Credit risk control decision method, device and equipment based on multi-source data fusion and medium

Abstract

The application provides a credit business risk control decision method, device, equipment and medium based on multi-source data fusion, which comprises the following steps: acquiring multi-source data related to the subject identifier of a credit subject; performing entity analysis and alignment on the multi-source data based on the semantic features of each entity in the multi-source data, and constructing an initial knowledge graph; performing fusion and conflict resolution on the multi-source data through a space-time decay fusion model based on the confidence score of each data source in the initial knowledge graph and the corresponding time decay weight, performing entity screening on the initial knowledge graph based on the fusion result, and obtaining an optimized knowledge graph; for each entity node associated with a heterogeneous data source, calling a scoring model adapted to the data characteristics thereof, and generating a comprehensive credit score through a model integration mechanism; and determining a risk control decision for the credit subject based on the comprehensive credit score. The method, device, equipment and medium provided by the application realize accurate and efficient risk control decision, and significantly improve the risk identification accuracy.

Description

Technical Field

[0001] This invention relates to the fields of financial technology and artificial intelligence, and in particular to a credit business risk control decision-making method, apparatus, equipment and medium based on multi-source data fusion. Background Technology

[0002] In the financial lending sector, accurately assessing the credit risk of borrowers is a core aspect of risk management for financial institutions. Traditional credit scoring models primarily rely on static indicators from single data sources (such as credit reports and bank statements), which suffer from limited data coverage and insufficient real-time data, making it difficult to comprehensively reflect the dynamic credit status of borrowers. With the rapid development of financial technology, multi-source heterogeneous data is gradually becoming an important source of information to supplement traditional credit reporting data.

[0003] However, cross-system data alignment accounts for over 60% of the time consumption and suffers from semantic conflicts (such as significant differences in the definition of "income stability" across different systems). This leads to a high false rejection rate for rule engine-based methods and an inability to identify cross-device associated risks, resulting in distorted risk assessments. Furthermore, the update frequency and reliability of data from different data sources vary significantly. For example, IoT device data may be updated at the second level but contains sensor noise, while credit data has a long update cycle but high authority, thus reducing the reliability of risk assessments. Summary of the Invention

[0004] This invention provides a credit business risk control decision-making method, apparatus, equipment, and medium based on multi-source data fusion, to solve the defects of poor reliability of risk assessment caused by multi-source heterogeneous data in the prior art.

[0005] This invention provides a credit business risk control decision-making method based on multi-source data fusion, comprising:

[0006] Acquire multi-source data related to the entity identifier of the credit entity, including credit data, IoT device data, and social network data;

[0007] Based on the semantic features of each entity in the multi-source data, entity parsing and alignment are performed on the multi-source data to construct an initial knowledge graph for the credit subject;

[0008] Based on the confidence scores of each data source in the initial knowledge graph and their corresponding time decay weights, the multi-source data is fused and conflict resolved through a spatiotemporal decay fusion model. Based on the fusion results, entity filtering is performed on the initial knowledge graph to obtain an optimized knowledge graph.

[0009] For each entity node in the optimized knowledge graph associated with a heterogeneous data source, a scoring model adapted to its data features is invoked, and a comprehensive credit score is generated through a model integration mechanism.

[0010] Based on the comprehensive credit score, risk control decisions are determined for the credit entity.

[0011] According to the credit business risk control decision-making method based on multi-source data fusion provided by the present invention, the step of performing entity parsing and alignment of the multi-source data based on the semantic features of each entity in the multi-source data includes:

[0012] Extract the semantic features of each entity in the multi-source data;

[0013] Treat each entity as a node, define the associated edges between nodes according to business rules, and calculate the topological distance weight of the edges:

[0014] Based on the semantic features of each node and the topological distance weight of the edges, the similarity between each pair of nodes is calculated.

[0015] Based on the similarity between each pair of nodes and a preset similarity threshold, the nodes are merged.

[0016] According to the credit business risk control decision-making method based on multi-source data fusion provided by the present invention, the calculation of the similarity between every two nodes based on the semantic features of each node and the topological distance weight of the edges includes:

[0017] The similarity between any two nodes is determined using the following formula:

[0018]

[0019] in, For nodes With nodes Similarity between them For nodes semantic features For nodes semantic features For similarity calculation function, For nodes With nodes The topological distance weight between edges.

[0020] According to the credit business risk control decision-making method based on multi-source data fusion provided by the present invention, the step of fusing and resolving conflicts of the multi-source data through a spatiotemporal decay fusion model based on the confidence scores of each data source in the initial knowledge graph and their corresponding time decay weights includes:

[0021] Based on the confidence scores of each data source and their corresponding time decay weights, the multi-source data is time-weightedly fused to obtain the time fusion confidence score.

