Enterprise customer accurate screening system and method based on big data portrait

Abstract

The invention relates to the technical field of big data analysis and customer management, and discloses an enterprise customer accurate screening system and method based on big data portraits. The system comprises a data acquisition module, a data preprocessing module, a customer portrait generation module, an enterprise relation modeling module, a customer screening module, a dynamic updating module and a result display module. The system obtains multi-source data of an enterprise, and feature representation is generated after cleaning, standardization and feature fusion; constructing a customer portrait containing static attributes and dynamic attributes, and forming an enhanced portrait in combination with a supply chain, investment and cooperation relationship; on the basis, calculating a model score, a rule score and a risk penalty term to obtain a comprehensive score, and grading according to a preset threshold value; and when the data is changed, a customer portrait and a comprehensive score are adjusted through an event-driven incremental updating mechanism, so that the real-time performance of a screening result is ensured. According to the method, multi-source data fusion, portrait multi-dimensional modeling and dynamic updating are realized, and the accuracy and interpretability of enterprise customer screening are improved.

Description

Technical Field

[0001] This invention relates to the field of big data analytics and customer management technology, and more specifically, to a precise screening system and method for enterprise customers based on big data profiling. Background Technology

[0002] With the increasing digitalization of industries such as finance, insurance, supply chain management, and government regulation, customer screening and tiered management have become crucial aspects of business operations for various institutions. Taking financial lending as an example, banks and financial institutions need to assess the value and identify the risks of potential loan recipients to determine whether to grant credit and the credit limit. In supply chain scenarios, core enterprises also need to screen upstream and downstream companies to select high-quality suppliers or distributors. However, existing technologies for customer screening suffer from the following shortcomings: First, enterprise data sources are complex and heterogeneous. Internally, enterprises may possess business transaction data and some financial information; externally available data includes business registration records, legal proceedings, and public opinion information; third-party data sources may also provide credit ratings and industry analysis. Because these data vary significantly in format, terminology, and update frequency, existing systems often utilize only one type of data, resulting in a lack of comprehensiveness in customer assessment results.

[0003] Secondly, the methods for constructing customer profiles are relatively simplistic. Existing methods typically use static attributes of a company (such as registered capital and industry category) as the primary basis for judgment, ignoring dynamically changing information (such as public opinion, quarterly fluctuations in financial indicators, and trends in transaction behavior). This type of customer assessment based on static data often fails to reflect changes in a company's risk and value in a timely manner, easily leading to delayed screening results.

[0004] Secondly, the value of inter-firm networks is not fully utilized. In reality, companies often have upstream and downstream supply chain relationships, equity investment relationships, and cooperative relationships. Even if a company is operating well, if its upstream or downstream partners or controlling shareholders pose high risks, it can significantly impact the company's future credit and development potential. Most existing customer screening methods lack the ability to model corporate network structures and therefore cannot capture this indirect risk.

[0005] Furthermore, existing customer screening lacks a dynamic update mechanism. Most systems generate customer scores but only recalculate them during the next batch processing, while enterprise data, especially public opinion data, can change at any time. This static update method results in non-real-time customer screening results, failing to meet the needs of high-frequency business scenarios.

[0006] Finally, regarding customer scoring methods, traditional approaches often rely on rules set by human experience, lacking effective integration with machine learning models. Even when models are introduced, they are often disconnected from the rule system, failing to achieve complementarity between rules and models. Furthermore, for risk assessment, most methods only consider static credit information, failing to incorporate dynamic risk factors such as public opinion and records of defaults, resulting in limited risk identification capabilities.

[0007] In summary, existing enterprise customer screening methods generally suffer from problems such as insufficient data utilization, incomplete profile construction, lack of relationship modeling, untimely updates, and imperfect scoring systems, making it difficult to meet the needs for accurate, dynamic, and interpretable screening of enterprise customers.

