Risk control credit assessment method and system

Abstract

The embodiment of the invention provides a risk control credit assessment method and system. The method comprises the steps of collecting multi-modal data of a user, performing deep semantic feature extraction on the preprocessed multi-modal data to generate a multi-dimensional semantic feature, generating a knowledge graph feature based on the multi-modal data of the user by adopting a financial knowledge domain graph, and extracting the knowledge graph feature based on the multi-modal data of the user. And fusing the multi-dimensional semantic features and the knowledge graph features by adopting a cross-modal alignment mechanism to generate fusion features, performing semantic enhancement and risk tag association on the fusion features, training a risk control credit extension model based on the processed fusion features, and performing risk control credit extension on the risk control credit extension model. And evaluating a risk score and a credit line of a to-be-tested user based on the risk control credit model. In this way, the risk control credit granting accuracy can be improved.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, and more particularly to the fields of artificial intelligence and financial technology. Background Technology

[0002] Traditional risk control models primarily rely on structured data (such as credit reports, income statements, and historical borrowing records) for credit assessment. However, these methods suffer from limitations such as limited data dimensions and reliance on manual feature extraction. Summary of the Invention

[0003] This disclosure provides a risk control credit assessment method and system.

[0004] According to a first aspect of this disclosure, a risk control credit assessment method is provided. The method includes:

[0005] Collect user multimodal data, which includes at least two of the following: text data, image data, voice data, and structured data. The structured data includes at least one of the following: credit report, transaction record, and device fingerprint.

[0006] Deep semantic feature extraction is performed on the preprocessed multimodal data to generate multidimensional semantic features. The preprocessing includes cleaning, normalization, desensitization, and format unification.

[0007] A knowledge graph feature is generated based on the user's multimodal data using a financial knowledge domain graph.

[0008] A cross-modal alignment mechanism is used to fuse the multidimensional semantic features and the knowledge graph features to generate fused features;

[0009] The fused features are semantically enhanced and associated with risk labels;

[0010] The risk control and credit granting model is trained based on the processed fusion features;

[0011] The risk score and credit limit of the user to be tested are evaluated based on the risk control and credit granting model.

[0012] According to a second aspect of this disclosure, a risk control credit assessment system is provided. The system includes:

[0013] The user data acquisition module collects multimodal data of users, including at least two of the following: text data, image data, voice data, and structured data. The structured data includes at least one of the following: credit report, transaction record, and device fingerprint.

[0014] The semantic feature extraction module performs deep semantic feature extraction on the preprocessed multimodal data to generate multidimensional semantic features. The preprocessing includes cleaning, normalization, desensitization, and format unification.

[0015] The knowledge graph feature generation module uses a financial knowledge domain graph to generate knowledge graph features based on the user's multimodal data;

[0016] The fusion feature generation module uses a cross-modal alignment mechanism to fuse the multidimensional semantic features and the knowledge graph features to generate fusion features;

[0017] The enhancement and risk association module performs semantic enhancement and associates the fused features with risk labels.

[0018] The risk control and credit granting model training module trains the risk control and credit granting model based on the processed fusion features.

[0019] The risk control and credit assessment module evaluates the user's risk score and credit limit based on the risk control and credit assessment model.

[0020] According to a third aspect of this disclosure, an electronic device is provided. The electronic device includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the program to implement the method described above.

[0021] According to a fourth aspect of this disclosure, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the methods according to the first and / or second aspects of this disclosure.

[0022] It should be understood that the description in the Summary of the Invention is not intended to limit the key or essential features of the embodiments of this disclosure, nor is it intended to restrict the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0023] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. The drawings are provided for a better understanding of the invention and are not intended to limit the scope of this disclosure. In the drawings, the same or similar reference numerals denote the same or similar elements, wherein:

[0024] Figure 1 A flowchart of a risk control credit assessment method according to an embodiment of the present disclosure is shown;

[0025] Figure 2 A block diagram of a risk control credit assessment system according to an embodiment of the present disclosure is shown;

[0026] Figure 3A block diagram of an electronic device used to implement the risk control credit assessment method of the present disclosure is shown. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0028] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0029] In recent years, with breakthroughs in large models in natural language processing, computer vision, and speech recognition, their powerful feature extraction and cross-modal understanding capabilities have provided new technical pathways for risk control and credit granting. Meanwhile, knowledge graphs, as a structured semantic knowledge base, can effectively represent complex relationships between entities, providing rich background knowledge for risk control models. This disclosure deeply integrates multimodal large models with knowledge graphs: input text, images, speech, and structured data are uniformly embedded and jointly reasoned with graph relationships to output risk scores, credit limits, and interpretable labels, achieving highly accurate and traceable risk control and credit granting.

