social relation reasoning model.

[0026] Furthermore, the trained social relationship inference model is implemented using a Relational Graph Convolutional Network (R-GCN). This R-GCN takes as input a heterogeneous knowledge graph built from basic data on companies and known companies, along with known social relationship data. Through continuous iterative learning, it infers the social relationships between companies and their shareholders.

[0027] The relational graph convolutional network (during the training process, the graph convolution operation of neighbor sampling and relationship differentiation is adopted, and the adjacent node features of different relationship types are used to construct a node representation with semantic expression ability, and further extract and update the embedding representation of each node in the graph layer by layer; combined with the relationship scoring function (such as DistMult), the given entity pair and relationship type triple Score it and judge whether it is established in the graph. Through positive and negative sample training, the model can infer the social relations between unknown companies and their shareholders, and finally train a social relationship inference model for inferring the social relations between companies and their shareholders. Each node in Extract attribute features, where: is a set of nodes, each node have dimensional initial eigenvector ,Right now , (superscript 0 indicates the 0th layer of the network), is the edge set, is a set of relationship types, .

[0028] During the training process, the relationship graph convolution network selects subgraphs through node sampling and obtains the associated nodes of each node under various relationships through neighbor sampling. , for nodes Sample neighbor set: ; in, Represents an edge node and nodes Previously had a relationship type , Representation node Correspondence The set of neighbor nodes of Define separate Real weight matrix ,Right now , used in the The model aggregates neighbors and distinguishes relationships at each layer, thereby enhancing the model's ability to express multiple semantics. During the graph convolution process, the model comprehensively considers the node's own characteristics and the structural information of its multi-relational neighbors, and updates the node embedding representation layer by layer: ; in, Represents a set of relations All relations in The summation operation, Indicates the node Correspondence The set of neighbor nodes All Neighbor nodes The summation operation, Indicates the Nodes in the layer The embedding representation of Indicates the Nodes in the layer Neighbor nodes In the The embedding representation in the layer, is a node Its own information weight matrix, is the normalization coefficient (usually or its square root), is a nonlinear function (such as ReLU); the final node embedding is: ; in, Indicates the Nodes in the layer The final embedding representation of is the total number of graph convolution layers; Furthermore, for any triple to be inferred , using the relationship scoring function Estimate its existence probability. For example, using the DistMult model, the scoring function is: ; in, express The transpose of and Respectively expressed in Nodes in the layer and nodes The final embedding representation is, Indicates that the relationship type Vector representation of Convert to a diagonal matrix.

[0029] It should be understood that each relationship (such as "investment", "holding", "partnership") has a unique , represents the semantic weight of the relationship in the embedding space; Furthermore, the score is mapped to the probability of existence through the Sigmoid function: ; in For the Sigmoid activation function, the parameters are continuously optimized through forward propagation and back propagation to make the prediction results Gradually approaching the true label ; Finally, a social relationship inference model is trained to infer the social relationship between enterprises and their shareholders.

[0030] Furthermore, the training process of the trained social relationship reasoning model includes: A first data set is constructed, comprising: basic characteristic data of enterprises with known complete social relationship data; known complete social relationship data includes: the existence of investment relationships, shareholding relationships, controlling relationships, kinship relationships, co-residence, shared telephone numbers, overlapping positions, and common legal persons; the complete social relationship data also includes: all real social relationship information between shareholders of the enterprises.

[0031] Based on the first data set, a heterogeneous knowledge graph is constructed, and all the real inter-enterprise relationship triples are extracted as positive sample sets. ; At the same time, a negative sample set is generated through negative sampling (such as replacing entities or relations) , used to construct a supervised learning training set; Input heterogeneous knowledge graphs into the social relationship reasoning model and train the model based on positive and negative samples; During the training process, the model parameters are continuously optimized by maximizing the positive sample scores and minimizing the negative sample scores; When the loss function value no longer decreases or the number of training iterations exceeds the preset upper limit, the training is stopped and the trained social relationship inference model is obtained.

[0032] Furthermore, the loss function of the social relationship inference model can be a binary cross entropy loss function: ; in, Represents the positive sample set All nodes in and nodes Relation triples The summation operation, Represents the negative sample set All nodes in and nodes Relation triples The summation operation, Representation node and nodes relation The scoring function, Representation node and nodes relation The scoring function, is the activation function.