[0008] Therefore, it is necessary to propose a new enterprise customer precision screening system and method based on big data profiling to solve these problems. Summary of the Invention

[0009] The purpose of this invention is to solve the technical problems mentioned in the background section above, and to provide a precise enterprise customer screening system based on big data profiling, comprising: Data acquisition module: used to acquire multi-source heterogeneous data from enterprise customers from internal business systems, external public databases, and third-party data services; Data preprocessing module: used to perform data cleaning, standardization, deduplication and feature alignment on the multi-source heterogeneous data to form structured feature vectors; Customer profile generation module: used to construct multi-dimensional profiles of enterprise customers based on the structured feature vectors, including static profiles to represent stable attributes such as business registration and industry category, and dynamic profiles to represent attributes that change over time such as financial indicators, news and public opinion, and business transactions. The static profiles and dynamic profiles are combined by the fusion unit to obtain a unified customer profile. Enterprise Relationship Modeling Module: Used to construct enterprise relationship graphs based on relationships between enterprises such as supply chains, investments, cooperation, and holding companies, and to generate graph-enhanced profiles containing upstream and downstream relationship information by weighted aggregation of enterprise nodes and their neighboring nodes; Customer screening module: Used to assess the value of enterprise customers, including scoring based on graph-enhanced profiling model, combining rule engine scores and risk penalty items to obtain the customer's total score, and further dividing customers into multiple levels according to preset stratification thresholds; Dynamic update module: Used to incrementally update the dynamic profile of customers when changes in enterprise data are detected, and to adjust the comprehensive score and level of customers in a synchronous manner to ensure the real-time nature of customer screening results; Results Display Module: This module displays customer screening results in a visual format and supports interactive analysis and queries.

[0010] As a preferred technical solution of the present invention, the data preprocessing module uses a robust statistical method to standardize numerical features in order to reduce the interference of extreme or outlier values ​​on enterprise feature vectors, and uses a vectorized representation method to convert categorical and textual features.

[0011] As a preferred technical solution of the present invention, the data preprocessing module assigns credibility weights to different data sources during multi-source data fusion, aligns features from different sources to a unified feature space through linear mapping, and then weights and fuses the features from each source according to their weights, thereby generating a more stable and reliable enterprise feature representation.

[0012] As a preferred technical solution of the present invention, the customer profile generation module includes a static channel, a dynamic channel, and a fusion unit. The static channel encodes information that does not change frequently over time, such as business registration and industry attributes, to obtain a static profile. The dynamic channel models time-series information such as financial data, news and public opinion, and contract transactions to obtain a dynamic profile. The fusion unit combines the static profile and the dynamic profile to form a unified customer profile representation.

[0013] As a preferred technical solution of the present invention, when constructing the enterprise relationship modeling module, the enterprise is treated as a node and the supply, investment, cooperation and other relationships between enterprises are treated as edges. The module assigns weights to different neighbor nodes through an attention mechanism and aggregates neighbor information to the target enterprise node, thereby generating a graph-enhanced profile that includes upstream and downstream dependencies and network structure features.

[0014] As a preferred technical solution of the present invention, the customer screening module obtains a model score for the enterprise based on the graph augmentation profile input of the enterprise into the trained machine learning model, which is used to reflect the comprehensive performance of the enterprise in terms of potential value, credit level or cooperation potential.

[0015] As a preferred technical solution of the present invention, the customer screening module performs weighted fusion of the model score and the business rule score calculated by the rule engine, and superimposes risk penalty items generated based on risk information such as public opinion and credit records, so as to obtain a comprehensive score of enterprise customers, thereby realizing the combination of model-driven and rule-driven approaches; the customer screening module divides customers into multiple levels according to the comprehensive score and preset stratification thresholds, including high-value low-risk customers, medium-value customers and high-risk customers, and marks each level of customers with different stratification labels.

[0016] As a preferred technical solution of the present invention, when the dynamic update module detects events such as updates to the company's financial data, changes in business registration, or new news and public opinion, it performs local incremental updates to the dynamic profile of the customer based on an event-driven mechanism, thereby adjusting the customer's comprehensive score and level in real time without the need for overall recalculation.