[0030] Figure 1 A flowchart of a risk control credit assessment method 100 according to an embodiment of the present disclosure is shown. Method 100 may include:

[0031] In box 110, multimodal data of the user is collected. The multimodal data includes at least two of the following: text data, image data, voice data, and structured data. The structured data includes at least one of the following: credit report, transaction record, and device fingerprint.

[0032] In box 120, deep semantic features are extracted from the preprocessed multimodal data to generate multidimensional semantic features. The preprocessing includes cleaning, normalization, desensitization, and format unification.

[0033] In box 130, a knowledge graph feature is generated based on the user's multimodal data using a financial knowledge domain graph.

[0034] In box 140, a cross-modal alignment mechanism is used to fuse the multidimensional semantic features and the knowledge graph features to generate fused features. A contrastive learning or cross-attention mechanism is employed to align the heterogeneous modal vectors and the knowledge graph vectors to the same semantic space. The fused vector serves as the sole input to the end-to-end risk control and credit granting model, directly outputting the risk score and credit limit.

[0035] In box 150, the fused features are semantically enhanced and associated with risk labels.

[0036] In box 160, the risk control and credit granting model is trained based on the processed fusion features.

[0037] In box 170, the risk score and credit limit of the user under test are evaluated based on the risk control and credit granting model. Taking the user's integrated characteristics as input, the risk control and credit granting model outputs the user's risk score and credit limit based on the customer's profile characteristics.

[0038] Furthermore, the risk control and credit granting model supports online learning and model update mechanisms to adapt to changes in data distribution.

[0039] First, semantic features are extracted from unstructured data using large text / visual / speech models; simultaneously, a knowledge graph in the financial field is constructed, and high-order association information is propagated on the graph through graph neural networks to generate graph-enhanced features; the two types of features are mapped to a unified embedding space and then fused to form a comprehensive representation for risk control and credit granting.

[0040] This disclosure provides a risk control and credit assessment method that combines feature mining from multimodal large models and knowledge graphs. It solves the problems of low data utilization, weak feature extraction capability, poor model generalization capability, and insufficient application of knowledge graphs in existing technologies, and achieves more accurate, efficient, and interpretable risk control and credit assessment decisions.

[0041] In some embodiments, in block 120, deep semantic feature extraction is performed on the preprocessed multimodal data, which may specifically include:

[0042] The semantic features of the text data are extracted using large language models, such as LLaMA (short for Large Language Model Meta AI) and ChatGLM (short for Chat Generative Language Model).

[0043] Visual features of the image data are extracted using large visual models, such as ViT (Vision Transformer) and Swin Transformer (Shifted Window Transformer).

[0044] Large speech models are used to extract the voiceprint and semantic features of the speech data, such as Whisper (speech-to-text model) and WavLM (Waveform Language Model). Whisper is an end-to-end Transformer speech processing model open-sourced by OpenAI in 2022. Its core is used for automatic speech recognition (ASR), multilingual speech-to-text, speech translation (to English), and language detection. It has strong generalization ability in noisy environments and with diverse accents and has become a general basic tool in the field of speech processing.

[0045] The structured features of the structured data are extracted by an encoder, such as TabNet (short for Tabular Network) or FT-Transformer (Feature Tokenizer Transformer) models.

[0046] According to embodiments of this disclosure, pre-trained multimodal large models (such as CLIP (Contrastive Language-Image Pre-training), BLIP (Bootstrapping Language-Image Pre-training), Whisper, LLaVA (Large Language and Vision Assistant), etc.) are used to extract deep semantic features from unstructured data and generate high-dimensional semantic vector representations.

