Bank flow risk assessment method and application system

Abstract

The invention discloses a bank flow risk assessment method and an application system. The method comprises the following steps: acquiring first multi-modal data; wherein the first multi-modal data comprises a bank statement, equipment information corresponding to the bank statement, a user portrait corresponding to the equipment information, a social network relationship and a geographic position; inputting the first multi-modal data into a risk scoring model to obtain a risk level corresponding to the first multi-modal data; and performing risk monitoring according to the risk level. In this way, the accuracy and efficiency of a financial institution for identifying the enterprise fund flow state and the potential risk can be improved.

Description

Technical Field

[0001] This application relates to the technical field of bank statement risk assessment, and particularly to a bank statement risk assessment method and an application system. Background Art

[0002] Bank statement data is important information reflecting the economic activities of enterprises and individuals. For financial institutions including banks, investment banks, securities firms, and auditing firms, by analyzing bank statement data, they can comprehensively understand the operating conditions, profitability, income and expenditure of enterprises, identify potential improper behaviors of the target company, and effectively identify potential risks.

[0003] Currently, in due diligence reviews, when conducting risk assessments, the following methods are mainly relied on:

[0004] (1) Expert scoring method: relying on expert experience, there are problems such as strong subjectivity and poor consistency;

[0005] (2) Credit scoring model: building a model based on historical data, but it is difficult to cover all risk scenarios;

[0006] (3) Financial analysis method: focusing on the financial status of enterprises, but it is difficult to reflect real-time risks.

[0007] Characteristics of bank statement data: Bank statement data has the following characteristics:

[0008] (1) Real-time: reflecting the real-time economic activities of enterprises and individuals;

[0009] (2) Comprehensiveness: covering multiple accounts and various transaction types;

[0010] (3) Relevance: being related to other financial data of enterprises, personal credit status, etc.

[0011] The existing technologies have the following limitations in bank statement risk assessment:

[0012] (1) Lack of an assessment method targeting the characteristics of bank statement data;

[0013] (2) Bank diversity, making it difficult to adapt to and collect all bank statement formats;

[0014] (3) Bank institutions cannot query the statements of other banks interconnectedly. All are provided by the enterprises themselves, making it difficult to verify the integrity and authenticity of the data;

[0015] (4) Unable to effectively identify abnormal transactions and potential risks. Summary of the Invention

[0016] The bank statement risk assessment method and application system provided by this application can improve the accuracy and efficiency of financial institutions in identifying the status of corporate cash flows and potential risks.

[0017] In a first aspect, this application provides a bank statement risk assessment method, which includes: obtaining first multi-modal data; wherein, the first multi-modal data includes bank statements, device information corresponding to the bank statements, user portraits corresponding to the device information, social network relationships, and geographical locations; inputting the first multi-modal data into a risk scoring model to obtain a risk level corresponding to the first multi-modal data; and performing risk monitoring based on the risk level.

[0018] Among them, the risk scoring model includes: a feature extraction module, a relationship learning module, a reinforcement learning module, a generative adversarial network, and a risk scoring engine; inputting the multi-modal data into the risk scoring model to obtain a risk level corresponding to the multi-modal data includes: inputting the first multi-modal data into the feature extraction module to obtain a first feature; inputting the first feature into the relationship learning module to obtain a second feature; inputting the first feature into the generative adversarial network to obtain a third feature; inputting the first feature and the second feature into the reinforcement learning module to obtain a fourth feature; and inputting the first feature, the second feature, the third feature, and the fourth feature into the risk scoring engine to obtain a risk level corresponding to the first multi-modal data.

[0019] Among them, the method further includes: obtaining second multi-modal data; the second multi-modal data includes regulatory agency data, first bank data, second bank data, and financial institution data; inputting the second multi-modal data into an anti-fraud model to obtain target fraud information corresponding to the second multi-modal data; inputting the target fraud information into an intelligent contract risk control module to obtain a risk level; and inputting the target fraud information into a heterogeneous blockchain network for storage; obtaining a regulatory report based on the risk level.

[0020] Among them, the anti-fraud model includes a federated learning privacy protection layer, an anti-fraud layer, and an interpretable AI layer; inputting the second multi-modal data into the anti-fraud model to obtain fraud information corresponding to the second multi-modal data includes: inputting the second multi-modal data into the federated learning privacy protection layer to obtain privacy-protected target multi-modal data; inputting the target multi-modal data into the anti-fraud layer to obtain predicted fraud information; and inputting the predicted fraud information into the interpretable AI layer to obtain target fraud information.

[0021] Among them, the method further includes: obtaining third multi-modal data; wherein, the third multi-modal data includes basic user information, social media data, e-commerce behavior data, bank statements, and geographical locations; inputting the third multi-modal data into a recommendation model to obtain personalized recommendations corresponding to the third multi-modal data.

[0022] Among them, the recommendation model includes: a federated learning privacy protection layer, a user portrait construction layer, a deep learning layer, and a personalized recommendation layer; inputting the third multimodal data into the recommendation model to obtain the personalized recommendation corresponding to the third multimodal data, including: inputting the third multimodal data into the federated learning privacy protection layer to obtain the target multimodal data after privacy protection; inputting the target multimodal data into the user portrait construction layer to obtain user portrait features; inputting the user portrait features into the deep learning layer to obtain a recommendation strategy; inputting the recommendation strategy into the personalized recommendation layer to obtain the personalized recommendation corresponding to the third multimodal data.

[0023] Among them, the method further includes: obtaining fourth multimodal data; wherein, the fourth multimodal data includes online user behavior, offline transaction data, social media interaction data, customer service logs, and application program data; inputting the fourth multimodal data into the marketing decision-making model to obtain the marketing decision corresponding to the fourth multimodal data.

[0024] Among them, the method further includes: obtaining fifth multimodal data; wherein, the fifth multimodal data includes business department requirements, historical resource usage data, risk grading data, external market signals, and real-time performance indicators; inputting the fifth multimodal data into the intelligent resource scheduling model to obtain the resource scheduling decision corresponding to the fifth multimodal data.

[0025] Among them, the method further includes: obtaining sixth multimodal data; wherein, the sixth multimodal data includes bank statements, news text analysis, social media public opinion, macroeconomic indicators, industry reports, and regulatory policy documents; inputting the sixth multimodal data into the capital flow prediction model to obtain the resource scheduling decision corresponding to the sixth multimodal data.

[0026] In a second aspect, the present application provides a bank statement risk assessment application system, which includes a processor and a memory connected to the processor; the memory is used to store a computer program, and when the computer program is executed by the processor, it is used to implement the method provided in the first aspect.