[0033] Furthermore, the construction process of the first data set includes: data cleaning and feature extraction of the basic data of the sample enterprises to obtain basic feature data; the data cleaning includes: missing value processing, outlier detection and standardization processing, etc.

[0034] Furthermore, constructing a heterogeneous knowledge graph based on the first data set includes: The nodes of the heterogeneous knowledge graph are enterprises and shareholders; The edges of the heterogeneous knowledge graph are the relationships between the enterprises, between the enterprises and shareholders, and between shareholders; the relationships include: investment, shareholding, senior management, controlling stakes, kinship, common address, shared telephone number, intersection of positions, common legal person, etc.

[0035] The beneficial effect of the above technical solution is that the social relationship reasoning model uses the basic data of the enterprise to infer the relevant social relationships of the enterprise to be evaluated and its shareholders, which helps to conduct a more comprehensive and accurate assessment of the enterprise's credit rating, and then helps to personalize and accurately match credit products with corporate needs.

[0036] Furthermore, in step S103, the server inputs the basic feature data and comprehensive social relationship data of the enterprise to be evaluated into the trained analysis network to obtain the credit rating evaluation result of the enterprise to be evaluated. The specific training process of the trained analysis network includes: Constructing a second data set, the second data set including: basic feature data and complete social relationship data of known enterprises that have obtained credit rating evaluation results; credit rating evaluation results, including: low risk level (1-3 points), medium risk level (4-6 points), and high risk level (7-10 points); The second data set is input into the analysis network, and the analysis network is trained to obtain a trained analysis network.

[0037] It should be understood that the analysis network is implemented using a fully connected neural network, which takes basic feature data and social relationship data as input and outputs the credit rating prediction results of the enterprise.

[0038] When the analysis network is used to perform credit rating assessment training, the loss function is calculated through forward propagation and the loss is minimized for back propagation to continuously optimize the model parameters.

[0039] Furthermore, the specific expression of the "single sample loss" function of the analysis network is: ; Among them, the superscript Representative samples, whose true category is , for The predicted probability of the “true category” of samples is ,parameter is the focusing factor, which is used to adjust the model’s attention to difficult-to-classify samples. is the class weight coefficient, which is used to further alleviate the impact of uneven class distribution on training results. Finally, the loss function of the network is analyzed as follows: ; in is the total number of samples.

[0040] Compared to traditional methods, this method eliminates the need for comprehensive collection and manual analysis of complex information, significantly reducing manual intervention and manpower input, and lowering operational costs. Furthermore, through automated data processing and model inference mechanisms, it can rapidly assess corporate credit ratings, significantly improving assessment efficiency and processing speed, and providing efficient and reliable support for financing decisions.

[0041] Furthermore, the step S103: determining a credit product matching the enterprise based on the evaluation result, includes: If the credit rating assessment result is low risk, a low-risk credit product will be recommended; If the credit rating assessment result is medium risk, medium-risk credit products will be recommended; If the credit rating assessment result is high risk, a high-risk credit product will be recommended.

[0042] Furthermore, low-risk credit products are credit products offered by financial institutions to corporate clients with high credit ratings and strong repayment capabilities, characterized by low default probability and minimal risk exposure. Medium-risk and high-risk credit products are offered to clients with moderate or low credit ratings, respectively, and have corresponding risk control strategies and product features.

[0043] The beneficial effects of the above technical solution are: based on the results of corporate credit assessment, the present invention automatically matches and recommends credit products with corresponding risk levels, avoiding the traditional process of relying on manual review and complex information analysis, significantly reducing manual intervention and operating costs, while improving the efficiency and accuracy of personalized recommendations for credit products, enhancing the intelligence level of credit decision-making, and greatly improving assessment efficiency and processing speed.

[0044] It should be understood that the "credit products" mentioned in the present invention refer to loan services provided by financial institutions based on the business needs of enterprises, mainly including working capital loans, fixed asset loans, project financing, trade financing, bill discounting, etc., which are widely used in the daily operations of enterprises, fixed asset purchases, import and export trade, and capital turnover.