[0022] Based on the differences in time decay weights among the various data sources and the time fusion confidence, the conflict resolution confidence is calculated.

[0023] The time fusion confidence score is fused with the conflict resolution confidence score to obtain the final fusion confidence score.

[0024] According to the credit business risk control decision-making method based on multi-source data fusion provided by the present invention, before performing time-weighted fusion of the multi-source data based on the confidence scores of each data source and their corresponding time decay weights, the method further includes:

[0025] Extract multi-source data from the initial knowledge graph, with each data source including a confidence score and a data timestamp;

[0026] Based on the data timestamp, a time decay weight is calculated for each data source based on the time decay function.

[0027] According to the credit business risk control decision-making method based on multi-source data fusion provided by the present invention, the step of calling a scoring model adapted to the data characteristics of each heterogeneous data source associated with each entity node in the optimized knowledge graph, and generating a comprehensive credit score through a model integration mechanism, includes:

[0028] Based on the data type of each data source, the scoring model adapted to the data type is called to obtain a separate credit score for each scoring model.

[0029] Adjust the dynamic weights of each scoring model;

[0030] The individual credit scores are weighted and fused based on the adjusted dynamic weights to obtain the comprehensive credit score.

[0031] According to the credit business risk control decision-making method based on multi-source data fusion provided by the present invention, adjusting the dynamic weights of each scoring model includes:

[0032] The dynamic weights of each scoring model are adjusted based on the following formula:

[0033]

[0034] In the formula, For the scoring model Dynamic weights, For the scoring model of value, For all A rating model Sum of values This is a characteristic drift index.

[0035] The present invention also provides a credit business risk control decision-making device based on multi-source data fusion, comprising:

[0036] The data acquisition unit is used to acquire multi-source data related to the entity identifier of the credit entity, including credit data, Internet of Things device data, and social network data.

[0037] The graph construction unit is used to perform entity parsing and alignment of the multi-source data based on the semantic features of each entity in the multi-source data, so as to construct an initial knowledge graph for the credit subject.

[0038] The data fusion unit is used to fuse and resolve conflicts of the multi-source data based on the confidence scores of each data source in the initial knowledge graph and their corresponding time decay weights through a spatiotemporal decay fusion model, and to perform entity screening on the initial knowledge graph based on the fusion results to obtain an optimized knowledge graph.

[0039] The credit scoring unit is used to call a scoring model that is adapted to the data features of each entity node associated with the heterogeneous data source in the optimized knowledge graph, and generate a comprehensive credit score through the model integration mechanism.

[0040] The decision-making unit is used to determine risk control decisions for the credit subject based on the comprehensive credit score.

[0041] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the credit business risk control decision-making method based on multi-source data fusion as described above.

[0042] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the credit business risk control decision-making method based on multi-source data fusion as described above.

[0043] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the credit business risk control decision-making method based on multi-source data fusion as described above.

[0044] The present invention provides a credit business risk control decision-making method, device, equipment, and medium based on multi-source data fusion. By integrating heterogeneous data from multiple sources such as credit reporting, IoT devices, and social networks, it constructs an initial knowledge graph based on the BERT-Graph model to achieve high-precision entity parsing and alignment. It dynamically adjusts data weights and resolves conflicts using a spatiotemporal decay fusion model. After filtering and optimizing the knowledge graph, it calls an adaptive scoring model to generate a comprehensive credit score for heterogeneous data sources. Ultimately, it achieves accurate and efficient risk control decisions, significantly improves the accuracy, real-time performance, and robustness of risk identification, reduces false rejection rates and fraud losses, and enhances the effectiveness of financial risk control. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0046] Figure 1 This is one of the flowcharts of the credit business risk control decision-making method based on multi-source data fusion provided by the present invention.

[0047] Figure 2 This is a schematic diagram of the structure of the credit business risk control decision-making device based on multi-source data fusion provided by the present invention.

[0048] Figure 3 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0050] To address the issue of unreliable risk assessment caused by multi-source heterogeneous data, this invention provides a knowledge-driven credit risk control decision-making method that integrates multi-source heterogeneous data and supports dynamic time decay and conflict resolution. This method first acquires multi-source data related to the credit entity's identifier, including credit data, IoT device data, and social network data. Based on the semantic features of each entity in the multi-source data, entity parsing and alignment are performed to construct an initial knowledge graph for the credit entity, transforming the multi-source heterogeneous data into a standardized entity-relationship graph and unifying fields from different data sources. This approach aims to eliminate data silos and provide a structured foundation for subsequent fusion computation. Based on the confidence scores of each data source in the initial knowledge graph and their corresponding time decay weights, a spatiotemporal decay fusion model is used to fuse and resolve conflicts among multi-source data, improving the model's sensitivity to dynamic risks. Based on the fusion results, entity screening is performed on the initial knowledge graph to obtain an optimized knowledge graph. For each entity node in the optimized knowledge graph associated with heterogeneous data sources, a scoring model adapted to its data characteristics is invoked, and a comprehensive credit score is generated through a model integration mechanism. Based on the comprehensive credit score, risk control decisions are determined for credit entities.