[0027] The beneficial effects of the present application are as follows: Different from the prior art, the bank statement risk assessment method and application system provided by the present application obtain first multi-modal data; wherein, the first multi-modal data includes bank statements, device information corresponding to the bank statements, user portraits corresponding to the device information, social network relationships, and geographical locations; input the first multi-modal data into a risk scoring model to obtain the risk level corresponding to the first multi-modal data; and perform risk monitoring based on the risk level, overcoming the problems of data collection for various formats of domestic and foreign bank statements, and based on the characteristics of bank transaction data, combined with multi-modal data in other dimensions, performing data cross-validation, constructing a risk assessment model, identifying the integrity and authenticity of the statement records provided by enterprises, and being able to merge and summarize various transaction relationships, identifying false flows in transactions and their suspected related transactions, thereby improving the accuracy and efficiency of financial institutions in identifying the status of enterprise cash flows and potential risks. Description of the Drawings

[0028] To more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings. Among them:

[0029] Figure 1 is a flowchart of an embodiment of the bank statement risk assessment method provided by the present application;

[0030] Figure 2 is a flowchart of an embodiment of step 12 provided by the present application;

[0031] Figure 3 is a flowchart of another embodiment of the bank statement risk assessment method provided by the present application;

[0032] Figure 4 is a flowchart of an embodiment of step 32 provided by the present application;

[0033] Figure 5 is a flowchart of another embodiment of the bank statement risk assessment method provided by the present application;

[0034] Figure 6 is a flowchart of an embodiment of step 52 provided by the present application;

[0035] Figure 7 is a flowchart of another embodiment of the bank statement risk assessment method provided by the present application;

[0036] Figure 8 is a flowchart of another embodiment of the bank statement risk assessment method provided by the present application;

[0037] Figure 9 It is a schematic flowchart of another embodiment of the bank statement risk assessment method provided by this application;

[0038] Figure 10 It is a schematic flowchart of another embodiment of the bank statement risk assessment method provided by this application;

[0039] Figure 11 It is a schematic flowchart of another embodiment of the bank statement risk assessment method provided by this application;

[0040] Figure 12 It is a schematic structural diagram of an embodiment of the bank statement risk assessment application system provided by this application. Detailed implementation manners

[0041] Next, the technical solutions in the embodiments of this application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of this application. It can be understood that the specific embodiments described herein are only used to explain this application, rather than limiting this application. Additionally, it should be noted that for the sake of description, only parts related to this application rather than all structures are shown in the accompanying drawings. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of this application.

[0042] Referring to "embodiments" herein means that the specific features, structures, or characteristics described in conjunction with the embodiments can be included in at least one embodiment of this application. The phrase appears in various places in the specification and does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. Those skilled in the art will explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments.

[0043] Bank statement data is important information reflecting the economic activities of enterprises and individuals. For financial institutions including banks, investment banks, securities, and auditing, by analyzing bank statement data, comprehensively understanding the enterprise operation status, profitability, income and expenditure situation, and identifying possible improper behaviors of the target company can effectively identify potential risks.

[0044] Currently, in their due diligence reviews and risk assessments, they mainly rely on the following methods:

[0045] (1) Expert scoring method: relying on expert experience, there are problems such as strong subjectivity and poor consistency;

[0046] (2) Credit scoring model: building a model based on historical data, but it is difficult to cover all risk scenarios;

[0047] (3) Financial analysis method: Focus on the financial condition of the enterprise, but it is difficult to reflect real-time risks.

[0048] Characteristics of bank statement data: Bank statement data has the following characteristics:

[0049] (1) Real-time: Reflect the real-time economic activities of enterprises and individuals;

[0050] (2) Comprehensiveness: Cover multiple accounts and various transaction types;

[0051] (3) Relevance: There is a connection with other financial data of the enterprise, personal credit status, etc.

[0052] The existing technologies have the following limitations in bank statement risk assessment:

[0053] (1) Lack of an assessment method for the characteristics of bank statement data;

[0054] (2) Bank diversity, making it difficult to adapt to and collect all bank statement formats;

[0055] (3) Bank institutions cannot interconnect to query the statements of other banks, and all are provided by the enterprise itself, making it difficult to verify the integrity and authenticity of the data;

[0056] (4) Unable to effectively identify abnormal transactions and potential risks.

[0057] Based on this, the present application proposes to obtain first multi-modal data; wherein, the first multi-modal data includes bank statements, device information corresponding to the bank statements, user portraits corresponding to the device information, social network relationships, and geographical locations; input the first multi-modal data into a risk scoring model to obtain the risk level corresponding to the first multi-modal data; and perform risk monitoring based on the risk level, overcoming the data collection problems of various formats of bank statements at home and abroad, and based on the characteristics of bank transaction data, combined with multi-modal data in other dimensions, perform data cross-verification, construct a risk assessment model, identify the integrity and authenticity of the statement records provided by the enterprise, and be able to merge and summarize various transaction relationships, identify false flows of transactions with suspected associated transactions, thereby improving the accuracy and efficiency of financial institutions in identifying the status of enterprise cash flows and potential risks. Specifically, refer to any one of the following embodiments or a combination of any embodiments.

[0058] Refer to Figure 1 , Figure 1 is a schematic flowchart of an embodiment of the bank statement risk assessment method provided by the present application. The method includes:

[0059] Step 11: Obtain first multi-modal data; wherein, the first multi-modal data includes bank statements, device information corresponding to the bank statements, user portraits corresponding to the device information, social network relationships, and geographical locations.

[0060] In some embodiments, a multi-modal input layer can be utilized to fuse different data sources to obtain first multi-modal data.

[0061] Step 12: Input the first multi-modal data into a risk scoring model to obtain the risk level corresponding to the first multi-modal data.

[0062] In an application scenario, the risk scoring model includes: a feature extraction module, a relationship learning module, a reinforcement learning module, a generative adversarial network, and a risk scoring engine. Refer to Figure 2 , step 12 can specifically be the following process:

[0063] Step 121: Input the first multi-modal data into the feature extraction module to obtain a first feature.

[0064] In some embodiments, the feature extraction module includes a ResNet feature extractor and an attention mechanism layer. Input the first multi-modal data into the corresponding features of the ResNet feature extractor, and then input the corresponding features into the attention mechanism layer to obtain a first feature. The ResNet feature extractor can capture complex non-linear features.

[0065] Step 122: Input the first feature into the relationship learning module to obtain a second feature.

[0066] In some embodiments, the relationship learning module includes a graph neural network and a relationship embedding layer. Input the first feature into the graph neural network, use the graph neural network to construct the transaction relationships in the first multi-modal data into a graph, and input the graph into the relationship embedding layer to obtain a second feature. The graph neural network can learn the relationships between accounts.