[0045] Example 2 This embodiment provides a personalized credit product recommendation system based on enterprise credit assessment, including: a client and a server; A personalized credit product recommendation system based on enterprise credit assessment, including: client and server; The server side obtains the basic data and known social relationship data of the enterprise to be evaluated and the known enterprises sent by the client to the server side, wherein the basic data includes internal data and external data; the server side extracts features from the basic data to obtain basic feature data; The server constructs a heterogeneous knowledge graph based on the basic feature data and known social relationships of the enterprise to be evaluated and known enterprises, and then inputs the heterogeneous knowledge graph into the trained social relationship inference model to obtain the comprehensive social relationship data of the enterprise to be evaluated; The server inputs the basic feature data and comprehensive social relationship data of the enterprise to be evaluated into the trained analysis network to obtain the credit rating evaluation result of the enterprise to be evaluated; further based on the evaluation results, it determines the credit products that match the enterprise and feeds back the recommended credit products to the client.

[0046] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A personalized recommendation method for credit products based on corporate credit assessment, characterized by: Applied to the server side, including: Obtaining basic data and known social relationship data of the enterprise to be evaluated and known enterprises sent by the client to the server, wherein the basic data includes internal data and external data; the server extracts features from the basic data to obtain basic feature data; Construct a heterogeneous knowledge graph based on the basic feature data and known social relationships of the enterprise to be evaluated and known enterprises, and then input the heterogeneous knowledge graph into the trained social relationship reasoning model to obtain the comprehensive social relationship data of the enterprise to be evaluated; The server inputs the basic feature data and comprehensive social relationship data of the enterprise to be evaluated into the trained analysis network to obtain the credit rating evaluation result of the enterprise to be evaluated; further based on the evaluation results, it determines the credit products that match the enterprise and feeds back the recommended credit products to the client.

2. The personalized credit product recommendation method based on enterprise credit assessment according to claim 1, characterized in that: The server extracts features from the basic data to obtain basic feature data, including: Standardize the structured data in the internal data to obtain the first type of feature data; Extract semantic vectors from unstructured data in internal data to obtain the second type of feature data; Extract semantic vectors from external data to obtain the third type of feature data; The first, second and third types of feature data are fused to obtain basic feature data.

3. The personalized credit product recommendation method based on enterprise credit assessment according to claim 1, characterized in that: A heterogeneous knowledge graph is constructed based on the basic feature data and known social relations of the enterprise to be evaluated and known enterprises. The heterogeneous knowledge graph is then input into the trained social relation reasoning model to obtain the comprehensive social relation data of the enterprise to be evaluated, where the comprehensive social relation data includes: known social relation data and potential social relation data obtained based on the social relation reasoning model; the trained social relation reasoning model is implemented using a relational graph convolutional network.

4. The personalized credit product recommendation method based on enterprise credit assessment according to claim 3 is characterized in that: During the training process, the relationship graph convolution network selects subgraphs through node sampling and obtains the associated nodes of each node under various relationships through neighbor sampling. , for nodes Sample neighbor set: ； in Represents an edge node and nodes Previously had a relationship type , Representation node Correspondence The set of neighbor nodes of Define separate Real weight matrix , , used in the The network aggregates neighbors and distinguishes relationships at each layer, thereby enhancing the network's ability to express multiple semantics. During the graph convolution process, the network comprehensively considers the node's own characteristics and the structural information of its multi-relational neighbors, and updates the node embedding representation layer by layer: ； in, Represents a set of relations All relations in The summation operation, Indicates the node Correspondence The set of neighbor nodes All Neighbor nodes The summation operation, Indicates the Nodes in the layer The embedding representation of Indicates the Nodes in the layer Neighbor nodes In the The embedding representation in the layer, is a node Its own information weight matrix, is the normalization coefficient, is a nonlinear function; the final node embedding is: ； in, Indicates the Nodes in the layer The final embedding representation of is the total number of graph convolution layers; For any triple to be inferred , using the relationship scoring function Estimate its existence probability; use the DistMult model, and the scoring function is: ； in, express The transpose of and Respectively expressed in Nodes in the layer and nodes The final embedding representation is, Indicates that the relationship type Vector representation of Convert to a diagonal matrix; The score is mapped to the probability of existence through the Sigmoid function: in For the Sigmoid activation function, the parameters are continuously optimized through forward propagation and back propagation to make the prediction results Gradually approaching the true label ; Finally, a social relationship inference model is trained to infer the social relationship between enterprises and their shareholders.