[0047] In some embodiments, method 100 may further include:

[0048] Determine the entity type and the association relationship of the entity, wherein the entity includes natural persons, enterprises, equipment, addresses and telephone numbers, and the association relationship includes: co-address, co-number, co-equipment and financial transactions;

[0049] Access to multimodal data, including: core structured bank transaction records, image ID cards, facial images, telephone recordings, and credit reports;

[0050] Based on the multimodal data, entities and relationships are extracted to construct a financial knowledge domain graph.

[0051] According to embodiments of this disclosure, the embodiments may further include: constructing a knowledge graph in the financial field, which includes entities such as users, enterprises, devices, addresses, and telephone numbers and their related relationships (such as shared address, shared number, shared device, fund transfers, etc.).

[0052] The construction of a financial knowledge domain graph follows these steps:

[0053] 1. Identify the core entities: five categories, namely users, enterprises, devices, addresses, and telephone numbers, and standardize their IDs.

[0054] 2. Access to multimodal data: core structured bank statements, OCR image ID cards, facial images, telephone recordings, and credit report PDF texts.

[0055] 3. Directly map fields to structured tables; extract entities from images, audio, and text using ViT / Whisper / BERT (Bidirectional Encoder Representations from Transformers) to generate "person-company-amount-date" triples.

[0056] 4. Establish ontology: Define the attributes, unique key, and timestamp of each type of entity, and define weighted edges such as co-address, co-number, co-device, and financial transactions.

[0057] 5. Entity alignment: Precisely merge using ID card / credit code, automatically normalize using name + address fuzzy similarity ≥ 0.9, and resolve conflicts based on time priority.

[0058] 6. Graph database storage: Vertices contain de-identified attributes, edges contain amount, number of transactions, and start date, and a joint index of address and telephone number is established.

[0059] 7. Real-time incremental processing: Kafka collects new data streams, OCR images, and speech-to-text data, which are then written to the graph by Flink with a latency of <5 seconds. Kafka, short for Apache Kafka, is a distributed stream processing platform developed by the Apache Software Foundation. Its core function is to provide high-throughput, low-latency message queues and real-time data pipelines, and it is widely used for the real-time acquisition, transmission, storage, and processing of large-scale data.

[0060] 8. Implanted Graph Algorithms: Louvain (community detection algorithm) for identifying groups, 3-step chordal detection for fund loops, and Personalized-PageRank for calculating the probability of default propagation. Personalized-PageRank (PPR) is an extended variant of the classic PageRank algorithm. Its core is to calculate the personalized relevance score of a node relative to a specific starting point / interest set through biased random walks on the graph structure, rather than global importance. It is a core algorithm in the field of graph analysis and recommendation.

[0061] 9. Output interface: gRPC obtains the user's 2-3 hop neighbor subgraph, REST returns the closed-loop path, which is then called by the risk control engine.

[0062] gRPC is short for gRPC Remote Procedure Call, a high-performance, cross-language remote procedure call (RPC) framework open-sourced by Google. It is designed based on the HTTP / 2 protocol and is primarily used for efficient communication between different services / systems, especially suitable for cross-service call scenarios in microservice architectures.

[0063] REST is an abbreviation for Representational State Transfer, a software architectural style (rather than a standard or protocol) used to design web application programming interfaces (i.e., RESTful APIs). Its core principle follows a resource-oriented approach, enabling client-server communication in a concise and scalable manner.

[0064] According to embodiments of this disclosure, constructing a financial knowledge domain graph using the above methods can improve the accuracy of feature extraction.

[0065] In some embodiments, generating knowledge graph features based on the user's multimodal data using a financial knowledge domain graph includes:

[0066] By using graph neural networks to perform neighbor sampling, message passing, and attention pooling on the financial knowledge domain graph, information on default, funding loops, and group structure is embedded and aggregated into user nodes, and the higher-order correlation features of the user are output.