[0067] Step 123: Input the first feature into the generative adversarial network to obtain a third feature.

[0068] In some embodiments, the generative adversarial network includes a generator and a discriminator. Input the first feature into the generator to obtain corresponding features, and then input these features into the discriminator. The generative adversarial network can enhance the generalization ability of the model.

[0069] Step 124: Input the first feature and the second feature into the reinforcement learning module to obtain a fourth feature.

[0070] In some embodiments, the reinforcement learning module includes a policy network, a value network, and a reward function. Input the first feature and the second feature into the policy network to obtain corresponding policies, input these policies into the value network to evaluate the value of each policy, and then input the policies and the values into the reward function for reward calculation, thereby obtaining a fourth feature. Among them, the reinforcement learning module is used to dynamically optimize the detection policy.

[0071] Step 125: Input the first feature, the second feature, the third feature, and the fourth feature into the risk scoring engine to obtain the risk level corresponding to the first multimodal data.

[0072] In some embodiments, the risk scoring engine includes a multi-layer perceptron, a confidence scoring module, and a risk classification module. Input the first feature, the second feature, the third feature, and the fourth feature into the multi-layer perceptron to obtain the perceptual features output by the multi-layer perceptron. Use the confidence scoring module to calculate the confidence of the perceptual features to obtain the corresponding confidence. And use the risk classification module to classify the perceptual features to obtain the corresponding risk type. Based on this risk type and the corresponding confidence, obtain the risk level corresponding to the first multimodal data. The multi-layer perceptron can comprehensively evaluate risks.

[0073] Step 13: Conduct risk monitoring according to the risk level.

[0074] In some embodiments, the risk level can be divided into low risk, medium risk, and high risk. In the case of low risk, release normally. In the case of medium risk, conduct manual review. In the case of high risk, freeze the account and / or alarm.

[0075] Furthermore, continuously learn according to the risk level and optimize the risk scoring model.

[0076] In this embodiment, by obtaining the first multimodal data; wherein, the first multimodal data includes bank statements, device information corresponding to the bank statements, user portraits corresponding to the device information, social network relationships, and geographical locations; input the first multimodal data into the risk scoring model to obtain the risk level corresponding to the first multimodal data; according to the risk level for risk monitoring methods, overcome the current problems of data collection for various formats of domestic and foreign bank statements, and based on the characteristics of bank transaction data, combined with other dimensions of multimodal data, conduct data cross-validation, construct a risk assessment model, identify the integrity and authenticity of the bank statement records provided by enterprises, and be able to merge and summarize various transaction relationships, identify false flows of transactions with their suspected related transactions, thereby improving the accuracy and efficiency of financial institutions in identifying the status of enterprise cash flows and potential risks. This embodiment can be applied to an abnormal behavior detection system based on deep learning.

[0077] Refer to Figure 3 , Figure 3 is a schematic flowchart of another embodiment of the bank statement risk assessment method provided by this application. The method includes:

[0078] Step 31: Obtain the second multimodal data; the second multimodal data includes regulatory agency data, first bank data, second bank data, and financial institution data.

[0079] Step 32: Input the second multimodal data into the anti-fraud model to obtain the target fraud information corresponding to the second multimodal data.

[0080] In an application scenario, the anti-fraud model includes a federated learning privacy protection layer, an anti-fraud layer, and an explainable AI layer. Refer to Figure 4 , Step 32 can be the following process:

[0081] Step 321: Input the second multimodal data into the federated learning privacy protection layer to obtain the target multimodal data after privacy protection.

[0082] In some embodiments, the federated learning privacy protection layer includes a federated learning aggregator, a homomorphic encryption module, and a differential privacy mechanism. Input the second multimodal data into the federated learning aggregator to obtain federated learning features, and input the federated learning features into the homomorphic encryption module for encryption to obtain encrypted features. Then use the differential privacy mechanism to process the encrypted features to obtain the target multimodal data after privacy protection.

[0083] Step 322: Input the target multimodal data into the anti-fraud layer to obtain predicted fraud information.

[0084] In some embodiments, the anti-fraud layer includes: a feature extraction module, a graph neural network, a generative adversarial network, and a reinforcement learning policy network. Input the target multimodal data into the feature extraction module to obtain multimodal features, and input the multimodal features into the graph neural network to obtain relational graph features, and input the relational graph features into the generative adversarial network for enhancement, and input the enhanced relational graph features into the reinforcement learning policy network for policy optimization to obtain predicted fraud information. Among them, the feature extraction module can be composed of a ResNet feature extractor. That is, operations such as ResNet feature extraction, graph neural network relational learning, generative adversarial network enhancement, and reinforcement learning dynamic optimization can be performed in the anti-fraud layer.

[0085] Step 323: Input the predicted fraud information into the explainable AI layer to obtain the target fraud information.

[0086] In some embodiments, the explainable AI layer includes a decision tree visualization module, a feature contribution decomposer, and a local interpretability builder. Input the predicted fraud information into the decision tree visualization module for decision process visualization. And input the predicted fraud information into the feature contribution decomposer to obtain the first target fraud information. Input the predicted fraud information into the local interpretability builder to obtain the second target fraud information.

[0087] Step 33: Input the target fraud information into the intelligent contract risk control module to obtain a risk level; and input the target fraud information into the heterogeneous blockchain network for storage.

[0088] In some embodiments, the intelligent contract risk control module includes a risk assessment contract module, an automatic early warning mechanism, and a transaction freezing module. Input the target fraud information into the risk assessment contract module to obtain risk information, and use the automatic early warning mechanism to automatically warn of the risk information. Then, perform a transaction freezing operation in the transaction freezing module according to the warning result, and perform a final risk scoring to obtain a risk level. And perform abnormal transaction marking.

[0089] The heterogeneous blockchain network includes modules such as consortium chains, public chain interfaces, private chain interactions, and cross-chain protocols, and can interconnect different blockchain networks to build a more secure distributed ledger system.

[0090] Step 34: Obtain a supervision report according to the risk level.

[0091] In this embodiment, the federated learning technology is introduced to achieve multi-party collaborative modeling and improve the generalization ability of the model on the premise of protecting data privacy. And combined with interpretable AI technology to explain the prediction results of the model and improve the transparency and credibility of the model. And design intelligent contracts to achieve automated risk assessment and response, reduce manual intervention, and overall achieve a comprehensive integration of data security, model intelligence, and accurate risk identification, providing a new technological paradigm for financial anti-fraud. This embodiment can be applied to an anti-fraud system that combines blockchain and AI.

[0092] Refer to Figure 5 , Figure 5 is a schematic flowchart of another embodiment of the bank statement risk assessment method provided by this application. The method includes:

[0093] Step 51: Obtain third multi-modal data; where the third multi-modal data includes basic user information, social media data, e-commerce behavior data, bank statements, and geographical locations.