[0051] This invention integrates heterogeneous data from multiple sources, including credit reporting, IoT devices, and social networks. Based on the BERT-Graph model, it achieves high-precision entity parsing and alignment to construct an initial knowledge graph. A spatiotemporal decay fusion model is used to dynamically adjust data weights and resolve conflicts. After filtering and optimizing the knowledge graph, an adaptive scoring model is called to generate a comprehensive credit score for heterogeneous data sources. Ultimately, this achieves accurate and efficient risk control decisions, significantly improving the accuracy, real-time performance, and robustness of risk identification, reducing false rejection rates and fraud losses, and enhancing the effectiveness of financial risk control.

[0052] The embodiments of this invention can be applied to risk control decision-making scenarios that require credit business, such as banking, supply chain finance, and cross-border payments. The executing entity of this method can be an electronic device such as a terminal device, computer, server, server cluster, or specially designed risk control decision-making device, or a risk control decision-making device installed in such an electronic device, which can be implemented through software, hardware, or a combination of both.

[0053] In the description of the embodiments of the present invention, "multiple" means two or more, unless otherwise explicitly specified.

[0054] Figure 1 This is one of the flowcharts illustrating the credit business risk control decision-making method based on multi-source data fusion provided by the present invention, such as... Figure 1 As shown, the method includes the following steps:

[0055] Step 110: Obtain multi-source data related to the entity identifier of the credit entity. The multi-source data includes credit data, IoT device data, and social network data.

[0056] Specifically, the credit subject refers to individuals or enterprises applying for credit services from financial institutions. Its core characteristics include subject identification, such as ID card number, unified social credit code, credit history, and financial status, which are the objects of credit assessment.

[0057] Multi-source data refers to data collections from different systems or platforms, characterized by heterogeneity (a mixture of structured and unstructured data) and multimodal features (text, numerical, and time-series data). This step involves three types: credit data, IoT device data, and social network data.

[0058] Credit data refers to authoritative credit records provided by credit reporting agencies, such as loan history, repayment status, and debt situation. Data sources may include credit reporting centers, third-party credit reporting agencies, etc. Credit reports of borrowers are periodically synchronized through API interfaces or data files (such as XML and JSON). Credit data includes: basic information, such as name, ID number, and occupation; credit records, such as loan amount, repayment status, and number of overdue payments; and public records, such as tax arrears and legal proceedings.

[0059] IoT device data refers to real-time behavioral data collected through sensors and smart terminals, such as GPS location, device usage time, temperature, and humidity. Data sources are sensors deployed on devices associated with the credit entity, such as smartphone GPS, vehicle OBD, and smart home devices. Device data is uploaded to the IoT platform in real time via MQTT and HTTP protocols. Specific IoT device data includes: device ID, geographical location (latitude and longitude), usage time, and abnormal events, such as device fault codes.

[0060] Social network data refers to user behavior and relationship data on social platforms, such as friend networks, interaction frequency, and consumption preferences. Friend relationship chains, interaction behaviors, and consumption records can be obtained through user-authorized APIs, or, under compliant conditions, through web scraping to crawl publicly available social graph data, such as the number of followers and community affiliation.

[0061] Preferably, preprocessing and standardization can also be performed on multi-source data, such as unifying field formats, filling in default values, removing outliers, and noise reduction.

[0062] Step 120: Based on the semantic features of each entity in the multi-source data, perform entity parsing and alignment of the multi-source data to construct an initial knowledge graph for the credit subject.

[0063] Specifically, in multi-source data, each entity refers to an object with a clear business meaning. For example, a customer entity is a credit applicant, which can be an individual or a company; a device entity is an IoT device associated with the customer, such as a smartphone or vehicle terminal; and a location entity is the geographical location involved in the customer's activities, such as a home address or workplace.

[0064] The semantic features of each entity refer to the semantic information contained within the entity in natural language or data fields. Semantic features can be implemented through text embedding based on the BERT model or through feature encoding.

[0065] Entity resolution refers to identifying entities in multi-source data that refer to the same real-world object, such as associating "Zhang San" in different systems as the same customer. Entity alignment refers to uniformly mapping the resolved entities to standardized nodes in a knowledge graph, eliminating the problem of homonyms.

[0066] This can be achieved through a pre-built semantic mapping rule base for collaborative entity alignment. The semantic mapping rule base contains alignment logic for 2000+ cross-system terms, such as defining "monthly income" as salary slips + e-commerce spending + tax data.