[0067] Graph Neural Networks (GNNs) are used for message propagation on knowledge graphs to mine higher-order association features of users. These higher-order association features include:

[0068] ①Statistics on "Default Rate / Overdue Amount" for steps 2-3;

[0069] ② Funding loop ≤ 3 steps, guarantee loop length;

[0070] ③ Core coefficient of the group and k-core size; k-core size refers to the total number of nodes contained in the k-core subgraph of the network. That is, after recursively deleting nodes with a degree less than k, the number of nodes remaining in the maximal subgraph where each node has a degree greater than or equal to k is the final number of nodes. It is a key indicator for measuring the tightness of the network core and the stability of the structure.

[0071] ④ PageRank - Probability of Default Propagation; PageRank - Probability of Default Propagation is a random walk logic based on PageRank (including personalized variants). In the user-account-transaction relationship graph, it quantifies the probability of default risk being transmitted from a known default node to other nodes through network links. Its core is to map the PageRank value of a node to the probability of exposure to default risk, while combining damping factors and edge weights to control the propagation strength and path decay.

[0072] ⑤ Structural indicators such as the number of multiple paths and the clustering coefficient.

[0073] Using a 3-layer GraphSAGE (short for Graph Sample and AggreGatE) to perform neighbor sampling, message passing, and attention pooling on the knowledge graph, multi-hop risks are embedded and aggregated into user nodes, and the output is an interpretable high-order feature vector.

[0074] Specifically, the following steps are used to mine features in graph neural networks:

[0075] 1. Sampling: Extract a 3-hop subgraph centered on the user, retaining node attributes and edge types.

[0076] 2. Encoding: Initial node embedding = attribute vector + relation type vector + hop count position encoding.

[0077] 3. Propagation: Aggregate by "neighbor → self", compressing information on defaults, funding loops, and gang structures of 2-3 hops to user nodes layer by layer.

[0078] 4. Pooling: Apply Attention-Pool to multi-layer embeddings to output a unified high-order vector. Attention-Pool is a feature aggregation method that combines attention mechanisms with traditional pooling operations. Its core is to assign dynamic weights to different input features, allowing the model to adaptively focus on key information and weaken irrelevant noise, ultimately generating more representative aggregated features.

[0079] According to embodiments of this disclosure, a knowledge graph in the financial field is constructed simultaneously, and high-order association information is propagated on the graph through a graph neural network to generate graph enhancement features, which are used to improve the diversity and accuracy of training data.

[0080] In some embodiments, the fusion of the multidimensional semantic features and the knowledge graph features using a cross-modal alignment mechanism includes:

[0081] The multidimensional semantic features and the knowledge graph features are mapped to a unified semantic space to achieve heterogeneous information fusion.

[0082] Deep aggregation of heterogeneous information is achieved by sharing semantic space and gating residuals.

[0083] In this embodiment, cross-modal feature fusion and alignment are achieved. Specifically, cross-modal alignment mechanisms (such as contrastive learning, attention mechanisms, and joint embedding spaces) are used to map features from different modalities and knowledge graph features to a unified semantic space. Then, cross-attention is used to weight the "content" key value of the "relationship" query. After gating the residual, a unified fusion feature that takes into account both "relationship" and "content" is obtained, thus realizing heterogeneous information fusion. The principle of heterogeneous information fusion is to first encode four types of heterogeneous data—text, image, speech, and structured data—into vectors of the same dimension through their respective large models, and then feed them together with the relation vectors output by the Graph Neural Network (GNN) into the cross-modal attention layer. By sharing the semantic space and gating residuals, deep aggregation of heterogeneous information is achieved, and consistent fusion features are output for subsequent decision-making.

[0084] According to an embodiment of this disclosure, in this embodiment, the map features are spliced ​​with or attention-fused with user multimodal features to achieve feature enhancement and improve the accuracy of training data.

[0085] In some embodiments, method 100 may further include: the risk control credit granting model outputting a key feature attribution analysis and risk interpretation report.

