Account credit assessment method, credit assessment device, storage medium and processor

Abstract

The invention provides an account credit evaluation method, a credit evaluation device, a storage medium and a processor. The method comprises the steps of obtaining multi-modal transaction data of a user, the multi-modal transaction data comprising transaction flow data and a plurality of related data, the transaction flow data being a transaction flow of an account provided by the user in advance, and the related data being data related to the transaction flow data; feature extraction is at least carried out on the multi-modal transaction data to obtain multiple target features, feature fusion is carried out on the multiple target features based on an attention mechanism to obtain fused features, and the target features are used for representing key flow and operation indexes of the multi-modal transaction data; according to the fusion features, credit evaluation is carried out on the account, a credit evaluation result is obtained, and the evaluation result comprises the key flow of the account and the operation condition of the account. According to the method, the accuracy of account credit evaluation can be improved.

Description

Technical Field

[0001] This application relates to the financial field, and more specifically, to an account credit assessment method, an account credit assessment device, a computer-readable storage medium, and a processor. Background Technology

[0002] Existing account credit assessment methods are mainly based on transaction flow analysis. Current transaction flow analysis is primarily done manually, which has significant limitations. It is not only time-consuming but also affects the quality of the final analysis, greatly increasing management difficulty and labor costs.

[0003] Therefore, there is an urgent need for a method to improve the accuracy of account credit assessment. Summary of the Invention

[0004] The main objective of this application is to provide an account credit assessment method, an account credit assessment device, a computer-readable storage medium, and a processor, so as to at least solve the problem of how to improve the accuracy of account credit assessment in the prior art.

[0005] To achieve the above objectives, according to one aspect of this application, an account credit assessment method is provided, comprising: acquiring multimodal transaction data of a user, wherein the multimodal transaction data includes transaction flow data and multiple related data, the transaction flow data being the transaction flow data of the user's pre-provided account, and the related data being data related to the transaction flow data; extracting features from at least the multimodal transaction data to obtain multiple target features, and fusing the multiple target features based on an attention mechanism to obtain fused features, wherein the target features are used to characterize the key flow and operating indicators of the multimodal transaction data; and conducting a credit assessment of the account based on the fused features to obtain a credit assessment result, wherein the assessment result includes the key flow of the account and the operating status of the account.

[0006] Optionally, acquiring the user's multimodal transaction data includes: acquiring structured table data related to the transaction flow data, invoice images related to the transaction flow data, and text and audio data related to the transaction flow data to obtain the related data; inputting the transaction flow data and multiple related data into a multi-channel encoder to uniformly represent the transaction flow data and multiple related data to obtain the multimodal transaction data.

[0007] Optionally, the transaction flow data and multiple related data are input into a multi-channel encoder to uniformly represent the transaction flow data and multiple related data to obtain the multimodal transaction data. This includes: performing feature transformation on the transaction flow data and multiple related data using encoding functions to obtain a first function and multiple second functions, wherein the first function is the feature-transformed transaction flow data, and the second function is the feature-transformed related data; obtaining a mapping matrix, wherein the mapping matrix is ​​used to adjust the spatial distribution of the first function and multiple second functions to unify the logic of the transaction flow data and the related data; calculating the product of each related data with the mapping matrix to obtain multiple first products; calculating the norm square of the difference between the first function and each first product to obtain multiple first squared values; calculating the sum of the multiple first squared values ​​to obtain a loss function, and determining the transaction flow data and multiple related data corresponding to the minimum value of the loss function as the multimodal transaction data.

[0008] Optionally, feature extraction is performed on the multimodal transaction data to obtain multiple target features, including: performing consistency verification on the multimodal transaction data; if the consistency verification fails, performing error comparison on each transaction data in the multimodal transaction data to obtain missing fields and abnormal entries, and supplementing the abnormal entries using statistical imputation and the missing fields using semantic imputation; if the consistency verification succeeds, performing data cleaning on the multimodal transaction data to obtain first data; and performing feature extraction on the first data using a residual fusion algorithm to obtain the target features.

[0009] Optionally, the multimodal transaction data is cleaned to obtain first data, including: a calculation step, calculating the confidence level of the multimodal transaction data and comparing the confidence level with a confidence threshold; a processing step, if the confidence level is less than the confidence threshold, resampling or semantic replacement processing is performed on the multimodal transaction data to obtain second data; and a repetition step, repeating the calculation step and the processing step at least once, and updating the multimodal transaction data in the calculation step with the second data obtained in the processing step in the previous repetition, until the confidence level is greater than or equal to the confidence threshold, and determining the second data corresponding to the confidence level greater than or equal to the confidence threshold as the first data.

[0010] Optionally, feature fusion is performed on multiple target features based on an attention mechanism to obtain fused features, including: calculating the correlation coefficient between the multimodal transaction data and the corresponding target features to obtain multiple correlation coefficients; normalizing each correlation coefficient to obtain multiple attention weights; calculating the product of each attention weight and the corresponding target feature to obtain multiple second products; and calculating the sum of the multiple second products to obtain the fused features.

[0011] Optionally, based on the fusion features, a credit assessment is performed on the account to obtain a credit assessment result, including: based on the fusion features, obtaining large transaction flows, periodic transaction flows, and abnormal transaction flows from the transaction flow data to obtain the key flows; based on the key flows, determining the account's operating status parameters, including capital turnover efficiency, operating stability coefficient, related transaction ratio, and risk hazard value; calculating the norm square of the difference between each operating status parameter and its corresponding average value to obtain multiple second square values; calculating the sum of the multiple second square values ​​to obtain a loss function value; determining the operating status parameter corresponding to the minimum loss function value as the target parameter to obtain the operating status.

[0012] According to another aspect of this application, an account credit assessment device is provided, comprising: an acquisition unit for acquiring multimodal transaction data of a user, wherein the multimodal transaction data includes transaction flow data and multiple related data, the transaction flow data being the transaction flow data of the user's pre-provided account, and the related data being other data related to the transaction flow data; an extraction unit for performing feature extraction on at least the multimodal transaction data to obtain multiple target features, and performing feature fusion on the multiple target features based on an attention mechanism to obtain fused features, wherein the target features are used to characterize the key flow and operating indicators of the multimodal transaction data; and an assessment unit for performing credit assessment on the account based on the fused features to obtain a credit assessment result, wherein the assessment result includes the key flow of the account and the operating status of the account.

