Risk assessment method and device for financial service, electronic equipment and storage medium

Abstract

The invention discloses a risk assessment method and device for a financial service, electronic equipment and a storage medium, and relates to the technical field of big data and artificial intelligence or other related fields, and the method comprises the steps: carrying out the information collection of a target service main body from N heterogeneous data sources after receiving a financial service application submitted by the target service main body, n is a positive integer; performing multi-modal feature extraction processing on the business main body information to obtain business main body features; the business main body features are input to a risk assessment model, a risk assessment result is output, the risk assessment model comprises a feature fusion layer and a risk quantification layer, the risk assessment model is used for cooperatively calculating assessment scores of M risk dimensions through the feature fusion layer and the risk quantification layer, and M is a positive integer; and triggering a risk early warning signal based on the risk assessment result. Through the method and the device, the technical problem of low efficiency of risk analysis of business data of different data sources and different structures in related technologies is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of big data and artificial intelligence or other related fields, in particular, to a financial business risk assessment method and device, electronic equipment and storage medium. BACKGROUND

[0002] Financial institutions are facing the urgent need to efficiently extract and analyze business data from diverse data sources for risk assessment, especially in the field of corporate financial services, which requires processing semi-structured and unstructured data from multiple channels such as financial statements, financial service applications, industry analysis reports, etc.

[0003] However, the existing technology has significant deficiencies in processing these heterogeneous data: 1. Low data processing efficiency, traditional manual review method is time-consuming and costly when facing massive data, especially for unstructured data such as PDF reports and Word documents with complex format and irregular information distribution, it is difficult to achieve full coverage; 2. Inaccurate information extraction, when analyzing complex descriptions and potential risk signals in the report, the existing data extraction technology cannot understand the semantic context of the text, resulting in missing or misidentifying key information. 3. Weak cross-data source integration capability, lack of effective correlation analysis between different data sources leads to isolated risk assessment, which cannot evaluate the credit status of the enterprise from a global perspective, affecting the accuracy and comprehensiveness of risk detection.

[0004] In summary, there is a problem of low efficiency in risk analysis of different data sources and different structured business data in the financial industry.

[0005] In view of the above problems, no effective solution has been proposed so far. SUMMARY

[0006] The main purpose of the present application is to provide a financial business risk assessment method and device, electronic equipment and storage medium, to at least solve the technical problem of low efficiency in risk analysis of different data sources and different structured business data in related technologies.

[0007] In order to achieve the above object, according to one aspect of the present application, a risk assessment method for a financial service is provided, which comprises: after receiving a financial service application submitted by a target service subject, collecting information of the target service subject from N heterogeneous data sources to obtain service subject information, wherein N is a positive integer; performing multi-modal feature extraction processing on the service subject information to obtain service subject features, wherein the multi-modal feature extraction includes text feature extraction and structured data feature extraction; inputting the service subject features into a risk assessment model to output a risk assessment result, wherein the risk assessment model comprises a feature fusion layer and a risk quantification layer, and the risk assessment model is configured to cooperatively calculate evaluation scores of M risk dimensions through the feature fusion layer and the risk quantification layer, wherein M is a positive integer; triggering a risk warning signal based on the risk assessment result.

[0008] Further, the step of collecting information of the target service subject from N heterogeneous data sources to obtain service subject information comprises: collecting application materials submitted by the target service subject, wherein the application materials include business description information in text form and financial data in table form; collecting fund flow records of the target service subject from a financial system database, wherein the fund flow records include transaction stream data sorted by time; collecting credit enhancement data of the target service subject from a third-party institution, wherein the credit enhancement data includes asset ownership proof information related to the target service subject; integrating the application materials, the fund flow records and the credit enhancement data according to a preset association rule to obtain structured service subject information.

[0009] Further, the step of performing multi-modal feature extraction processing on the service subject information to obtain service subject features comprises: performing text feature extraction on the business description information and the asset ownership proof information to obtain a text feature set; performing structured data feature extraction on the financial data and the transaction stream data to obtain a structured feature set; and generating the service subject features including the text feature set and the structured feature set.