[0017] As a preferred technical solution of the present invention, the risk penalty item includes a public opinion risk factor. The public opinion risk factor is weighted and accumulated based on negative public opinion information occurring at different times, and a time decay mechanism is adopted to make the weight of recent negative information greater than that of historical information, so as to improve the real-time performance and accuracy of risk assessment.

[0018] This invention also provides a method for precise screening of enterprise customers based on big data profiling, comprising the following steps: (1) Obtain multi-source heterogeneous data from enterprise customers, including data from internal business systems, external public databases, and third-party data services; (2) Perform data cleaning, standardization, deduplication and feature alignment on the multi-source data to form a structured feature vector, and perform weighted fusion of features from different sources according to the source credibility to obtain a unified enterprise feature representation; (3) Constructing a customer profile based on the unified feature representation, specifically including: encoding information that does not change frequently over time into a static profile, modeling information that changes over time into a dynamic profile, and combining the static profile and the dynamic profile through a fusion unit to form a unified customer profile. (4) Construct an enterprise relationship graph, with enterprises as nodes and supply chain relationships, investment holding relationships, and business cooperation relationships as edges; perform weighted aggregation of the target enterprise node and its neighboring nodes based on the attention mechanism to generate a graph-enhanced profile containing upstream and downstream relationship information; (5) Input the graph-enhanced profile into the trained machine learning model to obtain the model score; obtain the rule score based on the business rule engine; obtain the risk penalty items based on information such as public opinion, credit records, judicial risks, and customer concentration; (6) The model score, rule score and risk penalty item are weighted and fused to obtain the comprehensive score of the enterprise customer; the enterprise customer is divided into multiple levels according to the preset stratification threshold and output. (7) Monitor enterprise data change events, including financial data updates, business registration changes, new news and public opinion events, changes in enterprise relationships, etc.; when a change event is detected, perform event-driven incremental updates, partially reconstruct the dynamic profile of customers and its related subgraphs, and refresh the comprehensive score and level simultaneously; (8) Display the screening results in a visual interface and provide interactive query and analysis.

[0019] Beneficial effects The enterprise customer precision screening system and method based on big data profiling proposed in this invention can effectively overcome the shortcomings of existing technologies and has the following beneficial effects: First, this invention enables the fusion and utilization of multi-source heterogeneous data. By introducing data cleaning, standardization, feature alignment, and weighted fusion mechanisms, it not only eliminates format and quality differences between data from different sources but also dynamically adjusts the contribution of data based on its credibility, thereby forming a unified and stable representation of enterprise characteristics. This makes customer screening results more comprehensive, avoiding the bias caused by relying on a single data source.

[0020] Secondly, this invention constructs a customer profile that combines static and dynamic features. The static profile stably describes the company's basic attributes, while the dynamic profile reflects the company's changing business practices and risk information over time. These two features are organically combined through a fusion unit to form a unified customer profile, making the characterization of company value and risk more accurate and comprehensive, reflecting the company's latest status in real time, and avoiding delays in screening results.

[0021] Furthermore, this invention introduces a corporate relationship modeling mechanism. By constructing a network of supply chain relationships, investment holding relationships, and cooperative relationships among enterprises, and employing an attention-based weighted aggregation method, it is possible to fully utilize the interconnected information between enterprises to model the transmission effects of risk and value. This not only improves the ability to identify potential high-risk customers but also helps to discover high-quality customers in advantageous network positions.

[0022] Fourth, this invention proposes a comprehensive scoring method that combines model scores, rule scores, and risk penalty items. By organically integrating the predictive power of machine learning models with the business experience of human rule-makers, and introducing risk factors such as public opinion, credit records, and legal proceedings as penalty items, this method can improve interpretability while ensuring the scientific nature of the scoring results, and avoid biases caused by a single model or a single rule.

[0023] Fifth, this invention provides an event-driven dynamic update mechanism. When changes in enterprise data are detected, local incremental updates to the dynamic profile and related enterprise relationships can be performed without overall recalculation, thereby adjusting customer scores and ratings in real time. This mechanism significantly improves the real-time performance and computational efficiency of the screening results, meeting the needs of business scenarios with extremely high timeliness requirements, such as financial lending and risk monitoring.