[0013] According to another aspect of this application, a computer-readable storage medium is provided, the computer-readable storage medium including a stored program, wherein, when the program is executed, it controls the device on which the computer-readable storage medium is located to perform any of the methods described.

[0014] According to another aspect of this application, a processor is provided for running a program, wherein the program, when running, performs any of the methods described.

[0015] Applying the technical solution of this application, firstly, multimodal transaction data of the user is acquired. This multimodal transaction data includes transaction flow data and multiple related data. The transaction flow data is the pre-provided transaction flow data of the user's account, and the related data is data associated with the transaction flow data. Next, features are extracted from the multimodal transaction data to obtain multiple target features. These target features are then fused using an attention mechanism to obtain fused features. The target features are used to characterize the key flows and operational indicators of the multimodal transaction data. Then, based on the fused features, a credit assessment is performed on the account to obtain a credit assessment result. The assessment result includes the account's key flows and the account's operational status. This provides a customer transaction flow analysis method based on a multimodal fusion algorithm, and then generates a credit assessment result based on this transaction flow analysis. Firstly, using the customer-provided transaction flow as the core input, intelligent interaction, dynamic verification, and real-time correlation of multimodal data are achieved. The multimodal fusion algorithm improves the accuracy of key flow extraction and operational status analysis, ultimately achieving accurate account credit assessment. Attached Figure Description

[0016] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0017] Figure 1 A hardware structure block diagram of a mobile terminal for performing an account credit assessment method according to an embodiment of this application is shown;

[0018] Figure 2 A flowchart illustrating an account credit assessment method according to an embodiment of this application is shown;

[0019] Figure 3 A schematic diagram illustrating a specific process of an account credit assessment method according to an embodiment of this application is shown;

[0020] Figure 4 A schematic diagram illustrating a multimodal transaction data docking and alignment process according to an embodiment of this application is shown.

[0021] Figure 5 A schematic diagram of a data cleaning and feature extraction process according to an embodiment of this application is shown;

[0022] Figure 6 A schematic diagram of a closed-loop process for report output and MCPAgent interaction response provided according to an embodiment of this application is shown;

[0023] Figure 7 A structural block diagram of an account credit assessment device provided according to an embodiment of this application is shown.

[0024] The above figures include the following reference numerals:

[0025] 102. Processor; 104. Memory; 106. Transmission device; 108. Input / output device. Detailed Implementation

[0026] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0027] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0028] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of this application described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0029] For ease of description, the following explains some of the nouns or terms used in the embodiments of this application:

[0030] MCPAgent: This refers to the core intelligent component in this solution used to realize multimodal data collaborative processing, intelligent analysis and interactive response. It can complete the entire process of tasks such as unified representation, dynamic verification, cleaning and extraction, analysis report generation and user feedback of multi-source heterogeneous transaction data.

[0031] As described in the background section, the prior art suffers from inaccurate account credit assessments. To address these issues, embodiments of this application provide an account credit assessment method, an account credit assessment device, a computer-readable storage medium, and a processor.

[0032] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0033] The methods and embodiments provided in this application can be executed on a mobile terminal, computer terminal, or similar computing device. Taking running on a mobile terminal as an example, Figure 1 This is a hardware structure block diagram of a mobile terminal for an account credit assessment method according to an embodiment of the present invention. Figure 1 As shown, a mobile terminal may include one or more ( Figure 1 Only one is shown in the diagram. A processor 102 (which may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.) and a memory 104 for storing data are also shown. The mobile terminal may further include a transmission device 106 for communication functions and an input / output device 108. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the mobile terminal described above. For example, the mobile terminal may also include components that are more... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0034] The memory 104 can be used to store computer programs, such as application software programs and modules, like the computer program corresponding to the account credit assessment method in this embodiment of the invention. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, thereby implementing the above-described method. The memory 104 may include high-speed random access memory and non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the mobile terminal via a network. Examples of the aforementioned networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof. The transmission device 106 is used to receive or send data via a network. Specific examples of the aforementioned networks may include wireless networks provided by the mobile terminal's communication provider. In one example, the transmission device 106 includes a network interface controller (NIC), which can be connected to other network devices via a base station to communicate with the Internet. In one example, the transmission device 106 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0035] This embodiment provides an account credit assessment method that runs on a mobile terminal, computer terminal, or similar computing device. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Also, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0036] Figure 2 This is a flowchart of an account credit assessment method according to an embodiment of this application. Figure 2 As shown, the method includes the following steps:

[0037] Step S201: Obtain the user's multimodal transaction data, wherein the multimodal transaction data includes transaction flow data and multiple related data, the transaction flow data is the transaction flow of the user's pre-provided account, and the related data is data related to the transaction flow data.

[0038] Step S202: At least the above-mentioned multimodal transaction data is subjected to feature extraction to obtain multiple target features, and the multiple target features are fused based on the attention mechanism to obtain fused features, wherein the target features are used to characterize the key flow and operating indicators of the above-mentioned multimodal transaction data;

[0039] Step S203: Based on the above-mentioned fusion characteristics, perform a credit assessment on the above-mentioned account to obtain a credit assessment result, wherein the assessment result includes the key transaction data of the above-mentioned account and the operating status of the above-mentioned account.

[0040] Specifically, the above method can be implemented using MCPAgent, such as... Figure 3 As shown, the above steps can be achieved through the following methods: inputting multimodal data (including transaction records provided by customers) and connecting it with MCPAgent; data cleaning and cross-modal feature extraction assisted by MCPAgent (focusing on consistency verification of transaction records and related data); feature fusion based on attention mechanism (MCPAgent dynamically adjusts feature weights to strengthen key transaction recognition dimensions); key transaction extraction, business status analysis and report generation (MCPAgent completes transaction classification, anomaly identification, business indicator calculation and report compliance check); analysis report output and interactive response with MCPAgent (supporting users to query key transaction basis, modify analysis dimensions, and update report request processing).