[0067] The resulting initial knowledge graph is a structured network centered on the credit entity and constructed through entity relationships such as "customer-device binding" and "device-location association," which is used to support subsequent credit assessment and risk analysis.

[0068] In some possible implementations, step 120 specifically includes:

[0069] Step 121: Extract the semantic features of each entity in the multi-source data;

[0070] Step 122: Treat each entity as a node, define the association edges between nodes according to business rules, and calculate the topological distance weight of the edges:

[0071] Step 123: Calculate the similarity between any two nodes based on the semantic features of each node and the topological distance weight of the edges.

[0072] Step 124: Merge the nodes based on the similarity between each pair of nodes and a preset similarity threshold.

[0073] Specifically, for text-type entities, a pre-trained BERT model is used to convert customer names, device ID descriptions, address text, etc., into high-dimensional semantic vectors to obtain semantic features.

[0074] For numerical feature encoding, it can be standardized and mapped to a low-dimensional vector through a fully connected layer.

[0075] For categorical feature encoding, it is converted into a dense vector through an embedding layer.

[0076] An entity relationship graph is constructed using a graph neural network, where each entity, such as customer, device, and location, is treated as a node in the graph. The node features are its semantic feature vectors (BERT vectors + numerical / categorical embedding vectors). Relationships between nodes are defined according to business rules, such as "customer-device" binding relationships and "device-location" usage records, and the topological distance weights of the edges are calculated.

[0077] For direct relationships, such as customer A being explicitly bound to device X, the topology distance weight can be set to w=1.0; for indirect relationships, such as customer A→device X→location Y, the topology distance weight can be decayed according to the path length, such as set to w=0.6.

[0078] Considering that the constructed entity relationship graph may contain similar entities, in order to improve the accuracy of cross-system entity matching and eliminate the problems of homonyms and synonyms, a unified credit subject profile is constructed, which can calculate the similarity between node pairs composed of every two nodes.

[0079] Here, the similarity between node pairs can be calculated by combining semantic similarity and topological distance weights. Semantic similarity can be obtained by calculating the cosine similarity of their BERT vectors; the higher the cosine similarity, the closer the semantics of the two nodes are.

[0080] In some embodiments, the similarity between every two nodes is determined based on the following formula:

[0081]

[0082] in, For nodes With nodes Similarity between them For nodes semantic features For nodes semantic features For similarity calculation function, For nodes With nodes The topological distance weight between edges.

[0083] After obtaining the similarity between any two nodes, similar nodes can be merged. The similarity threshold can be set based on business needs, combined with statistical distribution or expert experience. If the similarity is greater than or equal to the similarity threshold, the two nodes are determined to be the same entity, and a merge is performed. High-confidence attributes are inherited during the merge, such as updating customer addresses with the latest data. All related edges of the merged entity are retained, and the weight of directly related edges is increased. If the similarity is less than the similarity threshold, the two nodes are determined to be different entities and are retained independently. Furthermore, conflicting fields are resolved using a rule base or voting mechanism. This constructs an initial knowledge graph representing the profile of the credit entity.

[0084] Step 130: Based on the confidence scores of each data source in the initial knowledge graph and their corresponding time decay weights, the multi-source data is fused and conflict resolved through a spatiotemporal decay fusion model. Based on the fusion results, entity screening is performed on the initial knowledge graph to obtain an optimized knowledge graph.

[0085] Specifically, the node attributes in the initial knowledge graph have been preliminarily standardized through semantic parsing and alignment. Before executing step 130, the following steps are included:

[0086] Multi-source data is extracted from the initial knowledge graph, with each data source including a confidence score and a data timestamp; based on the data timestamp, a time decay weight is calculated for each data source using a time decay function.

[0087] In this step, the initial knowledge graph nodes are traversed, recording the multi-source data associated with each entity and its source. For example, customer A is associated with credit data, and device X is associated with IoT data. The entity-data source mapping in the initial knowledge graph is extracted. Each data source includes a confidence score and a data timestamp. The confidence score is a prediction reliability index output by each data source, with a value range of [0, 1], reflecting the reliability of the data. For example, the credit score confidence of credit data output by the credit model is 0.9, indicating high reliability; IoT data is scored by sensor data quality; and the confidence score of social network data is obtained by scoring the authenticity of user behavior data.

[0088] The time decay weight is a dynamic weight that measures the freshness of data, decaying exponentially with the age of the data. The formula for calculating the time decay weight can be expressed as:

[0089]

[0090] In the formula, Indicates the first The data sources are at the data age of Time decay weight, =0.85.

[0091] After obtaining the confidence scores and corresponding time decay weights of each data source, a spatiotemporal decay fusion model is used to fuse and resolve conflicts among the multi-source data. The spatiotemporal decay fusion model dynamically adjusts the contribution of multi-source data using time decay weights and confidence scores, and resolves the weighted fusion model for conflicting data, achieving the effect of recent high-confidence data dominating while the influence of conflicting data is dynamically suppressed. In some embodiments, step 130 specifically includes:

[0092] Step 131: Based on the confidence scores of each data source and their corresponding time decay weights, perform time-weighted fusion of the multi-source data to obtain the time fusion confidence score.