[0010] Further, the step of performing text feature extraction on the business description information and the asset ownership proof information to obtain a text feature set comprises: performing paragraph-level semantic analysis on the business description information to obtain first-class text features, wherein the first-class text features are used to describe business attributes; performing text analysis on the asset ownership proof information to obtain second-class text features, wherein the second-class text features are used to describe ownership statements; and merging the first-class text features and the second-class text features according to semantic similarity to obtain the text feature set.

[0011] Further, the step of performing structured data feature extraction on the financial data and the transaction log data to obtain a structured feature set comprises: performing table parsing on the financial data to obtain a first type of structured feature, wherein the first type of structured feature is used to represent financial health; performing time series analysis on the transaction log data to obtain a second type of structured feature, wherein the second type of structured feature is used to represent a fund behavior pattern; and combining the first type of structured feature and the second type of structured feature to obtain the structured feature set.

[0012] Further, the step of inputting the business subject feature into a risk assessment model to output a risk assessment result comprises: performing cross-modal correlation analysis on the input text feature vector and the structured feature vector through the feature fusion layer to generate a multi-dimensional feature vector, wherein the cross-modal correlation analysis comprises: dimension alignment and weighted fusion; performing collaborative calculation of M risk dimensions based on the multi-dimensional feature vector through the risk quantification layer to obtain M evaluation scores of the risk dimensions, wherein the collaborative calculation comprises: independent score calculation of each risk dimension and association score weight adjustment between the risk dimensions; and outputting the risk assessment result containing the evaluation scores of the M risk dimensions.

[0013] Further, the step of triggering a risk warning signal based on the risk assessment result comprises: performing risk state classification determination based on the evaluation scores of the M risk dimensions recorded in the risk assessment result to obtain a determination result; generating a risk warning signal based on the risk level in the determination result and generating a risk disposal scheme corresponding to the risk level.

[0014] To achieve the above-mentioned purpose, according to another aspect of the present application, a financial business risk assessment device is also provided, which comprises: an acquisition unit configured to perform information acquisition on a target business subject from N heterogeneous data sources after receiving a financial business application submitted by the target business subject to obtain business subject information, wherein N is a positive integer; an extraction unit configured to perform multi-modal feature extraction processing on the business subject information to obtain business subject features, wherein the multi-modal feature extraction comprises text feature extraction and structured data feature extraction; an input unit configured to input the business subject features into a risk assessment model to output a risk assessment result, wherein the risk assessment model comprises a feature fusion layer and a risk quantification layer, and the risk assessment model is configured to collaboratively calculate evaluation scores of M risk dimensions through the feature fusion layer and the risk quantification layer, wherein M is a positive integer; and a triggering unit configured to trigger a risk warning signal based on the risk assessment result.

[0015] Further, the collection unit comprises: a first collection module, configured to collect application materials submitted by the target business subject, wherein the application materials comprise business description information in text form and financial data in table form; a second collection module, configured to collect fund flow records of the target business subject from a financial system database, wherein the fund flow records comprise transaction log data sorted by time; a third collection module, configured to collect credit enhancement data of the target business subject from a third-party institution, wherein the credit enhancement data comprises asset ownership proof information related to the target business subject; and an integration module, configured to integrate the application materials, the fund flow records and the credit enhancement data according to a preset association rule, to obtain structured business subject information.

[0016] Further, the extraction unit comprises: a first extraction module, configured to perform text feature extraction on the business description information and the asset ownership proof information, to obtain a text feature set; a second extraction module, configured to perform structured data feature extraction on the financial data and the transaction log data, to obtain a structured feature set; and a first generation module, configured to generate the business subject features comprising the text feature set and the structured feature set.