[0041] This embodiment first acquires the user's multimodal transaction data, including transaction flow data and multiple related data. The transaction flow data is the pre-provided transaction flow data of the user's account, and the related data is data associated with the transaction flow data. Next, features are extracted from the multimodal transaction data to obtain multiple target features. These target features are then fused using an attention mechanism to obtain fused features. The target features are used to characterize the key flow and operational indicators of the multimodal transaction data. Then, based on the fused features, a credit assessment is performed on the account to obtain a credit assessment result, which includes the account's key flow and operational status. This provides a customer transaction flow analysis method based on a multimodal fusion algorithm, and generates a credit assessment result based on this transaction flow analysis. First, using the customer-provided transaction flow as the core input, intelligent interaction, dynamic verification, and real-time correlation of multimodal data are achieved. The multimodal fusion algorithm improves the accuracy of key flow extraction and operational status analysis, ultimately achieving accurate account credit assessment.

[0042] In specific implementation, step S201 can be achieved through the following steps: Step S2011, acquire structured table data related to the transaction flow data, invoice images related to the transaction flow data, and text and audio data related to the transaction flow data to obtain the aforementioned related data; Step S2012, input the aforementioned transaction flow data and multiple related data into a multi-channel encoder to uniformly represent the aforementioned transaction flow data and multiple related data to obtain the aforementioned multimodal transaction data. This method can further accelerate the acquisition of the aforementioned multimodal transaction data.

[0043] Specifically, such as Figure 4 As shown, multimodal data sources are input into the MCPAgent multi-channel encoder. The multimodal data sources include core data and related data. The core data is the fund transaction flow, and the related data includes structured tables, semi-structured ticket images, and unstructured text or audio. Then, the data of different modalities are mapped to the shared latent space respectively. Based on timestamps, fund flow, and cross-modal alignment of counterparties, high-quality multi-source data input is obtained.

[0044] To further accelerate the unified representation of multi-source data, step S2012 of this application can be implemented through the following steps: Step S20121, using an encoding function to perform feature transformation on the transaction flow data and multiple related data respectively, to obtain a first function and multiple second functions, wherein the first function is the transaction flow data after feature transformation, and the second function is the related data after feature transformation; Step S20122, obtaining a mapping matrix, wherein the mapping matrix is ​​used to adjust the spatial distribution of the first function and multiple second functions to unify the logic of the transaction flow data and related data; Step S20123, calculating the product of each related data and the mapping matrix respectively, to obtain multiple first products; Step S20124, calculating the norm square of the difference between the first function and each first product, to obtain multiple first squared values; Step S20125, calculating the sum of the multiple first squared values ​​to obtain a loss function, and determining the transaction flow data and multiple related data corresponding to the minimum value of the loss function as the multimodal transaction data.

[0045] Specifically, the above encoding function can be: ,in, It can be either the first function or the second function. For transaction flow data or related data, The above encoding function is used. The formula for the above loss function can be: , where W is the cross-modal mapping matrix, used to adjust the spatial distribution of flow characteristics and associated data characteristics to make them consistent at the level of fund transaction logic. By minimizing This enables dynamic alignment of transaction logs with multimodal associated data, ensuring the stability and accuracy of subsequent feature fusion. Simultaneously, MCPAgent employs a streaming access mechanism to perform real-time incremental mapping updates on newly added transaction logs and associated data, ensuring the continuity and real-time nature of the input data. After integration, the system provides a high-quality input foundation, centered on transaction logs and featuring multi-source data collaboration, for the next stage of data cleaning and cross-modal feature extraction.

[0046] Step S202 can also be implemented in other ways, such as: Step S2021, performing consistency verification on the multimodal transaction data; Step S2022, if the consistency verification fails, comparing the errors of each transaction data in the multimodal transaction data to obtain missing fields and abnormal entries, and using statistical imputation to supplement the abnormal entries and semantic imputation to supplement the missing fields; Step S2023, if the consistency verification succeeds, performing data cleaning on the multimodal transaction data to obtain the first data; Step S2024, using a residual fusion algorithm to extract features from the first data to obtain the target features. The above method can further improve the accuracy of obtaining the target features.

[0047] Specifically, the above residual fusion algorithm can be used for cross-modal extraction using the following formula: ,in, To assign higher base weights to flow characteristics, To determine the number of feature vectors, modal complementarity is preserved, while the difference term is used. Strengthen key factors, To enhance the coefficients of key factors and avoid feature dilution, the final fused feature F focuses on key flow identification and operational analysis dimensions, providing a high-confidence, low-redundancy input foundation for the subsequent attention-based feature fusion stage.

[0048] In some embodiments, step S2023 can be implemented through the following steps: Step S20231, a calculation step, calculating the confidence level of the multimodal transaction data and comparing the confidence level with a confidence threshold; Step S20232, a processing step, when the confidence level is less than the confidence threshold, resampling or semantic replacement processing is performed on the multimodal transaction data to obtain second data; Step S20233, a repetition step, repeating the calculation step and the processing step at least once, and updating the multimodal transaction data in the calculation step with the second data obtained in the processing step in the previous repetition, until the confidence level is greater than or equal to the confidence threshold, and determining the second data corresponding to the confidence level greater than or equal to the confidence threshold as the first data. This method can further improve the accuracy of data cleaning.

[0049] The confidence level mentioned above can be calculated using the following formula: , Feature vectors that primarily represent transaction flow patterns It is a weighted average of transaction flow data and multiple related data mentioned above. This is used to measure the consistency between transaction flow features and the overall transaction semantics. If the confidence level is below a threshold, MCPAgent triggers an adaptive cleaning mechanism to resample or semantically replace the transaction flow features, ensuring the logical integrity and accuracy of the transaction flow data. The flowchart for the above data cleaning and cross-modal feature extraction is as follows: Figure 5 As shown, modal consistency detection is performed on the aligned multimodal features to identify missing or abnormal entries in the transaction flow. These entries are then supplemented using statistical imputation and semantic completion. Next, data confidence is measured. If the confidence is less than the threshold, adaptive cleaning is performed using resampling or semantic replacement until the confidence is greater than or equal to the threshold. Valid features are then retained. Finally, residual fusion operators are used to process and extract key transaction flow identification factors and operational status correlation factors.