[0093] Step 132: Calculate the conflict resolution confidence based on the differences in time decay weights among the various data sources and the time fusion confidence.

[0094] Step 133: Combine the time fusion confidence and the conflict resolution confidence to obtain the final fusion confidence.

[0095] For each entity, calculate its weighted average confidence score from the multi-source data, i.e., the time fusion confidence score. This can be expressed as: Time Fusion Confidence Score = ,in For the first Confidence scores of each data source This represents the total number of data sources.

[0096] Through the product term To quantify the degree of data conflict, if the time decay weights of each data source are similar and the differences are small, then... The smaller the time decay weight, the lower the contribution of the conflict resolution term; if the time decay weights of different data sources differ significantly, then... Increase the weight of low-weight conflicting data.

[0097] The confidence level for conflict resolution is expressed by the formula:

[0098] Confidence level of conflict resolution =

[0099] The final fusion confidence is the sum of the temporal fusion confidence and the conflict resolution confidence. Based on this, entity filtering is performed on the initial knowledge graph according to the fusion result to obtain an optimized knowledge graph. First, a dynamic threshold is set, and entity nodes and their associated edges with a final fusion confidence less than the set dynamic threshold are deleted. The knowledge graph structure is then updated to obtain the optimized knowledge graph.

[0100] In this embodiment, time decay weights are used to give more decision-making power to recent data, adapting to dynamic environments such as sudden changes in customer credit status. Dynamic suppression of highly conflicting data interference through product terms reduces scoring errors. Confidence filtering and redundancy resolution improve the accuracy and consistency of the graph structure and attributes, supporting more precise risk control decisions.

[0101] Step 140: For each entity node in the optimized knowledge graph associated with a heterogeneous data source, call the scoring model that is adapted to its data features and generate a comprehensive credit score through the model integration mechanism.

[0102] Specifically, the optimized knowledge graph processed by the spatiotemporal decay fusion model contains high-confidence entity nodes and their relationships. The data quality and consistency have been significantly improved, and it can serve as a reliable basis for credit assessment.

[0103] Each entity node is associated with heterogeneous data sources characterized by data type heterogeneity, semantic heterogeneity, and temporal dynamics. Data types include structured, unstructured, and semi-structured data. Therefore, scoring models are designed for specific data source characteristics to assess the credit risk of lending entities. For example, credit reporting data models, based on logistic regression or XGBoost, calculate credit scores using historical loan records and repayment behavior. IoT data models use temporal convolutional networks (TCNs) or LSTMs to mine device usage patterns, such as active device periods and the frequency of anomalous events. Social network data models, based on graph neural networks (GNNs), mine social credit risk through friend relationships and interactive behaviors.

[0104] The outputs of multiple scoring models can be combined into a comprehensive credit score through weighting or other strategies to improve the comprehensiveness and robustness of the assessment.

[0105] In some embodiments, step 140 specifically includes:

[0106] Step 141: Based on the data type of each data source, call the scoring model that is compatible with the data type to obtain a separate credit score for each scoring model.

[0107] Step 142: Adjust the dynamic weights of each scoring model;

[0108] Step 143: Based on the adjusted dynamic weights, the individual credit scores are weighted and merged to obtain a comprehensive credit score.

[0109] Specifically, the data types for each data source include:

[0110] Structured data: in tabular form with clearly defined fields, such as "loan amount" and "repayment status" in credit data.

[0111] Unstructured data: without a fixed format, such as text (social media comments) and images (photos taken by devices).

[0112] Semi-structured data: somewhere in between, such as IoT logs in JSON format.

[0113] Time-series data: Data arranged in chronological order, such as temperature values ​​recorded by a device's sensors every minute.

[0114] Graph data: Relationship networks represented by nodes and edges, such as friend relationships in a social network.

[0115] Data type-adapted scoring models refer to predictive models designed for specific data types, used to assess the credit risk of lending entities, for example:

[0116] Logistic regression: suitable for linear risk prediction of structured data.

[0117] XGBoost / LightGBM: An efficient tree model for processing structured data.

[0118] Temporal Convolutional Networks (TCNs): Model local patterns in temporal data (such as device sensor logs).

[0119] Long Short-Term Memory (LSTM) networks: capture long-term dependencies in time-series data.

[0120] Graph Neural Networks (GNNs): Analyze the risks associated with node relationships in graph data.

[0121] The weighting coefficients, adjusted in real time based on the model's recent performance, such as AUC, KS value, and data quality, are used to control the contribution ratio of each model in the overall credit score, i.e., to adjust the dynamic weights of each scoring model.