[0086] According to embodiments of this disclosure, during the credit decision-making stage, key features and knowledge graph paths are located using the SHAP / LIME attribution algorithm to generate a natural language explanation report. The system then feeds back the decision results and subsequent real default labels to the graph neural network, achieving a closed loop of "decision-feedback-graph update." A visual interface is provided for manual review and rule intervention, meeting regulatory auditability requirements and thus improving interpretability. SHAP is an abbreviation for Shapley Additive exPlanations, a model interpretability algorithm based on the "Shapley Value" in game theory. LIME is an abbreviation for Local Interpretable Model-agnostic Explanations, a model-agnostic machine learning interpretability algorithm. Its core logic is that a complex "black box" model can be approximated by a simple linear model within the local neighborhood of a single predicted sample, and the decision logic of the original model for that sample is explained by the coefficients of the linear model.

[0087] In some embodiments, method 100 may further include: obtaining decision results and user feedback, and optimizing the risk control credit granting model.

[0088] According to embodiments of this disclosure, decision results and user feedback are recorded for iterative model optimization, thereby improving the accuracy of the risk control and credit granting model.

[0089] According to the embodiments of this disclosure, the following technical effects are achieved:

[0090] 1. Improve credit granting accuracy: By deeply mining potential risk characteristics of users through multimodal large models and combining high-order correlation information from knowledge graphs, the ability to identify defaults is significantly improved.

[0091] 2. Enhanced generalization ability: Large models have strong transfer capabilities, and knowledge graphs provide rich background knowledge to adapt to different regions, groups, and product scenarios.

[0092] 3. Reduce labor costs: Automated feature extraction and fusion reduce reliance on manual rules and feature engineering.

[0093] 4. Improve interpretability: Provide visualized decision-making paths and feature attribution to meet regulatory and audit requirements.

[0094] 5. Supports real-time decision-making: The system supports millisecond-level response, making it suitable for online real-time credit granting scenarios.

[0095] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this disclosure is not limited to the described order of actions, because according to this disclosure, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to this disclosure.

[0096] The above is an introduction to the method embodiments. The following system embodiments will further illustrate the solution described in this disclosure.

[0097] Figure 2 A block diagram of a risk control credit assessment system 200 according to an embodiment of the present disclosure is shown. Figure 2 As shown, system 200 includes:

[0098] User data acquisition module 210 collects multimodal data of users, including at least two of the following: text data, image data, voice data, and structured data. The structured data includes at least one of the following: credit report, transaction record, and device fingerprint.

[0099] The semantic feature extraction module 220 performs deep semantic feature extraction on the preprocessed multimodal data to generate multidimensional semantic features. The preprocessing includes cleaning, normalization, desensitization and format unification.

[0100] The knowledge graph feature generation module 230 uses a financial knowledge domain graph to generate knowledge graph features based on the user's multimodal data;

[0101] The fusion feature generation module 240 uses a cross-modal alignment mechanism to fuse the multidimensional semantic features and the knowledge graph features to generate fusion features;

[0102] The enhancement and risk association module 250 performs semantic enhancement and association of the fused features with risk labels;

[0103] The risk control and credit granting model training module 260 trains the risk control and credit granting model based on the processed fusion features.

[0104] The risk control and credit assessment module 270 assesses the user's risk score and credit limit based on the risk control and credit assessment model.

[0105] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the described module can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0106] The acquisition, storage, and application of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0107] According to embodiments of this disclosure, this disclosure also provides an electronic device and a readable storage medium.

[0108] The electronic device includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the program to implement the method described above.

[0109] The readable storage medium stores a computer program that, when executed by a processor, implements the method described above.

[0110] Figure 3 A schematic block diagram of an electronic device 800 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0111] Device 800 includes a computing unit 801, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 802 or a computer program loaded from storage unit 808 into random access memory (RAM) 803. RAM 803 may also store various programs and data required for the operation of device 800. The computing unit 801, ROM 802, and RAM 803 are interconnected via bus 804. Input / output (I / O) interface 805 is also connected to bus 804.