[0050] Step S202 above can also be implemented in other ways, such as: step S2025, calculating the correlation coefficients between the multimodal transaction data and the corresponding target features to obtain multiple correlation coefficients; step S2026, normalizing each correlation coefficient to obtain multiple attention weights; step S2027, calculating the product of each attention weight and the corresponding target feature to obtain multiple second products; step S2028, calculating the sum of the multiple second products to obtain the fused feature. The above methods can further improve the accuracy of feature fusion.

[0051] Specifically, after cross-modal feature cleaning and extraction, MCPAgent performs deep fusion of multimodal features through an attention mechanism to ensure the accuracy and interpretability of key flow extraction and operational status analysis. Since different modal features contribute differently to key flow identification and operational indicator calculation, MCPAgent constructs a dynamic weight allocation function to further weight the input fused features according to the analysis objectives. Specifically, let the feature set after cross-modal cleaning be... MCPAgent first calculates the correlation score between each modal feature and the global analysis semantic vector F, defined as: , where U is a trainable correlation mapping matrix, This represents the matching strength of the i-th mode for key flow extraction and operational status assessment. for The transpose of the matrix. Then, MCPAgent normalizes the scores to obtain the dynamic attention weights: This ensures that modes that contribute significantly to the analysis objective receive higher weights. This represents the matching strength of the j-th modality for key flow extraction and operational status assessment. The final fusion output is represented as: In this process, the core features of high-weighted modes are enhanced, while low-weighted modes provide supplementary information. This process is achieved through weighting... The traceability is ensured, clarifying the contribution of each modality to the analysis results, and guaranteeing the semantic consistency, robustness, and interpretability of the fused representation Z. This fused representation will serve as input to the key transaction extraction, operational status analysis, and report generation stages, supporting subsequent transaction classification, anomaly identification, and compliance checks.

[0052] Step S203 above can be achieved through the following steps: Step S2031, based on the aforementioned fusion characteristics, obtain large-amount transaction flows, periodic transaction flows, and abnormal transaction flows from the aforementioned transaction flow data to obtain the aforementioned key flows; Step S2032, based on the aforementioned key flows, determine the operating status parameters of the aforementioned account, including capital turnover efficiency, operating stability coefficient, related-party transaction ratio, and risk hazard value; Step S2033, calculate the norm square of the difference between each of the aforementioned operating status parameters and its corresponding average value to obtain multiple second-square values; Step S2034, calculate the sum of the multiple second-square values ​​to obtain a loss function value, and determine the operating status parameter corresponding to the minimum loss function value as the target parameter to obtain the aforementioned operating status. The above method can further enrich the aforementioned credit assessment results.

[0053] After obtaining the fusion representation Z, MCPAgent enters the key transaction extraction, business status analysis and report generation stage. Its core objective is to extract key items from the original transaction data provided by the client, calculate business evaluation indicators, and generate structured analysis reports.

[0054] This stage is divided into three core sub-steps:

[0055] Key Flow Extraction: MCPAgent constructs a key flow identification function Based on the features in the fused representation Z, from the original flow Filter key entries: ,in This includes large transaction records, periodic operating records, and abnormal transaction records. By integrating transaction rule constraints with feature matching logic, the extracted results are made more targeted.

[0056] Business Performance Analysis: MCPAgent Builds a Model for Calculating Business Indicators Based on key flow Calculate core operating indicators, including capital turnover efficiency, operational stability, related-party transaction ratio, and potential risks, to generate an operational assessment result. ,in For capital turnover efficiency, For operational stability, For the proportion of related-party transactions, This represents the risk hazard value. Simultaneously, MCPAgent introduces an operational assessment loss function. ,in The average operating indicators of companies of similar size and industry are minimized. Ensure the industry comparability of business analysis results;

[0057] Analysis report generation: MCPAgent builds report generation functions , will key flow Integrate with operational assessment result A into a structured report The report contains three core parts: ① a list of key transaction records; ② an analysis of the operating conditions; and ③ risk warnings.

[0058] After completing the key transaction data extraction, operational status analysis, and report generation, MCPAgent is responsible for delivering the final analysis report. , For key transaction lists, For business performance analysis, To provide risk alerts and enable real-time interactive responses with users, the system supports querying key transaction data, modifying analysis dimensions, and updating report requests. The system assigns dynamic response weights to each module of the report. ,in For module data confidence, To ensure compliance risk, modules with high confidence and low risk are prioritized for responding to user queries. When a user initiates a modification request, MCPAgent triggers an incremental update process: ① Re-extract key transaction flows based on the new threshold. ② Update operating indicators synchronously ③ Generate an updated report And through the incremental mapping function To ensure update efficiency, among which Update the weights for the module. The vector is adjusted for module features. This process ensures that the modified report is logically consistent with the original data while also responding to user needs in real time, achieving closed-loop management of "query-feedback-update," and providing financial institutions with interactive and traceable customer fund transaction analysis services. The closed-loop diagram of the report output and MCPAgent interaction response is shown below. Figure 6 As shown, the analysis report output ( / / User operations include querying key transaction data, modifying analysis dimensions, and requesting report updates. Querying key transaction data can be dynamically responded to by MCPAgent (prioritizing high-confidence, low-risk modules). Modifying analysis dimensions and requesting report updates can be triggered by MCPAgent to perform incremental updates, re-extract key transaction data, update business indicators, and generate updated reports, ultimately achieving closed-loop management and meeting user needs.

[0059] This application also provides an account credit assessment device. It should be noted that the account credit assessment device of this application can be used to execute the account credit assessment method provided in this application. This device is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0060] The following describes the account credit assessment device provided in the embodiments of this application.

[0061] Figure 7 This is a schematic diagram of an account credit assessment device according to an embodiment of this application. Figure 7 As shown, the device includes:

[0062] The acquisition unit 10 is used to acquire the user's multimodal transaction data, wherein the multimodal transaction data includes transaction flow data and multiple related data, the transaction flow data is the transaction flow of the user's pre-provided account, and the related data is data related to the transaction flow data.

[0063] The extraction unit 20 is used to extract features from the multimodal transaction data to obtain multiple target features, and to fuse the multiple target features based on an attention mechanism to obtain fused features, wherein the target features are used to characterize the key flow and operating indicators of the multimodal transaction data.