[0024] Sixth, this invention possesses excellent scalability and versatility. The system can be applied to customer credit assessment in financial institutions such as banks and insurance companies, as well as to scenarios such as manufacturing supply chain management, merchant screening on internet platforms, and government risk supervision, demonstrating strong industry adaptability.

[0025] In summary, this invention significantly improves the comprehensiveness, accuracy, real-time performance, and interpretability of enterprise customer screening by organically combining multi-source data fusion, static and dynamic features, enterprise relationship modeling, a comprehensive scoring system, and a dynamic update mechanism. It can provide various institutions with a more scientific and effective tool for customer segmentation and risk management, and has outstanding practical value and promotional significance. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 This is a flowchart of a method for precise screening of enterprise customers based on big data profiling proposed in this invention; Figure 2 This is a flowchart illustrating the data processing and customer rating process of this invention. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of this invention clearer, the following description is provided in conjunction with embodiments and appendices. Figure 1-2 The present invention will be further described in detail below. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0029] Example 1: This invention proposes a precise screening system and method for enterprise customers based on big data profiling. Through steps such as multi-source data collection, data preprocessing, customer profile construction, relationship modeling, scoring and rating, dynamic updating and risk correction, it achieves refined screening of enterprise customers.

[0030] The system first acquires multi-source heterogeneous data from the enterprise through its data acquisition module, including data from internal business systems (such as CRM and ERP), publicly available external data (such as business registration information, annual reports, and industry statistics), and third-party data sources (such as public opinion monitoring and credit rating agencies). This data is denoted as... Because data from different sources varies in format and quality, it is necessary to standardize the data before it enters the data preprocessing module. This module processes the raw data... Standardization should be performed, and robust statistical methods should be used to avoid the interference of extreme values, such as through formulas.

[0031] median MAD represents the median. Indicates the absolute deviation of the median. It is a constant that avoids division by zero. The processed eigenvector is denoted as... It serves as the input for subsequent modeling.

[0032] Since the data comes from multiple different channels, further integration is needed for each enterprise. , in the The characteristics of each data source are: The system assigns credibility weights to different data sources. and through a linear mapping matrix Align them to a uniform space, and the blending formula is:

[0033] in This indicates the number of data sources. This weighted fusion method can both ensure the diversity of data sources and enhance the robustness of the fused data.

[0034] The customer profile generation module is based on the above feature vectors Constructing multi-dimensional enterprise profiles. The system encodes relatively stable information about enterprises (such as registered capital and industry category) into static vectors. By inputting information that changes over time (such as financial indicators and news sentiment sequences) into a time series model, a dynamic vector is obtained. Subsequently, a fusion network is used to combine the two to obtain a unified image representation:

[0035] in This indicates that the merged customer profile can simultaneously represent both static and dynamic features.

[0036] Building upon this foundation, the system further considers the relationships between enterprises. In reality, enterprises are often closely connected through supply chains, investments, or collaborations; therefore, an enterprise relationship modeling module is introduced. This module treats enterprises as nodes in a graph and relationships as edges, employing an attention mechanism to calculate the influence between nodes. For enterprises... and his neighbors In relation types The attention weights are calculated as follows:

[0037] in Indicates enterprise In relation types The following is a collection of neighbors. This is the output of the scoring function. Node profiles are updated through neighbor-weighted aggregation:

[0038] in and This is the weight matrix. This is the activation function. The result obtained through this step... This is a graph-augmented profile that can capture the impact of upstream and downstream relationships and network structure. The customer screening module utilizes this graph-augmented profile. Calculate the model score for the enterprise:

[0039] in For the weight vector, This is a bias term. At the same time, the system can also calculate rule scores based on manually set business rules (such as industry category, registered capital threshold). It also generates risk penalty items based on indicators such as public opinion risk and credit record. The final overall score is:

[0040] in This is the balance coefficient between the model score and the rule score. This is the penalty coefficient.