[0094] Step 52: Input the third multi-modal data into the recommendation model to obtain personalized recommendations corresponding to the third multi-modal data.

[0095] In an application scenario, the recommendation model includes: a federated learning privacy protection layer, a user portrait construction layer, a deep learning layer, and a personalized recommendation layer. Refer to Figure 6 , Step 52 can be the following process:

[0096] Step 521: Input the third multi-modal data into the federated learning privacy protection layer to obtain the target multi-modal data after privacy protection.

[0097] In some embodiments, the federated learning privacy protection layer includes a federated learning aggregator, a homomorphic encryption module, and a differential privacy mechanism. The third multi-modal data is input into the federated learning aggregator to obtain federated learning features, and the federated learning features are input into the homomorphic encryption module for encryption to obtain encrypted features. Then, the differential privacy mechanism is used to process the encrypted features to obtain the target multi-modal data after privacy protection.

[0098] Step 522: Input the target multi-modal data into the user profile construction layer to obtain user profile features.

[0099] In some embodiments, the user profile construction layer includes a demographic feature module, an interest tag extraction module, a behavior preference analysis module, and a user profile embedding module. The target multi-modal data is input into the demographic feature module to obtain demographic features. The demographic features are input into the interest tag extraction module to obtain interest tags. The interest tags and demographic features are input into the behavior preference analysis module to obtain behavior preference features. The behavior preference features are input into the user profile embedding module to obtain user profile features.

[0100] Step 523: Input the user profile features into the deep learning layer to obtain a recommendation strategy.

[0101] In some embodiments, the deep learning layer includes: a feature extraction module, a graph neural network, a generative adversarial network, and a reinforcement learning policy network. The user profile features are input into the feature extraction module to obtain multi-modal features, and the multi-modal features are input into the graph neural network to obtain relational graph features, and the relational graph features are input into the generative adversarial network for enhancement, and the enhanced relational graph features are input into the reinforcement learning policy network for policy optimization to obtain a recommendation strategy. Among them, the feature extraction module can be composed of a ResNet feature extractor. That is, operations such as ResNet feature extraction, graph neural network relational learning, generative adversarial network enhancement, and reinforcement learning dynamic optimization can be performed in the deep learning layer.

[0102] Step 524: Input the recommendation strategy into the personalized recommendation layer to obtain the personalized recommendation corresponding to the third multi-modal data.

[0103] In some embodiments, the personalized recommendation layer includes a collaborative filtering module, a context-related recommendation module, and a real-time recommendation adjustment module. The recommendation strategy is input into the collaborative filtering module to obtain a filtered recommendation strategy. The filtered recommendation strategy is input into the context-related recommendation module and the real-time recommendation adjustment module to optimize and adjust the filtered recommendation strategy to obtain the personalized recommendation corresponding to the third multi-modal data.

[0104] Furthermore, it is possible to perform operations such as predicting the click-through rate of personalized recommendations output by the recommendation model and rating the satisfaction of recommendations, and then continuously optimize and learn the recommendation model based on the recommendation satisfaction rating. Also, after outputting personalized recommendations, it is possible to obtain user interaction feedback and continuously optimize and learn the recommendation model.

[0105] In this embodiment, a graph neural network is used in the intelligent recommendation engine to model the complex relationships between customers, products, and services, providing more accurate recommendations. And reinforcement learning is used to continuously optimize the recommendation strategy through interaction with users, achieving long-term optimization of personalized recommendations.

[0106] Also, by combining customer portrait data such as age, gender, and occupation, the accuracy of recommendations is improved. External data such as social media data and e-commerce data is introduced to enrich the user portrait.

[0107] Federated learning is used to achieve collaborative modeling of multi-party data while protecting data privacy. And differential privacy is used to add noise to protect user privacy.

[0108] Refer to Figure 7 , Figure 7 is a schematic flowchart of another embodiment of the bank statement risk assessment method provided by this application. The method includes:

[0109] Step 71: Obtain fourth multi-modal data; where the fourth multi-modal data includes online user behavior, offline transaction data, social media interaction data, customer service logs, and application data.

[0110] After obtaining the fourth multi-modal data, the fourth multi-modal data can be processed using a streaming computing and event processing module. For example, the fourth multi-modal data is processed using a scalable distributed stream platform, a distributed stream processing system, a real-time event analysis engine, and an event trigger rule engine.

[0111] Step 72: Input the fourth multi-modal data into the marketing decision-making model to obtain a marketing decision corresponding to the fourth multi-modal data.

[0112] In some embodiments, the marketing decision-making model includes a user segmentation model, a personalized recommendation module, a marketing strategy generator, and a natural language generation module. The fourth multi-modal data is input into the user segmentation model to obtain a corresponding user segmentation result, and the user segmentation result is input into the personalized recommendation module and the marketing strategy generator to obtain corresponding strategy features, and the strategy features are input into the natural language generation module to obtain a marketing decision corresponding to the fourth multi-modal data.

[0113] Furthermore, multi-channel marketing execution is performed on the marketing decision. Such as online advertising placement, SMS / email marketing, personalized pages / content, and intelligent customer service recommendations.

[0114] Furthermore, evaluate the marketing effect. For example, use technologies such as causal inference models, data-driven decision-making frameworks, A / B testing, marketing ROI analysis, and deep conversion rate analysis to evaluate the marketing effect.

[0115] Furthermore, use a continuous optimization mechanism to optimize the whole, such as using reinforcement learning optimizers, model auto-tuning, marketing strategy evolution and other technologies. Then, conduct marketing strategy feedback.

[0116] In this embodiment, by using stream computing, it is possible to process massive amounts of data in real time and respond to customer behavior in a timely manner. And by using an event-driven architecture, marketing activities can be triggered based on events, improving the response speed. And multi-channel marketing can split bank statements into online and offline statements, integrate online and offline channels, achieve omni-channel marketing, and improve the marketing effect. And by using personalized content generation, customized content can be generated for different channels based on natural language generation technology. And for marketing effect evaluation, causal inference is used to evaluate the real impact of marketing activities on user behavior. And A / B testing is used to compare the effects of different marketing strategies and optimize marketing strategies.

[0117] That is, by breaking through data barriers and introducing intelligent technologies, the system has realized a new data-driven marketing paradigm for the future, and can provide more accurate, personalized, and efficient marketing solutions. This embodiment can be applied to a data-driven marketing automation system.