[0064] The assessment unit 30 is used to conduct a credit assessment on the account based on the aforementioned fusion characteristics and obtain a credit assessment result, wherein the assessment result includes the account's key transaction history and the account's operating status.

[0065] Specifically, the above method can be implemented using MCPAgent, such as... Figure 3As shown, the above steps can be achieved through the following methods: inputting multimodal data (including transaction records provided by customers) and connecting it with MCPAgent; data cleaning and cross-modal feature extraction assisted by MCPAgent (focusing on consistency verification of transaction records and related data); feature fusion based on attention mechanism (MCPAgent dynamically adjusts feature weights to strengthen key transaction recognition dimensions); key transaction extraction, business status analysis and report generation (MCPAgent completes transaction classification, anomaly identification, business indicator calculation and report compliance check); analysis report output and interactive response with MCPAgent (supporting users to query key transaction basis, modify analysis dimensions, and update report request processing).

[0066] In this embodiment, the acquisition unit acquires the user's multimodal transaction data, which includes transaction flow data and multiple related data. The transaction flow data is the pre-provided transaction flow data of the user's account, and the related data is data related to the transaction flow data. The extraction unit extracts features from the multimodal transaction data to obtain multiple target features, and fuses these target features based on an attention mechanism to obtain fused features. The target features are used to characterize the key flow and operational indicators of the multimodal transaction data. The evaluation unit performs a credit assessment on the account based on the fused features to obtain a credit assessment result, which includes the account's key flow and operational status. This provides a customer transaction flow analysis method based on a multimodal fusion algorithm, and then generates a credit assessment result based on this transaction flow analysis. First, using the customer-provided transaction flow as the core input, intelligent interaction, dynamic verification, and real-time correlation of multimodal data are achieved. The multimodal fusion algorithm is combined to improve the accuracy of key flow extraction and operational status analysis, ultimately achieving accurate account credit assessment.

[0067] In its specific implementation, the aforementioned acquisition unit includes a first acquisition module and an input module. The first acquisition module acquires structured table data related to the transaction flow data, invoice images related to the transaction flow data, and text and audio data related to the transaction flow data, thus obtaining the aforementioned related data. The input module inputs the transaction flow data and multiple related data into a multi-channel encoder to uniformly represent the transaction flow data and multiple related data, thereby obtaining the aforementioned multimodal transaction data. This method can further accelerate the acquisition of the aforementioned multimodal transaction data.

[0068] Specifically, such as Figure 4As shown, the multimodal data source is input into the MCPAgent multi-channel encoder. The multimodal data source includes core data and related data. The core data is the fund transaction flow, and the related data includes structured tables, semi-structured ticket images, and unstructured text or audio. Then, the data of different modalities are mapped to the shared latent space and cross-modal alignment is performed to obtain high-quality multi-source data input.

[0069] To further accelerate the unified representation of multi-source data, the input module of this application includes a transformation submodule, an acquisition submodule, a first calculation submodule, a second calculation submodule, and a third calculation submodule. The transformation submodule uses encoding functions to perform feature transformations on the transaction flow data and multiple related data, respectively, to obtain a first function and multiple second functions. The first function represents the feature-transformed transaction flow data, and the second functions represent the feature-transformed related data. The acquisition submodule acquires a mapping matrix, which is used to adjust the spatial distribution of the first function and the multiple second functions to unify the logic of the transaction flow data and the related data. The first calculation submodule calculates the product of each related data point with the mapping matrix, obtaining multiple first products. The second calculation submodule calculates the norm square of the difference between the first function and each first product, obtaining multiple first squared values. The third calculation submodule calculates the sum of the multiple first squared values ​​to obtain a loss function, and determines the transaction flow data and multiple related data corresponding to the minimum value of the loss function as the multimodal transaction data.

[0070] Specifically, the above encoding function can be: ,in, It can be either the first function or the second function. For transaction flow data or related data, The above encoding function is used. The formula for the above loss function can be: , where W is the cross-modal mapping matrix, used to adjust the spatial distribution of flow characteristics and associated data characteristics to make them consistent at the level of fund transaction logic. By minimizing This enables dynamic alignment of transaction logs with multimodal associated data, ensuring the stability and accuracy of subsequent feature fusion. Simultaneously, MCPAgent employs a streaming access mechanism to perform real-time incremental mapping updates on newly added transaction logs and associated data, ensuring the continuity and real-time nature of the input data. After integration, the system provides a high-quality input foundation, centered on transaction logs and featuring multi-source data collaboration, for the next stage of data cleaning and cross-modal feature extraction.

[0071] The extraction unit includes a verification module, a comparison module, a cleaning module, and an extraction module. The verification module performs consistency verification on the multimodal transaction data. The comparison module, in the event of a failed consistency verification, performs error comparison on each transaction data point within the multimodal transaction data to identify missing fields and abnormal entries. It then uses statistical imputation to fill in the abnormal entries and semantic imputation to fill in the missing fields. The cleaning module, in the event of a successful consistency verification, performs data cleaning on the multimodal transaction data to obtain first data. The extraction module uses a residual fusion algorithm to extract features from the first data to obtain the target features. This device can further improve the accuracy of target feature acquisition.

[0072] Specifically, the above residual fusion algorithm can be used for cross-modal extraction using the following formula: ,in, To assign higher base weights to flow characteristics, To determine the number of feature vectors, modal complementarity is preserved, while the difference term is used. Strengthen key factors, To enhance the coefficients of key factors and avoid feature dilution, the final fused feature F focuses on key flow identification and operational analysis dimensions, providing a high-confidence, low-redundancy input foundation for the subsequent attention-based feature fusion stage.

[0073] In some embodiments, the cleaning module includes a fourth calculation submodule, a processing submodule, and a repetition submodule. The fourth calculation submodule performs a calculation step, calculating the confidence level of the multimodal transaction data and comparing the confidence level with a confidence threshold. The processing submodule performs a processing step, resampling or semantically replacing the multimodal transaction data when the confidence level is less than the confidence threshold to obtain second data. The repetition submodule performs a repetition step, sequentially repeating the calculation and processing steps at least once, updating the multimodal transaction data in the calculation step with the second data obtained in the processing step during the previous repetition, until the confidence level is greater than or equal to the confidence threshold. The second data corresponding to the confidence level greater than or equal to the confidence threshold is then determined as the first data. This device can further improve the accuracy of data cleaning.