5. The personalized credit product recommendation method based on enterprise credit assessment according to claim 1 is characterized in that: The trained social relationship inference model. The training process includes: Constructing a first data set, the first data set comprising: basic characteristic data of enterprises with known complete social relationship data; the known complete social relationship data including: the existence of investment relationships, shareholding relationships, controlling relationships, kinship relationships, co-residence, shared telephone numbers, overlapping positions, and common legal persons; the complete social relationship data also includes: all real social relationship information between shareholders of the enterprises; Based on the first data set, a heterogeneous knowledge graph is constructed, and all the real inter-enterprise relationship triples are extracted as positive sample sets. ; At the same time, a negative sample set is generated by negative sampling , used to construct a supervised learning training set; Input heterogeneous knowledge graphs into the social relationship reasoning model and train the model based on positive and negative samples; During the training process, the model parameters are continuously optimized by maximizing the positive sample scores and minimizing the negative sample scores; When the loss function value no longer decreases or the number of training iterations exceeds the preset upper limit, the training is stopped and the trained social relationship inference model is obtained.

6. The personalized credit product recommendation method based on enterprise credit assessment according to claim 5, characterized in that: The loss function of the social relationship inference model can be a binary cross entropy loss function: ； in, Represents the positive sample set All nodes in and nodes Relation triples The summation operation, Represents the negative sample set All nodes in and nodes Relation triples The summation operation, Representation node and nodes relation The scoring function, Representation node and nodes relation The scoring function, is the activation function.

7. The personalized credit product recommendation method based on enterprise credit assessment according to claim 5, characterized in that: The step of constructing a heterogeneous knowledge graph based on the first data set includes: The nodes of the heterogeneous knowledge graph are enterprises and shareholders; The edges of the heterogeneous knowledge graph are the relationships between the enterprises, between the enterprises and shareholders, and between shareholders; the relationships include: investment, shareholding, senior management, holding, kinship, common address, shared telephone number, position intersection, and common legal person.

8. The personalized credit product recommendation method based on enterprise credit assessment according to claim 1, characterized in that: The server inputs the basic feature data and comprehensive social relationship data of the enterprise to be evaluated into the trained analysis network to obtain the credit rating evaluation result of the enterprise to be evaluated; The specific training process of the trained analysis network includes: Constructing a second data set, the second data set including: basic feature data and complete social relationship data of known enterprises that have obtained credit rating evaluation results; the credit rating evaluation results include: low risk level, medium risk level, and high risk level; The second data set is input into the analysis network, and the analysis network is trained to obtain a trained analysis network.

9. The personalized credit product recommendation method based on enterprise credit assessment according to claim 8, characterized in that: The specific expression of the "single sample loss function" of the analysis network is: ； in, Representative samples, whose true category is , for The predicted probability of the "true category" of the sample is ,parameter is the focusing factor, which is used to adjust the model’s attention to difficult-to-classify samples. is the category weight coefficient, which is used to further alleviate the impact of uneven class distribution on training results. Finally, the loss function of the network is analyzed as follows: ； in, is the total number of samples.

10. A personalized recommendation system for credit products based on corporate credit assessment, characterized by: include: Client and server; The server side obtains the basic data and known social relationship data of the enterprise to be evaluated and the known enterprises sent by the client to the server side, wherein the basic data includes internal data and external data; the server side extracts features from the basic data to obtain basic feature data; The server constructs a heterogeneous knowledge graph based on the basic feature data and known social relationships of the enterprise to be evaluated and known enterprises, and then inputs the heterogeneous knowledge graph into the trained social relationship inference model to obtain the comprehensive social relationship data of the enterprise to be evaluated; The server inputs the basic feature data and comprehensive social relationship data of the enterprise to be evaluated into the trained analysis network to obtain the credit rating evaluation result of the enterprise to be evaluated; further based on the evaluation results, it determines the credit products that match the enterprise and feeds back the recommended credit products to the client.

Images