[0118] Refer to Figure 8 , Figure 8 is a schematic flowchart of another embodiment of the bank statement risk assessment method provided by this application. The method includes:

[0119] Step 81: Obtain fifth multi-modal data; where the fifth multi-modal data includes business department requirements, historical resource usage data, risk grading data, external market signals, and real-time performance indicators.

[0120] Step 82: Input the fifth multi-modal data into the intelligent resource scheduling model to obtain a resource scheduling decision corresponding to the fifth multi-modal data.

[0121] In some embodiments, the intelligent resource scheduling model includes a deep learning intelligent analysis layer, a multi-objective optimization engine, a risk control module, and an intelligent scheduling decision layer.

[0122] Input the fifth multimodal data into the deep learning intelligent analysis layer for the preliminary scheduling strategy. Input the preliminary scheduling strategy into the multi-objective optimization engine to obtain the corresponding resource allocation weights. Input the preliminary scheduling strategy and the resource allocation weights into the risk control module to obtain the corresponding intermediate scheduling strategy. Input the resource allocation weights and the intermediate scheduling strategy into the intelligent scheduling decision layer to obtain the resource scheduling decision corresponding to the fifth multimodal data.

[0123] Among them, the deep learning intelligent analysis layer includes a sequence prediction module, a feature extraction module, a graph neural network, a reinforcement learning policy network, and a generative adversarial network. Input the fifth multimodal data into the sequence prediction module to obtain the target sequence, input the target sequence into the feature extraction module to obtain multimodal features, and input the multimodal features into the graph neural network to obtain relational graph features, and input the relational graph features into the reinforcement learning policy network for policy optimization. Input the optimized policy into the generative adversarial network for enhancement to obtain the preliminary scheduling strategy. Among them, the feature extraction module can be composed of a ResNet feature extractor.

[0124] Among them, the multi-objective optimization engine includes a Pareto optimal solution module, a multi-objective genetic algorithm module, a constraint condition evaluation module, and a resource allocation weight calculation module. That is, use the Pareto optimal solution module, the multi-objective genetic algorithm module, the constraint condition evaluation module, and the resource allocation weight calculation module to process the preliminary scheduling strategy to obtain the corresponding resource allocation weights.

[0125] Among them, the risk control module includes an anomaly detection algorithm unit, a risk assessment model, a risk warning mechanism, and an emergency resource allocation unit. That is, use the anomaly detection algorithm unit, the risk assessment model, the risk warning mechanism, and the emergency resource allocation unit, combined with the preliminary scheduling strategy and the resource allocation weights, to obtain the corresponding intermediate scheduling strategy.

[0126] Among them, the intelligent scheduling decision layer includes a resource allocation strategy module, a real-time scheduling execution module, a dynamic load balancing module, and a resource utilization optimization module. That is, use the resource allocation strategy module, the real-time scheduling execution module, the dynamic load balancing module, and the resource utilization optimization module to process the input resource allocation weights and the intermediate scheduling strategy to obtain the resource scheduling decision corresponding to the fifth multimodal data.

[0127] Furthermore, the performance feedback loop mechanism, the model automatic tuning mechanism, and the decision-making knowledge precipitation mechanism can be used to continuously learn and optimize the model, and then obtain a resource scheduling optimization report for the decision-making layer to refer to for decision-making.

[0128] In this embodiment, the above method can be applied to an intelligent resource scheduling system. Among them, the sequence prediction module can be composed of a sequence model (RNN / LSTM), which can consider time series data, predict future resource requirements, and achieve more accurate scheduling. Reinforcement learning is used to transform the resource scheduling problem into a Markov decision process, and through trial-and-error learning, an optimal scheduling strategy is found. And a graph neural network is used to model the complex relationships between various business departments within the bank to achieve more fine-grained resource allocation. And Pareto optimization can consider multiple optimization goals simultaneously, such as maximizing revenue, minimizing risk, balancing load, etc. And a multi-objective genetic algorithm can find a set of non-dominated solutions to provide more choices for decision-makers. And an anomaly detection algorithm can identify abnormal resource usage situations and take timely measures. And a risk assessment model is used to evaluate the risks under different scheduling strategies to ensure the stable operation of the system. That is, overall, the intelligent resource scheduling system can optimize the bank's resource allocation, improve the efficiency of capital use, and reduce operating costs.

[0129] Refer to Figure 9 , Figure 9 is a schematic flowchart of another embodiment of the bank statement risk assessment method provided by this application. The method includes:

[0130] Step 91: Obtain the sixth multimodal data; among them, the sixth multimodal data includes bank statements, news text analysis, social media sentiment, macroeconomic indicators, industry reports, and regulatory policy documents.

[0131] In some embodiments, the sixth multimodal data can be subjected to multimodal data cleaning, the cleaned multimodal data can be characterized using feature engineering, and data standardization can be performed, and then data embedding can be performed to obtain the final features. And the final features are input into the capital flow prediction model.

[0132] Step 92: Input the sixth multimodal data into the capital flow prediction model to obtain the resource scheduling decision corresponding to the sixth multimodal data.

[0133] Among them, the capital flow prediction model includes a deep learning prediction module, a causal inference and analysis module, a dynamic model update module, a prediction and risk assessment module, and a visualization and reporting module.

[0134] In some embodiments, the deep learning prediction module includes: a feature extraction module, a graph neural network, a generative adversarial network, a reinforcement learning prediction policy network, and a long short-term memory network. The sixth multimodal data is input into the feature extraction module to obtain multimodal features, and the multimodal features are input into the graph neural network to obtain relational graph features. The relational graph features are input into the generative adversarial network for enhancement, and the enhanced relational graph features are input into the reinforcement learning prediction policy network for policy optimization to obtain a prediction policy. The prediction policy is input into the long short-term memory network for temporal modeling to obtain the modeled prediction policy. Among them, the feature extraction module can be composed of a ResNet feature extractor. That is, operations such as ResNet feature extraction, graph neural network relational learning, generative adversarial network enhancement, and reinforcement learning dynamic optimization can be performed in the deep learning prediction module.

[0135] Among them, the causal inference and analysis module includes causal graph construction, counterfactual reasoning module, key influencing factor identification, and scenario simulation engine. The modeled prediction policy is input into the causal inference and analysis module, and the modeled prediction policy is processed by using the causal graph construction, counterfactual reasoning module, key influencing factor identification, and scenario simulation engine to obtain an analysis result.

[0136] Among them, the dynamic model update module includes an online learning mechanism, a transfer learning mechanism, a model performance evaluation mechanism, and an adaptive parameter adjustment mechanism. The modeled prediction policy and the analysis result are input into the dynamic model update module, and the modeled prediction policy and the analysis result are processed by using the online learning mechanism, the transfer learning mechanism, the model performance evaluation mechanism, and the adaptive parameter adjustment mechanism, thereby updating the model and obtaining the corresponding target result.