[0074] The confidence level mentioned above can be calculated using the following formula: , Feature vectors that primarily represent transaction flow patterns It is a weighted average of transaction flow data and multiple related data mentioned above. This is used to measure the consistency between transaction flow features and the overall transaction semantics. If the confidence level is below a threshold, MCPAgent triggers an adaptive cleaning mechanism to resample or semantically replace the transaction flow features, ensuring the logical integrity and accuracy of the transaction flow data. The flowchart for the above data cleaning and cross-modal feature extraction is as follows: Figure 5 As shown, modal consistency detection is performed on the aligned multimodal features to identify missing or abnormal entries in the flow chart. These entries are then supplemented using statistical imputation and semantic completion. Next, data confidence is measured. If the confidence is less than the threshold, adaptive cleaning is performed using resampling or semantic replacement until the confidence is greater than or equal to the threshold. Valid features are then retained. Finally, residual fusion operators are used to process and extract key flow chart identification factors and operational status correlation factors.

[0075] The extraction unit further includes a first calculation module, a processing module, a second calculation module, and a third calculation module. The first calculation module calculates the correlation coefficients between the multimodal transaction data and the corresponding target features, obtaining multiple correlation coefficients. The processing module normalizes each correlation coefficient, obtaining multiple attention weights. The second calculation module calculates the product of each attention weight and the corresponding target feature, obtaining multiple second products. The third calculation module calculates the sum of the multiple second products, obtaining the fused feature. This device can further improve the accuracy of feature fusion.

[0076] Specifically, after cross-modal feature cleaning and extraction, MCPAgent performs deep fusion of multimodal features through an attention mechanism to ensure the accuracy and interpretability of key flow extraction and operational status analysis. Since different modal features contribute differently to key flow identification and operational indicator calculation, MCPAgent constructs a dynamic weight allocation function to further weight the input fused features according to the analysis objectives. Specifically, let the feature set after cross-modal cleaning be... MCPAgent first calculates the correlation score between each modal feature and the global analysis semantic vector F, defined as: , where U is a trainable correlation mapping matrix, This represents the matching strength of the i-th mode for key flow extraction and operational status assessment. for The transpose of the matrix. Then, MCPAgent normalizes the scores to obtain the dynamic attention weights: This ensures that modes that contribute significantly to the analysis objective receive higher weights. This represents the matching strength of the j-th modality for key flow extraction and operational status assessment. The final fusion output is represented as: In this process, the core features of high-weighted modes are enhanced, while low-weighted modes provide supplementary information. This process is achieved through weighting... The traceability is ensured, clarifying the contribution of each modality to the analysis results, and guaranteeing the semantic consistency, robustness, and interpretability of the fused representation Z. This fused representation will serve as input to the key transaction extraction, operational status analysis, and report generation stages, supporting subsequent transaction classification, anomaly identification, and compliance checks.

[0077] The aforementioned assessment unit includes a second acquisition module, a determination module, a fourth calculation module, and a fifth calculation module. The second acquisition module is used to acquire large-amount transaction flows, periodic transaction flows, and abnormal transaction flows from the aforementioned transaction flow data based on the aforementioned fusion characteristics, thus obtaining the aforementioned key flows. The determination module is used to determine the operating status parameters of the aforementioned account based on the aforementioned key flows. The operating status parameters include capital turnover efficiency, operating stability coefficient, related-party transaction ratio, and risk hazard value. The fourth calculation module is used to calculate the norm square of the difference between each of the aforementioned operating status parameters and its corresponding average value, obtaining multiple second square values. The fifth calculation module is used to calculate the sum of the multiple second square values, obtaining a loss function value, and determining the operating status parameter corresponding to the minimum loss function value as the target parameter, thus obtaining the aforementioned operating status. The aforementioned device can further enrich the aforementioned credit assessment results.

[0078] After obtaining the fusion representation Z, MCPAgent enters the key transaction extraction, business status analysis and report generation stage. Its core objective is to extract key items from the original transaction data provided by the client, calculate business evaluation indicators, and generate structured analysis reports.

[0079] This stage is divided into three core sub-steps:

[0080] Key Flow Extraction: MCPAgent constructs a key flow identification function Based on the features in the fused representation Z, from the original flow Filter key entries: ,in This includes large transaction records, periodic operating records, and abnormal transaction records. By integrating transaction rule constraints with feature matching logic, the extracted results are made more targeted.

[0081] Business Performance Analysis: MCPAgent Builds a Model for Calculating Business Indicators Based on key flow Calculate core operating indicators, including capital turnover efficiency, operational stability, related-party transaction ratio, and potential risks, to generate an operational assessment result. ,in For capital turnover efficiency, For operational stability, For the proportion of related-party transactions, This represents the risk hazard value. Simultaneously, MCPAgent introduces an operational assessment loss function. ,in The average operating indicators of companies of similar size and industry are minimized. Ensure the industry comparability of business analysis results;

[0082] Analysis report generation: MCPAgent builds report generation functions , will key flow Integrate with operational assessment result A into a structured report The report contains three core parts: ① a list of key transaction records; ② an analysis of the operating conditions; and ③ risk warnings.

[0083] After completing the key transaction data extraction, operational status analysis, and report generation, MCPAgent is responsible for delivering the final analysis report. , For key transaction lists, For business performance analysis, To provide risk alerts and enable real-time interactive responses with users, the system supports querying key transaction data, modifying analysis dimensions, and updating report requests. The system assigns dynamic response weights to each module of the report. ,in For module data confidence, To ensure compliance risk, modules with high confidence and low risk are prioritized for responding to user queries. When a user initiates a modification request, MCPAgent triggers an incremental update process: ① Re-extract key transaction flows based on the new threshold. ② Update operating indicators synchronously ③ Generate an updated report And through the incremental mapping function To ensure update efficiency, among which Update the weights for the module. The vector is adjusted for module features. This process ensures that the modified report is logically consistent with the original data while also responding to user needs in real time, achieving closed-loop management of "query-feedback-update," and providing financial institutions with interactive and traceable customer fund transaction analysis services. The closed-loop diagram of the report output and MCPAgent interaction response is shown below. Figure 6 As shown, the analysis report output ( / / User operations include querying key transaction data, modifying analysis dimensions, and requesting report updates. Querying key transaction data can be dynamically responded to by MCPAgent (prioritizing high-confidence, low-risk modules). Modifying analysis dimensions and requesting report updates can be triggered by MCPAgent to perform incremental updates, re-extract key transaction data, update business indicators, and generate updated reports, ultimately achieving closed-loop management and meeting user needs.