[0112] Multiple components in device 800 are connected to I / O interface 805, including: input unit 806, such as keyboard, mouse, etc.; output unit 807, such as various types of monitors, speakers, etc.; storage unit 808, such as disk, optical disk, etc.; and communication unit 809, such as network card, modem, wireless transceiver, etc. Communication unit 809 allows device 800 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0113] The computing unit 801 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 801 performs the various methods and processes described above, such as risk control credit assessment methods. For example, in some embodiments, the risk control credit assessment method can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 808. In some embodiments, part or all of the computer program can be loaded and / or installed on device 800 via ROM 802 and / or communication unit 809. When the computer program is loaded into RAM 803 and executed by the computing unit 801, one or more steps of the risk control credit assessment method described above can be performed. Alternatively, in other embodiments, the computing unit 801 can be configured to perform risk control credit assessment methods by any other suitable means (e.g., by means of firmware).

[0114] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0115] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0116] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0117] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0118] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0119] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0120] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0121] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A risk control and credit assessment method, comprising: Collect user's multimodal data, which includes at least two of the following: text data, image data, voice data, and structured data. The structured data includes at least one of the following: credit report, transaction record, and device fingerprint. Deep semantic feature extraction is performed on the preprocessed multimodal data to generate multidimensional semantic features. The preprocessing includes cleaning, normalization, desensitization, and format unification. A knowledge graph feature is generated based on the user's multimodal data using a financial knowledge domain graph. A cross-modal alignment mechanism is used to fuse the multidimensional semantic features and the knowledge graph features to generate fused features; The fused features are semantically enhanced and associated with risk labels; The risk control and credit granting model is trained based on the processed fusion features; The risk score and credit limit of the user to be tested are evaluated based on the risk control and credit granting model.

2. The method according to claim 1, wherein, The deep semantic feature extraction of the preprocessed multimodal data includes: The semantic features of the text data are extracted using a large language model; Visual features of the image data are extracted using a large visual model. The large speech model is used to extract the voiceprint and semantic features of the speech data; The structured features of the structured data are extracted using an encoder.

3. The method according to claim 1, further comprising: Determine the entity type and the association relationship of the entity, wherein the entity includes natural persons, enterprises, equipment, addresses and telephone numbers, and the association relationship includes: co-address, co-number, co-equipment and financial transactions; Access to multimodal data, including: core structured bank transaction records, image ID cards, facial images, telephone recordings, and credit reports; Based on the multimodal data, entities and relationships are extracted to construct a financial knowledge domain graph.

4. The method according to claim 1, wherein, The method of generating knowledge graph features based on the user's multimodal data using a financial knowledge domain graph includes: By using graph neural networks to perform neighbor sampling, message passing, and attention pooling on the financial knowledge domain graph, information on default, funding loops, and group structure is embedded and aggregated into user nodes, and the higher-order correlation features of the user are output.

5. The method according to claim 1, wherein, The method of fusing the multidimensional semantic features and the knowledge graph features using a cross-modal alignment mechanism includes: The multidimensional semantic features and the knowledge graph features are mapped to a unified semantic space to achieve heterogeneous information fusion. Deep aggregation of heterogeneous information is achieved by sharing semantic space and gating residuals.

6. The method according to claim 1, further comprising: The risk control and credit granting model outputs key feature attribution analysis and risk interpretation reports.

7. The method according to claim 1, further comprising: The decision-making results and user feedback are obtained to optimize the risk control and credit granting model.

8. A risk control and credit assessment system, comprising: The user data acquisition module collects multimodal data of users, including at least two of the following: text data, image data, voice data, and structured data. The structured data includes at least one of the following: credit report, transaction record, and device fingerprint. The semantic feature extraction module performs deep semantic feature extraction on the preprocessed multimodal data to generate multidimensional semantic features. The preprocessing includes cleaning, normalization, desensitization, and format unification. The knowledge graph feature generation module uses a financial knowledge domain graph to generate knowledge graph features based on the user's multimodal data; The fusion feature generation module uses a cross-modal alignment mechanism to fuse the multidimensional semantic features and the knowledge graph features to generate fusion features; The enhancement and risk association module performs semantic enhancement and associates the fused features with risk labels. The risk control and credit granting model training module trains the risk control and credit granting model based on the processed fusion features. The risk control and credit assessment module evaluates the user's risk score and credit limit based on the risk control and credit assessment model.

9. An electronic device, comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.

10. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-7.

11. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1-7.

Images