[0137] Among them, the prediction and risk assessment module includes functions such as fund flow prediction, risk scoring, anomaly detection, and decision support. The target result is input into the prediction and risk assessment module to obtain a resource scheduling decision corresponding to the sixth multimodal data.

[0138] Furthermore, after obtaining the resource scheduling decision corresponding to the sixth multimodal data, the visualization and reporting module is used to interpret the prediction result, give a risk warning, and provide decision suggestions and visualize them through an interactive dashboard.

[0139] In this embodiment, a fund flow prediction system based on big data can be applied. Among them, text data can be used to analyze text information such as news and reports to capture market trends. Social media data can be used to analyze public opinion on social media to predict market fluctuations. External economic data can be used to combine macroeconomic data to improve prediction accuracy.

[0140] And establish a causal relationship diagram between capital flows and influencing factors using a causal diagram model to deeply explore the causal mechanism. And simulate capital flow situations under different scenarios using counterfactual reasoning to evaluate the potential impact of decisions. And through online learning that dynamically updates the model, as new data continuously pours in, the model can be updated in a timely manner to ensure the timeliness of predictions. And through transfer learning, knowledge from other fields or banks can be transferred to the current model to accelerate model training.

[0141] See Figure 10 , Figure 10 is a schematic flowchart of another embodiment of the bank statement risk assessment method provided by this application. The method includes:

[0142] Step 101: Obtain seventh multi-modal data; where the seventh multi-modal data includes external credit data, bank statements, social media public opinion, news, market data, and regulatory reports.

[0143] In some embodiments, the seventh multi-modal data can be subjected to multi-modal data cleaning, feature engineering on the cleaned multi-modal data for feature extraction, and data standardization, and then data embedding is performed to obtain the final features. And the final features are input into the risk analysis model.

[0144] Step 102: Input the seventh multi-modal data into the risk analysis model to obtain the risk analysis result corresponding to the seventh multi-modal data.

[0145] In some embodiments, the risk analysis model includes a deep learning prediction module, a risk detection and assessment module, a real-time warning mechanism module, a visualization interaction module, and a continuous learning and optimization module.

[0146] In some embodiments, the deep learning prediction module includes: a feature extraction module, a graph neural network, a generative adversarial network, a reinforcement learning risk strategy network, and a multi-dimensional risk assessment model. Input the seventh multi-modal data into the feature extraction module to obtain multi-modal features, and input the multi-modal features into the graph neural network to obtain relationship graph features, and input the relationship graph features into the generative adversarial network for enhancement, input the enhanced relationship graph features into the reinforcement learning risk strategy network for strategy optimization to obtain a prediction strategy. And input the prediction strategy into the multi-dimensional risk assessment model to obtain a preliminary risk result. Among them, the feature extraction module can be composed of a ResNet feature extractor. That is, ResNet feature extraction, graph neural network relationship learning, generative adversarial network enhancement, and reinforcement learning dynamic optimization and other operations can be performed in the deep learning prediction module.

[0147] In some embodiments, the risk detection and assessment module includes a credit risk model, a market risk model, an operational risk model, a reputation risk model, and an anomaly detection model. The preliminary risk results are input into the risk detection and assessment module, and the credit risk model, market risk model, operational risk model, reputation risk model, and anomaly detection model are used to analyze the preliminary risk results to obtain the assessment results for the corresponding risk types.

[0148] Among them, the real-time warning mechanism module includes a risk scoring engine, an alarm trigger rule module, a priority classification module, and an automatic response process module. The assessment results of the risk types are input into the real-time warning mechanism module. The risk scoring engine, alarm trigger rule module, priority classification module, and automatic response process module are used to process the assessment results of the risk types to obtain the risk analysis results corresponding to the seventh multi-modal data.

[0149] Among them, the visualization interaction module includes functions such as an interactive dashboard, geographic information visualization, a risk heat map, multi-dimensional charts, and risk trend analysis. After obtaining the risk analysis results corresponding to the seventh multi-modal data, the functions such as the interactive dashboard, geographic information visualization, risk heat map, multi-dimensional charts, and risk trend analysis are used to perform corresponding visual displays on the risk analysis results.

[0150] Among them, the continuous learning and optimization module has functions such as model performance evaluation, automatic parameter tuning, knowledge graph update, and decision feedback loop for continuous learning and optimization.

[0151] In this embodiment, it can be applied to a real-time data monitoring platform. In the real-time data monitoring platform, the data source is extended. In addition to traditional bank transaction data, external data sources such as social media, news reports, and market conditions can also be introduced to build a more comprehensive risk monitoring system.

[0152] And for optimizing the risk model, a multi-dimensional risk model is constructed using a multi-dimensional risk model covering credit risk, market risk, operational risk, reputation risk, etc.

[0153] And by using the method of dynamic risk assessment, based on real-time data, the risk model parameters are dynamically adjusted to improve the adaptability of the model. And an anomaly detection algorithm is adopted to timely detect abnormal fluctuations in the data, such as large transactions, frequent operations, etc.

[0154] And for visual enhancement, a user-friendly interactive dashboard is provided using the interactive dashboard, supporting custom views and alarm settings. And geographic information visualization, visualizing risk events on the map to intuitively display the risk distribution.

[0155] Refer to Figure 11 , Figure 11It is a schematic flowchart of another embodiment of the bank statement risk assessment method provided by this application. The method includes:

[0156] Step 111: Obtain the eighth multimodal data; wherein, the eighth multimodal data includes social network data, user pictures, text and document data, and bank statements.

[0157] Step 112: Input the social network data into the deep learning model to obtain the first feature.

[0158] In some embodiments, the GNN module in the deep learning model can be used to perform relationship learning on the social network data to obtain the first feature. Mine the association relationships from the social network data to understand complex interest relationships or behavior patterns.

[0159] Step 113: Input the user pictures into the deep learning model to obtain the second feature.

[0160] In some embodiments, the ResNet feature extraction module in the deep learning model can be used to extract features from the user pictures to obtain the second feature. For example, extract the features in the user pictures to provide visual support for compliance analysis (such as identity verification or fraud detection).

[0161] Step 114: Input the text and document data, bank statements, the first feature, and the second feature into the compliance analysis module to obtain a compliance scenario.

[0162] In some embodiments, the text and document data can be input into the compliance analysis module after natural language processing.

[0163] In some embodiments, input the text and document data, bank statements, the first feature, and the second feature into the compliance analysis module to construct a compliance knowledge graph, and then perform natural language processing enhancement through multilingual support, causal relationship analysis, and sentiment analysis, and then perform text and knowledge compliance analysis on the enhanced data and enrich the compliance background, and then perform scenario simulation to obtain a compliance scenario. Among them, use reinforcement learning to optimize the compliance strategy of the business process. And use GAN to generate high-quality virtual data and scenarios to supplement the shortage of real data.