[0084] The aforementioned account credit assessment device includes a processor and a memory. The acquisition unit, extraction unit, and assessment unit are all stored as program units in the memory, and the processor executes these program units to achieve their respective functions. All of the above modules reside in the same processor; alternatively, the modules may be located in different processors in any combination.

[0085] The processor contains a kernel, which retrieves the corresponding program units from memory. One or more kernels can be configured, and account credit can be assessed by adjusting kernel parameters.

[0086] The memory may include non-permanent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0087] This invention provides a computer-readable storage medium including a stored program, wherein, when the program is executed, it controls the device where the computer-readable storage medium is located to perform the account credit assessment method.

[0088] Specifically, account credit assessment methods include:

[0089] Step S201: Obtain the user's multimodal transaction data, wherein the multimodal transaction data includes transaction flow data and multiple related data, the transaction flow data is the transaction flow of the user's pre-provided account, and the related data is data related to the transaction flow data.

[0090] Step S202: At least the above-mentioned multimodal transaction data is subjected to feature extraction to obtain multiple target features, and the multiple target features are fused based on the attention mechanism to obtain fused features, wherein the target features are used to characterize the key flow and operating indicators of the above-mentioned multimodal transaction data;

[0091] Step S203: Based on the above-mentioned fusion characteristics, perform a credit assessment on the above-mentioned account to obtain a credit assessment result, wherein the assessment result includes the key transaction data of the above-mentioned account and the operating status of the above-mentioned account.

[0092] This invention provides a processor for running a program, wherein the program executes the account credit assessment method described above.

[0093] Specifically, account credit assessment methods include:

[0094] Step S201: Obtain the user's multimodal transaction data, wherein the multimodal transaction data includes transaction flow data and multiple related data, the transaction flow data is the transaction flow of the user's pre-provided account, and the related data is data related to the transaction flow data.

[0095] Step S202: At least the above-mentioned multimodal transaction data is subjected to feature extraction to obtain multiple target features, and the multiple target features are fused based on the attention mechanism to obtain fused features, wherein the target features are used to characterize the key flow and operating indicators of the above-mentioned multimodal transaction data;

[0096] Step S203: Based on the above-mentioned fusion characteristics, perform a credit assessment on the above-mentioned account to obtain a credit assessment result, wherein the assessment result includes the key transaction data of the above-mentioned account and the operating status of the above-mentioned account.

[0097] This invention provides a device including a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it performs at least the following steps:

[0098] Step S201: Obtain the user's multimodal transaction data, wherein the multimodal transaction data includes transaction flow data and multiple related data, the transaction flow data is the transaction flow of the user's pre-provided account, and the related data is data related to the transaction flow data.

[0099] Step S202: At least the above-mentioned multimodal transaction data is subjected to feature extraction to obtain multiple target features, and the multiple target features are fused based on the attention mechanism to obtain fused features, wherein the target features are used to characterize the key flow and operating indicators of the above-mentioned multimodal transaction data;

[0100] Step S203: Based on the above-mentioned fusion characteristics, perform a credit assessment on the above-mentioned account to obtain a credit assessment result, wherein the assessment result includes the key transaction data of the above-mentioned account and the operating status of the above-mentioned account.

[0101] The devices mentioned in this article can be servers, PCs, tablets, mobile phones, etc.

[0102] This application also provides a computer program product, which, when executed on a data processing device, is suitable for executing an initialization program having at least the following method steps:

[0103] Step S201: Obtain the user's multimodal transaction data, wherein the multimodal transaction data includes transaction flow data and multiple related data, the transaction flow data is the transaction flow of the user's pre-provided account, and the related data is data related to the transaction flow data.

[0104] Step S202: At least the above-mentioned multimodal transaction data is subjected to feature extraction to obtain multiple target features, and the multiple target features are fused based on the attention mechanism to obtain fused features, wherein the target features are used to characterize the key flow and operating indicators of the above-mentioned multimodal transaction data;

[0105] Step S203: Based on the above-mentioned fusion characteristics, perform a credit assessment on the above-mentioned account to obtain a credit assessment result, wherein the assessment result includes the key transaction data of the above-mentioned account and the operating status of the above-mentioned account.

[0106] It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. They can be implemented using computer-executable program code, and thus can be stored in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those described herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0107] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0108] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0109] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0110] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0111] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0112] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0113] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0114] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0115] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0116] As can be seen from the above description, the embodiments of this application achieve the following technical effects:

[0117] 1) The account credit assessment method of this application firstly acquires the user's multimodal transaction data, which includes transaction flow data and multiple related data. The transaction flow data is the pre-provided transaction flow of the user's account, and the related data is data related to the transaction flow data. Next, features are extracted from the multimodal transaction data to obtain multiple target features. These target features are then fused based on an attention mechanism to obtain fused features. The target features are used to characterize the key flows and operational indicators of the multimodal transaction data. Then, based on the fused features, the account is credit-assessed to obtain a credit assessment result, which includes the account's key flows and operational status. This provides a customer transaction flow analysis method based on a multimodal fusion algorithm, and then generates a credit assessment result based on this transaction flow analysis. Firstly, using the customer-provided transaction flow as the core input, intelligent interaction, dynamic verification, and real-time correlation of multimodal data are achieved. The multimodal fusion algorithm improves the accuracy of key flow extraction and operational status analysis, ultimately achieving accurate account credit assessment.