[0122] In some embodiments, the performance metrics of each model on recent data are first calculated, including the AUC value: measuring the model's ability to distinguish between positive and negative samples, with higher values ​​indicating better performance; the KS value: evaluating the credit score's ability to distinguish between good and bad samples; and the KS drift metric: detecting the impact of changes in feature distribution on model performance. The dynamic weights of each scoring model comprehensively consider both model performance and data stability.

[0123] The dynamic weights of each scoring model can be adjusted based on the following formula:

[0124]

[0125] In the formula, For the scoring model Dynamic weights, For the scoring model of value, For all A rating model Sum of values This is a characteristic drift index.

[0126] The individual credit scores of each model are dynamically weighted and summed to obtain the comprehensive credit score.

[0127] In some embodiments, the scoring model can be iteratively optimized using incremental learning and quantum optimization techniques. The feature selection task is transformed into a quadratic unconstrained binary optimization (QUBO) problem, and the QUBO problem is solved efficiently using a quantum processor (such as a quantum annealing machine) to obtain the optimal feature subset, which significantly improves the feature selection speed and model iteration efficiency.

[0128] Step 150: Based on the comprehensive credit score, determine the risk control decision for the credit subject.

[0129] After obtaining the comprehensive credit score, risk control decisions are made for the borrower based on the comprehensive credit score. If the comprehensive credit score indicates low risk, the risk control decision may be automatic loan disbursement and dynamic increase of the loan amount; if the comprehensive credit score indicates medium risk, the risk control decision may be biometric authentication plus manual review; if the comprehensive credit score indicates high risk, the risk control decision may be blockchain blacklisting plus automatic circuit breaker for transaction links.

[0130] The method provided in this invention integrates heterogeneous data from multiple sources, including credit reporting, IoT devices, and social networks. Based on the BERT-Graph model, it achieves high-precision entity parsing and alignment to construct an initial knowledge graph. A spatiotemporal decay fusion model is used to dynamically adjust data weights and resolve conflicts. After filtering and optimizing the knowledge graph, an adaptive scoring model is called to generate a comprehensive credit score for heterogeneous data sources. Ultimately, this achieves accurate and efficient risk control decisions, significantly improving the accuracy, real-time performance, and robustness of risk identification, reducing false rejection rates and fraud losses, and enhancing the effectiveness of financial risk control.

[0131] It should be noted that the intranet (production data) and the extranet (Internet) of the edge computing platform can be separated by a security barrier to block external attacks and improve data security.

[0132] The following describes the credit business risk control decision-making device based on multi-source data fusion provided by the present invention. The credit business risk control decision-making device based on multi-source data fusion described below can be referred to in correspondence with the credit business risk control decision-making method based on multi-source data fusion described above.

[0133] Based on any of the above implementations, Figure 2 This is a schematic diagram of the structure of the credit business risk control decision-making device based on multi-source data fusion provided by the present invention, as shown below. Figure 2 As shown, the device includes:

[0134] The data acquisition unit 210 is used to acquire multi-source data related to the entity identifier of the credit entity, including credit data, Internet of Things device data and social network data;

[0135] The graph construction unit 220 is used to perform entity parsing and alignment of the multi-source data based on the semantic features of each entity in the multi-source data, so as to construct an initial knowledge graph for the credit subject.

[0136] The data fusion unit 230 is used to fuse and resolve the multi-source data based on the confidence scores of each data source in the initial knowledge graph and their corresponding time decay weights through a spatiotemporal decay fusion model, and to perform entity screening on the initial knowledge graph based on the fusion results to obtain an optimized knowledge graph.

[0137] Credit scoring unit 240 is used to call scoring models adapted to the data features of each heterogeneous data source associated with each entity node in the optimized knowledge graph, and generate a comprehensive credit score through a model integration mechanism.

[0138] The decision-making unit 250 is used to determine risk control decisions for the credit subject based on the comprehensive credit score.

[0139] Based on the above embodiments, the map construction unit is specifically used for:

[0140] Extract the semantic features of each entity in the multi-source data;

[0141] Treat each entity as a node, define the associated edges between nodes according to business rules, and calculate the topological distance weight of the edges:

[0142] Based on the semantic features of each node and the topological distance weight of the edges, the similarity between each pair of nodes is calculated.

[0143] Based on the similarity between each pair of nodes and a preset similarity threshold, the nodes are merged.

[0144] Based on the above embodiments, the map construction unit is specifically used for:

[0145] The similarity between any two nodes is determined using the following formula:

[0146]

[0147] in, For nodes With nodes Similarity between them For nodes semantic features For nodes semantic features For similarity calculation function, For nodes With nodes The topological distance weight between edges.