[0017] Further, the first extraction module comprises: a first analysis submodule, configured to perform paragraph-level semantic analysis on the business description information, to obtain first text features, wherein the first text features are used to describe business attributes; a first parsing submodule, configured to perform text parsing on the asset ownership proof information, to obtain second text features, wherein the second text features are used to describe ownership statements; and a merging submodule, configured to merge the first text features and the second text features according to semantic similarity, to obtain the text feature set.

[0018] Further, the second extraction module comprises: a second parsing submodule, configured to perform table parsing on the financial data, to obtain first structured features, wherein the first structured features are used to represent financial health; a second analysis submodule, configured to perform time series analysis on the transaction log data, to obtain second structured features, wherein the second structured features are used to represent fund behavior patterns; and a combination submodule, configured to combine the first structured features and the second structured features according to a standardization rule, to obtain the structured feature set.

[0019] Further, the input unit comprises: an analysis module, configured to perform cross-modal correlation analysis on the input text feature vector and the structured feature vector through the feature fusion layer to generate a multi-dimensional feature vector, wherein the cross-modal correlation analysis comprises dimension alignment and weighted fusion; a calculation module, configured to perform collaborative calculation of the M risk dimensions based on the multi-dimensional feature vector through the risk quantification layer to obtain the M evaluation scores of the risk dimensions, wherein the collaborative calculation comprises independent score calculation of each risk dimension and correlation score weight adjustment between the risk dimensions; and an output module, configured to output the risk assessment result comprising the evaluation scores of the M risk dimensions.

[0020] Further, the trigger unit comprises: a determination module, configured to perform risk state classification determination based on the evaluation scores of the M risk dimensions recorded in the risk assessment result to obtain a determination result; and a second generation module, configured to generate a risk warning signal based on the risk level in the determination result and generate a risk disposal scheme corresponding to the risk level.

[0021] To achieve the above object, according to another aspect of the present application, there is also provided a computer readable storage medium comprising a stored computer program, wherein the computer readable storage medium controls the device where the computer readable storage medium is located to perform the risk assessment method of the financial business according to any one of the above aspects when the computer program is running.

[0022] To achieve the above object, according to another aspect of the present application, there is also provided an electronic device comprising one or more processors and a memory, the memory being configured to store one or more programs, wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the risk assessment method of the financial business according to any one of the above aspects.

[0023] To achieve the above object, according to another aspect of the present application, there is also provided a computer program product comprising computer instructions, wherein the computer instructions, when executed by a processor, implement the steps of the risk assessment method of the financial business according to any one of the above aspects.

[0024] The application provides a risk assessment method for a financial service.

[0025] In the application, the potential risk information of a target service subject is efficiently and accurately mined from multiple heterogeneous data sources by means of automatic and intelligent combination, integrated data acquisition, multi-modal feature extraction, deep feature fusion and risk quantification evaluation, so that the technical effects of rapid response and accurate risk assessment of the financial service application are achieved. Specifically, the application receives a financial service application submitted by a target service subject, and then starts an automatic information acquisition process to collect information from multiple data sources by using OCR, natural language processing and database access technology, so as to ensure the comprehensiveness and real-time performance of the data. Then, multi-modal feature extraction is performed on the service subject information, the semantic information of the text content and the quantitative indicators of the structured data are fused, the feature fusion layer under the deep learning framework is used for deep integration, and the service subject feature representation is generated. Then, the feature representation is input into a risk assessment model, the risk quantification layer built in the model can cooperatively calculate the evaluation scores of multiple risk dimensions, and the multi-dimensional fine risk assessment is realized. Finally, a risk warning signal is triggered based on the risk assessment result, and the technical problem of low efficiency of risk analysis of different data sources and different structured service data in the related art is solved. BRIEF DESCRIPTION OF DRAWINGS

[0026] The accompanying drawings, which form a part of the present application, are intended to provide further understanding of the present application, and the illustrative embodiments of the present application and their description serve the purpose of explaining the present application. In the drawings:

[0027] Figure 1 Fig. 1 shows a hardware structure block diagram of a computer terminal (or a mobile device) for implementing the risk assessment method for the financial service;