[0118] 2) The account credit assessment device of this application includes an acquisition unit that acquires multimodal transaction data of users, including transaction flow data and multiple related data. The transaction flow data is the transaction flow of the user's pre-provided account, and the related data is data related to the transaction flow data. The extraction unit extracts features from the multimodal transaction data to obtain multiple target features, and fuses these target features based on an attention mechanism to obtain fused features. The target features are used to characterize the key flow and operating indicators of the multimodal transaction data. The assessment unit performs credit assessment on the account based on the fused features to obtain a credit assessment result, which includes the account's key flow and operating status. This provides a customer transaction flow analysis method based on a multimodal fusion algorithm, and then generates a credit assessment result based on this transaction flow analysis. First, using the customer-provided transaction flow as the core input, intelligent interaction, dynamic verification, and real-time correlation of multimodal data are achieved. The multimodal fusion algorithm is combined to improve the accuracy of key flow extraction and the accuracy of operating status analysis, ultimately achieving accurate account credit assessment.

[0119] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method of account credit assessment, characterized by, The method comprises the following steps: obtaining multi-modal transaction data of a user, wherein the multi-modal transaction data comprises transaction flow data and a plurality of related data, the transaction flow data is transaction flow of an account provided in advance for the user, and the related data is data related to the transaction flow data; performing feature extraction on at least the multi-modal transaction data to obtain a plurality of target features, and performing feature fusion on the plurality of target features based on an attention mechanism to obtain a fusion feature, wherein the target features are used to represent key flow and business indicators of the multi-modal transaction data; performing credit evaluation on the account according to the fusion feature to obtain a credit evaluation result, wherein the evaluation result comprises key flow of the account and business status of the account.

2. The method of claim 1, wherein, Obtaining multi-modal transaction data of a user comprises: obtaining structured table data related to the transaction flow data, bill pictures related to the transaction flow data, and text data and audio data related to the transaction flow data to obtain the related data; inputting the transaction flow data and the plurality of related data into a multi-path encoder to uniformly represent the transaction flow data and the plurality of related data to obtain the multi-modal transaction data.

3. The method of claim 2, wherein, Inputting the transaction flow data and the plurality of related data into a multi-path encoder to uniformly represent the transaction flow data and the plurality of related data to obtain the multi-modal transaction data comprises: performing feature conversion on the transaction flow data and the plurality of related data respectively by using an encoding function to obtain a first function and a plurality of second functions, wherein the first function is the transaction flow data after feature conversion, and the second function is the related data after feature conversion; obtaining a mapping matrix, wherein the mapping matrix is used to adjust the spatial distribution of the first function and the plurality of second functions to unify the logic of the transaction flow data and the related data; calculating the product of each related data and the mapping matrix respectively to obtain a plurality of first products; calculating the norm square of the difference value of the first function and each first product to obtain a plurality of first square values; calculating the sum of the plurality of first square values to obtain a loss function, and determining that the transaction flow data and the plurality of related data corresponding to the minimum value of the loss function are the multi-modal transaction data.

4. The method of claim 1, wherein, Performing feature extraction on at least the multi-modal transaction data to obtain a plurality of target features comprises: performing consistency checking on the multi-modal transaction data; in the case that the consistency checking is unsuccessful, performing error comparison on each transaction data in the multi-modal transaction data to obtain missing fields and abnormal items, and supplementing the abnormal items by using a statistical filling method and supplementing the missing fields by using a semantic filling method; in the case that the consistency checking is successful, performing data cleaning on the multi-modal transaction data to obtain first data; performing feature extraction on the first data by using a residual fusion algorithm to obtain the target features.

5. The method of claim 4, wherein, Performing data cleaning on the multi-modal transaction data to obtain first data comprises: A computing step of computing a confidence of the multi-modal transaction data and comparing the confidence with a confidence threshold; A processing step of, in a case where the confidence is less than the confidence threshold, performing resampling processing or semantic replacement processing on the multi-modal transaction data to obtain second data; A repeating step of sequentially repeating the computing step and the processing step at least once, and updating, in a repeating process, the multi-modal transaction data in the computing step to the second data in the processing step obtained in a previous repeating process, until the confidence is greater than or equal to the confidence threshold, and determining that the second data corresponding to the confidence greater than or equal to the confidence threshold is the first data.

6. The method of claim 1, wherein, The feature fusion based on the attention mechanism includes: Computing a correlation coefficient of the multi-modal transaction data and the corresponding target feature to obtain a plurality of correlation coefficients; Respectively normalizing each correlation coefficient to obtain a plurality of attention weights; Computing a product of each attention weight and the corresponding target feature to obtain a plurality of second products; Computing a sum of the plurality of second products to obtain the fusion feature.

7. The method of claim 1, wherein, According to the fusion feature, the credit evaluation result of the account is obtained, including: According to the fusion feature, obtaining large transaction flow, periodic transaction flow and abnormal transaction flow in the transaction flow data to obtain the key flow; According to the key flow, determining the operating condition parameters of the account, the operating condition parameters including the fund turnover efficiency, the operating stability coefficient, the correlation transaction proportion and the risk hidden value; Computing the norm square of the difference between each operating condition parameter and the corresponding average value to obtain a plurality of second square values; Computing the sum of the plurality of second square values to obtain a loss function value, determining the target parameter corresponding to the minimum loss function value to obtain the operating condition.

8. An account credit assessment apparatus, characterized by, Including: An acquisition unit is configured to acquire multi-modal transaction data of a user, wherein the multi-modal transaction data includes transaction flow data and a plurality of related data, the transaction flow data is transaction flow of an account provided by the user in advance, and the related data is other data related to the transaction flow data; An extraction unit is configured to perform feature extraction on at least the multi-modal transaction data to obtain a plurality of target features, and perform feature fusion on the plurality of target features based on an attention mechanism to obtain a fusion feature, wherein the target features are used to represent key flow and operating indicators of the multi-modal transaction data; An evaluation unit is configured to perform credit evaluation on the account according to the fusion feature to obtain a credit evaluation result, wherein the evaluation result includes key flow of the account and operating condition of the account.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium includes a stored program, wherein the program controls the device where the computer readable storage medium is located to execute the method of any one of claims 1 to 7 when the program is running.

10. A processor, comprising: The processor is configured to run a program, wherein the program, when running, performs the method of any one of claims 1 to 7.

Images