[0164] Step 115: Input the compliance scenario into the risk scoring engine to obtain a compliance risk assessment.

[0165] In some embodiments, the compliance scenario can be input into the reinforcement learning module in the deep learning model for reinforcement learning, and the data after reinforcement learning is input into the risk scoring engine for real-time compliance risk assessment to obtain a compliance risk assessment. The reinforcement learning module optimizes the decision-making of the simulated scenario and predicts potential risks. Combining with the virtual data generated by GAN improves the accuracy of model evaluation.

[0166] In this embodiment, the above method can be applied to an intelligent compliance management system. Among them, by using the compliance knowledge graph construction, a comprehensive compliance knowledge graph covering laws and regulations, regulatory policies, industry standards, etc. can be established to improve the accuracy of compliance analysis. And the multi-language support function can support compliance document analysis in multiple languages. And use sentiment analysis to analyze the sentiment tendency in the text and identify potential compliance risks. And use causal relationship analysis to mine the causal relationship in the text and deeply understand the causes of compliance events. And use compliance scenario simulation to build a virtual compliance scenario, simulate and test different business processes, and discover potential compliance issues in advance.

[0167] Furthermore, in order to improve the speed and accuracy of banking business data processing and realize intelligent analysis and automatic processing of transaction data, the following methods can be used:

[0168] First, in the multimodal data part, in addition to traditional bank transaction data, consider integrating external data sources (such as social media, third-party payment, etc.) to enrich data dimensions and enhance analysis depth.

[0169] Secondly, a distributed data processing system based on cloud computing can be formed.

[0170] Introducing real-time stream processing technology into distributed data processing systems to quickly process massive amounts of real-time data and achieve real-time monitoring and early warning. And using edge computing, delegating some computing tasks to edge nodes, reducing network latency and improving system response speed.

[0171] At the same time, pay attention to data security and privacy protection: adopt more advanced encryption algorithms and access control mechanisms to ensure data security and comply with relevant regulatory requirements.

[0172] Specific technical points may include:

[0173] Distributed storage: HDFS, S3, Ceph, etc. Use HDFS, S3, Ceph and other distributed storage systems to provide efficient and reliable data storage capabilities.

[0174] Distributed computing: Spark, Flink, Hadoop MapReduce, etc. Batch processing tasks use Spark, Flink or Hadoop MapReduce to calculate large-scale offline data.

[0175] Stream processing: Kafka, Storm, etc. Real-time data stream processing uses Kafka or Storm to achieve fast response and real-time monitoring.

[0176] Data Security: Encryption algorithms (such as AES, RSA), access control (RBAC), data masking, etc. Advanced encryption algorithms such as AES and RSA can be used for data encryption to ensure the security of data storage and transmission. Role-based access control (RBAC) can be adopted for access control to restrict access rights to sensitive data. Data masking can shield sensitive information during data processing to ensure privacy compliance.

[0177] Finally, the analysis results are stored in the database to support business applications and generate compliance reports. The real-time warning module provides instant risk alerts.

[0178] Specifically, it can be as follows:

[0179] Obtain heterogeneous data sources, preprocess the heterogeneous data sources and input them into the edge computing layer. Use the edge nodes in the edge computing layer to distribute computing tasks for distributed computing and real-time data processing. Use a stream processing engine for access control, and combine it with a real-time processing system for real-time monitoring and analysis, and conduct event detection for real-time warning. Output the warning information to the business application system. And output a compliance report to the business application system through the data security layer. For distributed computing, store its analysis results in the database and feedback them to the business application system.

[0180] During this process, third-party payment data, social media data, and traditional bank data in the heterogeneous data sources are distributedly stored, and these data are distributedly calculated and input into the stream processing engine. Encrypted storage is used during storage.

[0181] In some embodiments, the distributed computing architecture can adopt the form of a distributed data center. Among them, the asynchronous federated learning training workflow in the data center is as follows:

[0182] 1. Model Distribution: The container manager selects an idle container for model training. The selection is based on the version of the target model (i.e., the number of training times) and the historical training logs of the container.

[0183] 2. Resource Allocation: The resource manager allocates the most cost-effective hardware resources to the container according to the model version and the historical running time of the container.

[0184] 3. Model Training: The container uses its data to train the received planetary model.

[0185] 4. Model Collection: The container sends the trained planetary model to the model manager, and the container manager updates the container information.

[0186] 5. Model Update: The model manager aggregates the trained planetary model and the stellar model using specific weights and updates the corresponding planetary model in the model list.

[0187] This asynchronous training process will continue until the training time reaches the threshold.

[0188] Among them, the model rotation process between data centers is as follows:

[0189] 1. Model distribution: When the training time within a data center reaches the threshold, the model manager in each data center will aggregate all planetary models to generate a master model, and send this model together with the average number of training times of all planetary models to the target data center. Each data center will take turns to select the target center in different rotation rounds.

[0190] 2. Model routing: The output port transmits the model to the target data center according to the current routing table. The routing table selects the shortest path for model transmission.

[0191] 3. Model update: After the input port receives the master model, the model manager uses this model to replace the old stellar model.

[0192] Before the model distribution step, the model manager will reset the timer and wait for the next round of model rotation.

[0193] Key step: Direct communication is carried out between the stellar model and the master model through the model communicator.

[0194] The resource manager is responsible for allocating resources to the training containers.

[0195] The container selector selects a model among the planetary models and uses the allocated resources to train in the container.

[0196] This structure ensures the asynchrony and efficiency of model training, while reducing the communication overhead and the complexity of resource scheduling.

[0197] For example, the distributed computing framework includes Data Center 1, Data Center 2, Data Center 3, and Data Center 4. Each data center is involved in the update and distribution of the stellar model. And when the model rotates, through model distribution, model routing, and model update methods, the old stellar model is replaced with a new stellar model.

[0198] In some embodiments, multimodal data can also be obtained. The multimodal data includes business process data, user pictures, document / text data, and social network data.

[0199] Input the business process data into the unsupervised learning module to output the first feature, input the user picture into the ResNet feature extraction module to output the second feature. Input the document / text data into the Transformer model to output the third feature. Input the social network data into the GNN relationship learning module to output the fourth feature.

[0200] Input the first feature, second feature, third feature, and fourth feature into the feature fusion module to obtain a fused feature. Perform data pattern analysis on the fused feature to discover data patterns and anomalies, and conduct anomaly detection on the fused feature to identify potential data problems. Input the data pattern analysis results and anomaly detection into the reinforcement learning module to optimize the analysis strategy and generate virtual data.