[0041] Based on the overall score The system categorizes enterprise customers into different levels:

[0042] in, This is a preset threshold. High-level customers often imply high potential value or low risk.

[0043] Considering that customer information is dynamically changing, the system is designed with a dynamic update module. When a change event is detected in enterprise data, the system updates the dynamic profile. Perform incremental updates:

[0044] in For the newly input data sample, For feature extraction function, The updated coefficients have a range of values. The updated version This will reshape the fused portrait. Enhanced image This will lead to an update of the overall score. Tier-level rating. Risk penalty items. Special consideration was given to public opinion risks, and calculations were performed using a time decay mechanism:

[0045] in This represents the time difference between the current moment and the moment the negative event occurred. NegScore is the attenuation coefficient. For the first The score is calculated based on the number of negative public opinion events. This design ensures that recent negative public opinion events have a greater impact on the company profile, while the impact of past events gradually diminishes.

[0046] During training, the system employs a multi-task learning objective function to jointly optimize different business tasks. The task set is divided into a classification task set. and regression task set The loss function is defined as:

[0047] in, For the task The true label, Let BCE(·) be the predicted value, and BCE(·) be the binary cross-entropy function. Through multi-task joint training, the system can simultaneously optimize multiple dimensions such as customer value prediction, risk identification, and conversion probability assessment, thereby improving overall performance.

[0048] In summary, this system starts with raw data, and through preprocessing and multi-source fusion, constructs a graph-augmented profile by combining static and dynamic profiles with enterprise relationship modeling. It then calculates model scores and rule scores, introduces risk factors to form a comprehensive score, and ultimately achieves level classification and dynamic updates. This solution not only fully utilizes multi-source heterogeneous data but also improves the accuracy of screening through relationship modeling and public opinion correction, and maintains the real-time nature of the screening results through incremental updates.

[0049] Example 2: In a practical application scenario, the system of the present invention can be deployed in the credit business department of a financial institution to screen potential corporate loan customers. Financial institutions typically possess some internal customer information, but to comprehensively assess the credit risk and development potential of corporate customers, it is necessary to combine publicly available external data and third-party service data.

[0050] This embodiment details the specific application of the system in a financial lending scenario. Financial institutions first acquire multi-source heterogeneous data from enterprises through the data acquisition module. This data includes internal transaction records, loan history, and repayment records, as well as external business registration information, court announcements, bidding data, and third-party public opinion monitoring and credit rating data. Because the data from different sources varies significantly in format and quality, the system needs to send it to the data preprocessing module.

[0051] In the data preprocessing module, the system uses robust statistical methods to standardize the original numerical features to avoid interference from extreme values.

[0052] median It is the median of this feature, MAD It is the absolute median deviation. It is a small constant. The processing result yields a standardized feature vector. Given the complexity of data sources, the system will use a weighted fusion mechanism to unify multi-source data:

[0053] For example, data from a financial institution's internal credit system may have higher weighting, followed by publicly available business registration data, and third-party public opinion monitoring data has lower weighting. In this way, the final result is the enterprise's... Unified feature representation In the customer profile generation module, the system will... It is divided into two parts: static features and dynamic features.

[0054] A company's registered capital, industry classification, legal representative, etc., are used as static features and encoded into a static vector. The company's financial statements, news and public opinion sequences, and loan repayment behavior are used as dynamic features, which are then input into a recurrent neural network to obtain a dynamic vector. Subsequently, the static and dynamic vectors are synthesized into a unified customer profile through a fusion network:

[0055] in It incorporates both stable and temporally changing attributes of enterprises, providing a more comprehensive representation of their status. Financial institutions also focus on the position of enterprises within supply chains and investment networks; therefore, the system introduces an enterprise relationship modeling module. The system uses enterprises as nodes in a graph, and upstream and downstream procurement relationships and investment holding relationships as edges, constructing an enterprise relationship graph. For a given enterprise node... Its neighboring nodes The weight of its influence is calculated using an attention mechanism:

[0056] in For enterprises The neighborhood group, For the scoring function. Node representation updated as follows:

[0057] In this way, financial institutions can identify clients strongly associated with high-risk enterprises, even if the target enterprise itself has a sound financial condition, they can still detect potential risks. In the client screening module, the system uses graph-based augmented profiling. Calculate the model score:

[0058] in For the weight vector, This is a bias term. Simultaneously, the system combines a rules engine with scoring. (e.g., the registered capital must exceed a certain threshold, and whether the industry belongs to a priority industry supported by financial institutions), and calculate risk penalty items. The formula for calculating public opinion risk is as follows:

[0059] here It is the time difference of the event. This is a score for each negative news event. Recent negative news has a greater impact on risk, while the impact of older negative news gradually diminishes. Ultimately, the company's overall score is calculated using the following formula:

[0060] in It is the balance coefficient between model and rule scores. This is the risk penalty coefficient. Based on the comprehensive score... Financial institutions divide their customers into four levels:

[0061] in, These are tiered thresholds. Grade A customers may be key marketing and lending targets, while Grade D customers are marked as high-risk customers.

[0062] In the dynamic update module, the system monitors changes to enterprise data, such as financial report updates, new news and public opinion updates, and changes to business registration information. When new data is detected... At that time, the system updates the dynamic profile:

[0063] in It is the update coefficient. It is a feature extraction function. Thus, the company's overall score... The Tier rating can be dynamically updated at any time during the model training phase, and financial institutions train the system parameters using historical loan customer data.

[0064] A multi-task learning approach is employed to predict both a company's eligibility for a loan (classification task) and its future probability of default (regression task). The loss function is:

[0065] in It is a collection of categorized tasks. It is a set of regression tasks. It's a real label. These are predicted values. A system trained in this way can perform both customer screening and risk prediction, meeting the actual needs of financial lending.

[0066] This embodiment enables financial institutions to quickly identify high-value, low-risk corporate clients and prioritize them for credit. Simultaneously, the system dynamically updates client profiles, ensuring that scores and ratings reflect the latest status of the businesses in real time. Compared to traditional methods relying solely on financial statements and human experience, this invention significantly improves the accuracy, real-time nature, and interpretability of client screening.

[0067] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A precise customer screening system for enterprises based on big data profiling, characterized in that, include: Data acquisition module: used to acquire multi-source heterogeneous data of enterprise customers from internal business systems, external public databases and third-party data services; Data preprocessing module: used to perform data cleaning, standardization, deduplication and feature alignment on the multi-source heterogeneous data to form structured feature vectors; Customer profile generation module: used to construct multi-dimensional profiles of enterprise customers based on the structured feature vectors, including static profiles to represent business registration and industry category attributes, and dynamic profiles to represent financial indicators, news and public opinion, and business transactions that change over time. The static profiles and dynamic profiles are combined by the fusion unit to obtain a unified customer profile. Enterprise Relationship Modeling Module: Used to construct enterprise relationship graphs based on supply chain, investment, cooperation, and holding relationships between enterprises, and generate graph-enhanced profiles containing upstream and downstream relationship information by weighted aggregation of enterprise nodes and their neighboring nodes; Customer screening module: Used to assess the value of enterprise customers, including scoring based on graph-enhanced profiling model, combining rule engine scores and risk penalty items to obtain the customer's total score, and further dividing customers into multiple levels according to preset stratification thresholds; Dynamic update module: Used to incrementally update the dynamic profile of customers when changes in enterprise data are detected, and to adjust the comprehensive score and level of customers in a synchronous manner to ensure the real-time nature of customer screening results; Results Display Module: This module displays customer screening results in a visual format and supports interactive analysis and queries.

2. The enterprise customer precision screening system based on big data profiling as described in claim 1, characterized in that, The data preprocessing module uses robust statistical methods to standardize numerical features to reduce the interference of extreme or outlier values ​​on enterprise feature vectors, and uses vectorized representation to convert categorical and textual features.

3. The system according to claim 1, characterized in that, The data preprocessing module assigns credibility weights to different data sources during multi-source data fusion, aligns features from different sources to a unified feature space through linear mapping, and then weights and fuses the features from each source to generate a more stable and reliable enterprise feature representation.