[0148] Based on the above embodiments, the data fusion unit is specifically used for:

[0149] Based on the confidence scores of each data source and their corresponding time decay weights, the multi-source data is time-weightedly fused to obtain the time fusion confidence score.

[0150] Based on the differences in time decay weights among the various data sources and the time fusion confidence, the conflict resolution confidence is calculated.

[0151] The time fusion confidence score is fused with the conflict resolution confidence score to obtain the final fusion confidence score.

[0152] Based on the above embodiments, the data fusion unit is specifically used for:

[0153] Extract multi-source data from the initial knowledge graph, with each data source including a confidence score and a data timestamp;

[0154] Based on the data timestamp, a time decay weight is calculated for each data source based on the time decay function.

[0155] Based on the above embodiments, the credit scoring unit is specifically used for:

[0156] Based on the data type of each data source, the scoring model adapted to the data type is called to obtain a separate credit score for each scoring model.

[0157] Adjust the dynamic weights of each scoring model;

[0158] The individual credit scores are weighted and fused based on the adjusted dynamic weights to obtain the comprehensive credit score.

[0159] Based on the above embodiments, the credit scoring unit is specifically used for:

[0160] The dynamic weights of each scoring model are adjusted based on the following formula:

[0161]

[0162] In the formula, For the scoring model Dynamic weights, For the scoring model of value, For all A rating model Sum of values This is a characteristic drift index.

[0163] Figure 3 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 3 As shown, the electronic device may include: a processor 310, a communications interface 320, a memory 330, and a communications bus 340, wherein the processor 310, the communications interface 320, and the memory 330 communicate with each other through the communications bus 340. The processor 310 can call logical instructions in the memory 330 to execute a credit business risk control decision-making method based on multi-source data fusion. This method includes: acquiring multi-source data related to the entity identifier of the credit subject, including credit data, IoT device data, and social network data; performing entity parsing and alignment on the multi-source data based on the semantic features of each entity in the multi-source data to construct an initial knowledge graph for the credit subject; fusing and resolving conflicts of the multi-source data using a spatiotemporal decay fusion model based on the confidence scores and corresponding time decay weights of each data source in the initial knowledge graph; filtering entities in the initial knowledge graph based on the fusion results to obtain an optimized knowledge graph; calling scoring models adapted to the data features of each heterogeneous data source associated with each entity node in the optimized knowledge graph, and generating a comprehensive credit score through a model integration mechanism; and determining a risk control decision for the credit subject based on the comprehensive credit score.

[0164] Furthermore, the logical instructions in the aforementioned memory 330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0165] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the credit business risk control decision-making method based on multi-source data fusion provided by the above methods. The method includes: acquiring multi-source data related to the subject identifier of the credit subject, the multi-source data including credit data, IoT device data, and social network data; performing entity parsing and alignment on the multi-source data based on the semantic features of each entity in the multi-source data to construct an initial knowledge graph for the credit subject; fusing and resolving conflicts of the multi-source data through a spatiotemporal decay fusion model based on the confidence scores of each data source in the initial knowledge graph and their corresponding time decay weights, and performing entity screening on the initial knowledge graph based on the fusion results to obtain an optimized knowledge graph; calling a scoring model adapted to the data features of each heterogeneous data source associated with each entity node in the optimized knowledge graph, and generating a comprehensive credit score through a model integration mechanism; and determining a risk control decision for the credit subject based on the comprehensive credit score.

[0166] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program is implemented to perform the credit business risk control decision-making method based on multi-source data fusion provided by the above methods. The method includes: acquiring multi-source data related to the entity identifier of the credit subject, wherein the multi-source data includes credit data, IoT device data, and social network data; performing entity parsing and alignment on the multi-source data based on the semantic features of each entity in the multi-source data to construct an initial knowledge graph for the credit subject; fusing and resolving conflicts of the multi-source data through a spatiotemporal decay fusion model based on the confidence scores of each data source in the initial knowledge graph and their corresponding time decay weights; performing entity screening on the initial knowledge graph based on the fusion results to obtain an optimized knowledge graph; calling a scoring model adapted to the data features of each heterogeneous data source associated with each entity node in the optimized knowledge graph, and generating a comprehensive credit score through a model integration mechanism; and determining a risk control decision for the credit subject based on the comprehensive credit score.