[0201] Furthermore, perform risk prediction and decision-making on the data output by the reinforcement learning module, and then use the interpretable AI module to generate an analysis report and provide a basis for decision-making.

[0202] Furthermore, use the automated data analysis system to display the analysis report, monitor risks in real time, and issue warnings.

[0203] Among them, the above can be applied to an automated data analysis system. The deep learning model introduces Transformer to enhance the model's expressive ability and generalization ability. Use unsupervised learning to explore unsupervised learning algorithms and discover potential patterns and anomalies in the data. Use interpretable AI to enhance the interpretability of the model and improve users' trust in the model results.

[0204] Among them, ResNet feature extraction extracts high-level visual features from images for analyzing user behavior or risks.

[0205] The Transformer model processes complex texts, enhances language understanding ability, and supports multilingual and context modeling.

[0206] GNN relationship learning analyzes social networks and graph-structured data to discover node relationships and graph characteristics.

[0207] The unsupervised learning module discovers potential patterns and anomalies in the data through clustering and dimensionality reduction.

[0208] The feature fusion module integrates features from different data sources to enhance the model's expressive ability and generalization ability.

[0209] The GAN generation module generates virtual data for simulation testing and data augmentation.

[0210] The reinforcement learning module optimizes the strategy and provides dynamic risk prediction and decision-making capabilities.

[0211] The interpretable AI module enhances the interpretability of the model and improves users' trust by generating highly readable analysis reports.

[0212] The automated data analysis system provides real-time monitoring, data visualization, and alert functions to support the automated risk management of the business.

[0213] Refer to Figure 12 , Figure 12It is a schematic structural diagram of an embodiment of the bank statement risk assessment application system provided by this application. The bank statement risk assessment application system 120 includes a processor 121 and a memory 122 connected to the processor 121; the memory 122 is used to store a computer program, and when the computer program is executed by the processor 121, it is used to implement the method of any of the above embodiments.

[0214] In several implementation manners provided by this application, it should be understood that the disclosed method and device can be implemented in other ways. For example, the device implementation manner described above is only illustrative. For example, the division of the modules or units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0215] If the integrated unit in the above other embodiments is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of this 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 enable a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the methods described in various embodiments of this application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0216] The above description is only the implementation manner of this application, and does not limit the patent scope of this application. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of this application, or directly or indirectly applied in other related technical fields, is equally included in the patent protection scope of this application.

Claims

1. A bank flow risk assessment method, characterized in that: The method comprises: Acquire first multimodal data; wherein the first multimodal data includes bank statements, device information corresponding to the bank statements, user portraits corresponding to the device information, social network relationships, and geographic locations; Inputting the first multimodal data into a risk scoring model to obtain a risk level corresponding to the first multimodal data; Risk monitoring is performed according to the risk levels.

2. The method according to claim 1, characterized in that The risk scoring model includes: a feature extraction module, a relationship learning module, a reinforcement learning module, a generative adversarial network and a risk scoring engine; the multimodal data is input into the risk scoring model to obtain the risk level corresponding to the multimodal data, including: Inputting the first multimodal data into the feature extraction module to obtain a first feature; Inputting the first feature into the relationship learning module to obtain a second feature; Inputting the first feature into the generative adversarial network to obtain a third feature; Inputting the first feature and the second feature into the reinforcement learning module to obtain a fourth feature; The first feature, the second feature, the third feature, and the fourth feature are input into the risk scoring engine to obtain a risk level corresponding to the first multimodal data.

3. The method according to claim 1, characterized in that The method further comprises: Acquire second multimodal data; the second multimodal data includes regulatory agency data, first bank data, second bank data, and financial institution data; Inputting the second multimodal data into an anti-fraud model to obtain target fraud information corresponding to the second multimodal data; Input the target fraud information into the smart contract risk control module to obtain the risk level; and inputting the target fraud information into a heterogeneous blockchain network for storage; Obtain regulatory reporting based on the risk level stated.

4. The method according to claim 3, characterized in that The anti-fraud model includes a federated learning privacy protection layer, an anti-fraud layer, and an explainable AI layer; the step of inputting the second multimodal data into the anti-fraud model to obtain fraud information corresponding to the second multimodal data includes: Inputting the second multimodal data into the federated learning privacy protection layer to obtain target multimodal data after privacy protection; Inputting the target multimodal data into the anti-fraud layer to obtain predicted fraud information; The predicted fraud information is input into the explainable AI layer to obtain the target fraud information.

5. The method according to claim 1, characterized in that: The method further comprises: Acquire third multimodal data; wherein the third multimodal data includes basic user information, social media data, e-commerce behavior data, bank statements, and geographic location; The third multimodal data is input into a recommendation model to obtain personalized recommendations corresponding to the third multimodal data.

6. The method according to claim 5, characterized in that The recommendation model includes: a federated learning privacy protection layer, a user portrait construction layer, a deep learning layer, and a personalized recommendation layer; the inputting the third multimodal data into the recommendation model to obtain personalized recommendations corresponding to the third multimodal data includes: Inputting the third multimodal data into the federated learning privacy protection layer to obtain target multimodal data after privacy protection; Inputting the target multimodal data into the user portrait construction layer to obtain user portrait features; Inputting the user portrait features into the deep learning layer to obtain a recommendation strategy; The recommendation strategy is input into the personalized recommendation layer to obtain personalized recommendations corresponding to the third multimodal data.

7. The method according to claim 1, characterized in that The method further comprises: Acquire fourth multimodal data; wherein the fourth multimodal data includes online user behavior, offline transaction data, social media interaction data, customer service logs, and application data; The fourth multimodal data is input into a marketing decision model to obtain a marketing decision corresponding to the fourth multimodal data.

8. The method according to claim 1, characterized in that The method further comprises: Acquiring fifth multimodal data; wherein the fifth multimodal data includes business department needs, historical resource usage data, risk classification data, external market signals, and real-time performance indicators; The fifth multimodal data is input into an intelligent resource scheduling model to obtain a resource scheduling decision corresponding to the fifth multimodal data.

9. The method according to claim 1, characterized in that: The method further comprises: Acquiring sixth multimodal data; wherein the sixth multimodal data includes bank statements, news text analysis, social media public opinion, macroeconomic indicators, industry reports, and regulatory policy documents; The sixth multimodal data is input into a capital flow prediction model to obtain a resource scheduling decision corresponding to the sixth multimodal data.

10. A bank flow risk assessment application system, characterized in that: The bank flow risk assessment application system includes a processor and a memory connected to the processor; the memory is used to store a computer program, and when the computer program is executed by the processor, it is used to implement the method described in any one of claims 1-9.

Images