4. The system according to claim 1, characterized in that, The customer profile generation module includes a static channel, a dynamic channel, and a fusion unit. The static channel encodes information such as business registration and industry attributes that do not change frequently over time to obtain a static profile. The dynamic channel models financial data, news sentiment, and contract transaction time-series information to obtain dynamic profiles; The fusion unit combines static and dynamic profiles to form a unified customer profile representation.

5. The system according to claim 1, characterized in that, When constructing the enterprise relationship modeling module, enterprises are treated as nodes, and supply, investment, and cooperation relationships between enterprises are treated as edges. The module assigns weights to different neighbor nodes through an attention mechanism and aggregates neighbor information to the target enterprise node, thereby generating a graph-enhanced profile that includes upstream and downstream dependencies and network structure features.

6. The system according to claim 1, characterized in that, The customer screening module uses a pre-trained machine learning model based on the enterprise's graph augmented profile input to obtain the enterprise's model score, which reflects the enterprise's overall performance in terms of potential value, credit level, or cooperation potential.

7. The enterprise customer precision screening system based on big data profiling according to claim 6, characterized in that, The customer screening module weights and merges the model score with the business rule score calculated by the rule engine, and superimposes risk penalty items generated based on public opinion and credit record risk information to obtain a comprehensive score for enterprise customers, thereby achieving a combination of model-driven and rule-driven approaches. The customer screening module divides customers into multiple levels based on the comprehensive score and preset stratification thresholds, including high-value low-risk customers, medium-value customers, and high-risk customers, and marks each level of customers with different tier labels.

8. The enterprise customer precision screening system based on big data profiling according to claim 1, characterized in that, When the dynamic update module detects updates to the company's financial data, business registration changes, or new news and public opinion events, it performs partial incremental updates to the customer's dynamic profile based on an event-driven mechanism, thereby adjusting the customer's comprehensive score and level in real time without the need for a complete recalculation.

9. The system according to claim 7, characterized in that, The risk penalty items include public opinion risk factors, which are weighted and accumulated based on negative public opinion information occurring at different times, and a time decay mechanism is used to make the weight of recent negative information greater than that of historical information, so as to improve the real-time performance and accuracy of risk assessment.

10. A method for precise screening of enterprise customers based on big data profiling, characterized in that, Includes the following steps: (1) Obtain multi-source heterogeneous data from enterprise customers, including data from internal business systems, external public databases, and third-party data services; (2) Perform data cleaning, standardization, deduplication and feature alignment on the multi-source data to form a structured feature vector, and perform weighted fusion of features from different sources according to the source credibility to obtain a unified enterprise feature representation; (3) Constructing a customer profile based on the unified feature representation, specifically including: encoding information that does not change frequently over time into a static profile, modeling information that changes over time into a dynamic profile, and combining the static profile and the dynamic profile through a fusion unit to form a unified customer profile. (4) Construct an enterprise relationship graph, with enterprises as nodes and supply chain relationships, investment holding relationships, and business cooperation relationships as edges; perform weighted aggregation of the target enterprise node and its neighboring nodes based on the attention mechanism to generate a graph-enhanced profile containing upstream and downstream relationship information; (5) Input the image-enhanced profile into the trained machine learning model to obtain the model score; obtain the rule score based on the business rule engine; obtain the risk penalty items based on public opinion, credit records, judicial risks, and customer concentration information; (6) The model score, rule score and risk penalty item are weighted and fused to obtain the comprehensive score of the enterprise customer; the enterprise customer is divided into multiple levels according to the preset stratification threshold and output. (7) Monitor enterprise data change events, including financial data updates, business registration changes, new news and public opinion events, and changes in enterprise relationships; when a change event is detected, perform event-driven incremental updates, partially reconstruct the dynamic profile of the customer and its related sub-graphs, and simultaneously refresh the comprehensive score and rating; (8) Display the screening results in a visual interface and provide interactive query and analysis.

Images