[0167] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0168] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0169] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A credit business risk control decision-making method based on multi-source data fusion, characterized in that, include: Acquire multi-source data related to the entity identifier of the credit entity, including credit data, IoT device data, and social network data; Based on the semantic features of each entity in the multi-source data, entity parsing and alignment are performed on the multi-source data to construct an initial knowledge graph for the credit subject; Based on the confidence scores of each data source in the initial knowledge graph and their corresponding time decay weights, the multi-source data is fused and conflict resolved through a spatiotemporal decay fusion model. Based on the fusion results, entity filtering is performed on the initial knowledge graph to obtain an optimized knowledge graph. For each entity node in the optimized knowledge graph associated with a heterogeneous data source, a scoring model adapted to its data features is invoked, and a comprehensive credit score is generated through a model integration mechanism. Based on the comprehensive credit score, risk control decisions are determined for the credit entity; The process of fusing and resolving conflicts of the multi-source data using a spatiotemporal decay fusion model, based on the confidence scores of each data source in the initial knowledge graph and their corresponding time decay weights, includes: Based on the confidence scores of each data source and their corresponding time decay weights, the multi-source data is time-weightedly fused to obtain a time fusion confidence score; based on the differences between the time decay weights of each data source and the time fusion confidence score, a conflict resolution confidence score is calculated; the time fusion confidence score and the conflict resolution confidence score are fused to obtain the final fusion confidence score. The process involves calling a scoring model adapted to the data features of each heterogeneous data source associated with each entity node in the optimized knowledge graph, and generating a comprehensive credit score through a model integration mechanism, including: Based on the data type of each data source, a scoring model adapted to the data type is invoked to obtain an individual credit score for each scoring model; the dynamic weights of each scoring model are adjusted; based on the adjusted dynamic weights, the individual credit scores are weighted and fused to obtain the comprehensive credit score.

2. The credit business risk control decision-making method based on multi-source data fusion according to claim 1, characterized in that, The entity parsing and alignment of the multi-source data based on the semantic features of each entity in the multi-source data includes: Extract the semantic features of each entity in the multi-source data; Treat each entity as a node, define the associated edges between nodes according to business rules, and calculate the topological distance weight of the edges: Based on the semantic features of each node and the topological distance weight of the edges, the similarity between each pair of nodes is calculated. Based on the similarity between each pair of nodes and a preset similarity threshold, the nodes are merged.

3. The credit business risk control decision-making method based on multi-source data fusion according to claim 2, characterized in that, The similarity between any two nodes is calculated based on the semantic features of each node and the topological distance weights of the edges, including: The similarity between any two nodes is determined using the following formula: in, For nodes With nodes Similarity between them For nodes semantic features For nodes semantic features For similarity calculation function, For nodes With nodes The topological distance weight between edges.

4. The credit business risk control decision-making method based on multi-source data fusion according to claim 1, characterized in that, Before performing time-weighted fusion of the multi-source data based on the confidence scores of each data source and their corresponding time decay weights, the method further includes: Extract multi-source data from the initial knowledge graph, with each data source including a confidence score and a data timestamp; Based on the data timestamp, a time decay weight is calculated for each data source based on the time decay function.

5. The credit business risk control decision-making method based on multi-source data fusion according to claim 1, characterized in that, The adjustment of the dynamic weights for each scoring model includes: The dynamic weights of each scoring model are adjusted based on the following formula: In the formula, For the scoring model Dynamic weights, For the scoring model of value, For all A rating model Sum of values This is a characteristic drift index.

6. A credit business risk control decision-making device based on multi-source data fusion, characterized in that, include: The data acquisition unit is used to acquire multi-source data related to the entity identifier of the credit entity, including credit data, Internet of Things device data, and social network data. The graph construction unit is used to perform entity parsing and alignment of the multi-source data based on the semantic features of each entity in the multi-source data, so as to construct an initial knowledge graph for the credit subject. The data fusion unit is used to fuse and resolve conflicts of the multi-source data based on the confidence scores of each data source in the initial knowledge graph and their corresponding time decay weights through a spatiotemporal decay fusion model, and to perform entity screening on the initial knowledge graph based on the fusion results to obtain an optimized knowledge graph. The credit scoring unit is used to call a scoring model that is adapted to the data features of each entity node associated with the heterogeneous data source in the optimized knowledge graph, and generate a comprehensive credit score through the model integration mechanism. The decision-making unit is used to determine risk control decisions for the credit subject based on the comprehensive credit score. The data fusion unit is specifically used for: Based on the confidence scores of each data source and their corresponding time decay weights, the multi-source data is time-weightedly fused to obtain a time fusion confidence score; based on the differences between the time decay weights of each data source and the time fusion confidence score, a conflict resolution confidence score is calculated; the time fusion confidence score and the conflict resolution confidence score are fused to obtain the final fusion confidence score. The credit scoring unit is specifically used for: Based on the data type of each data source, the scoring model adapted to the data type is called to obtain a separate credit score for each scoring model; the dynamic weights of each scoring model are then adjusted. The individual credit scores are weighted and fused based on the adjusted dynamic weights to obtain the comprehensive credit score.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the credit business risk control decision-making method based on multi-source data fusion as described in any one of claims 1 to 5.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the credit business risk control decision-making method based on multi-source data fusion as described in any one of claims 1 to 5.

Images