[0028] Figure 2 Fig. 2 is a flowchart of an optional risk assessment method for the financial service according to an embodiment of the present application;

[0029] Figure 3is a functional module diagram of an optional information processing and risk assessment system according to an embodiment of the present application;

[0030] Figure 4 is a schematic diagram of an optional financial service risk assessment device according to an embodiment of the present application;

[0031] Figure 5 is a structural block diagram of an electronic device for performing a financial service risk assessment method according to an embodiment of the present application. DETAILED DESCRIPTION

[0032] In order to make the personnel in the art better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should belong to the scope of protection of the present application.

[0033] It should be noted that the terms "first", "second", and the like in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device that includes a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0034] It should be noted that the financial service risk assessment method and device in the present application can be used in the field of big data and artificial intelligence technology for automatic risk assessment of financial service application materials, and can also be used in any field other than the field of big data and artificial intelligence technology for automatic risk assessment of financial service application materials. The application field of the financial service risk assessment method and device in the present application is not limited.

[0035] It should be noted that the related information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, storage, processing, transmission, provision, disclosure, use and processing of related data comply with the laws, regulations and standards of the relevant region, necessary security measures are taken, do not violate public order and good customs, and provide corresponding operation portal for user to choose authorization or refusal. For example, an interface is provided between the system and the related user or institution, and before obtaining the related information, the interface needs to send a request to the aforementioned user or institution, and after receiving the consent information feedback from the aforementioned user or institution, the related information is obtained.

[0036] The information collection (for example, user voice, video, text collection) and analysis operation involved in the present application have provided corresponding operation portal for the user to choose to agree or refuse the automatic decision result when executing; if the user chooses to refuse, the expert decision process is entered.

[0037] The following embodiments of the present application can be applied to various systems / applications / devices that need to perform enterprise financial business risk assessment and multi-source heterogeneous data processing, and can realize a financial business risk intelligent assessment platform based on multi-modal feature fusion and deep learning technology. The present application uses a deep learning model to perform multi-modal feature extraction on business subject information collected from multiple heterogeneous data sources, and then cooperatively calculates the evaluation scores of multiple risk dimensions through a feature fusion layer and a risk quantification layer, which can better understand the complexity and multi-dimensionality of enterprise operating conditions, and accurately identify and quantify potential risks.

[0038] The present application also accurately locates the key information of the business subject through automatic information collection and preprocessing, and performs interrupt insertion type analysis on text and structured data, so that the entire risk assessment process is not only convenient and fast, but also locates risks quickly and accurately, effectively improving the risk management ability of financial institutions in processing a large number of business applications.

[0039] The present application will be described in detail below in conjunction with various embodiments.

[0040] Embodiment one

[0041] According to the embodiments of the present application, an embodiment of a financial business risk assessment method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a group of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that here.

[0042] The risk assessment method for the financial service provided in the embodiment one of the present application can be executed in a mobile terminal, a computer terminal or similar computing device. Figure 1 A hardware structure block diagram of a computer terminal (or mobile device) for implementing the risk assessment method for the financial service is shown. As shown in Figure 1 The computer terminal 10 (or mobile device) can include one or more processors 102 (the processor 102 can include but not limited to a microprocessor MCU or a programmable logic device FPGA and other processing devices), a memory 104 for storing data, and a transmission device 106 for communication function. In addition, it can also include a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which can be included as one of the ports in the BUS bus), a network interface, a power supply and / or a camera. Those skilled in the art can understand that Figure 1 The structure shown is only schematic, which does not limit the structure of the above-mentioned electronic device. For example, the computer terminal 10 can include more or less components than those shown in Figure 1 or have a different configuration than that shown in Figure 1 .

[0043] It should be noted that the one or more processors 102 and / or other data processing circuits described above can be referred to as "data processing circuits" herein. The data processing circuit can be embodied in whole or in part as software, hardware, firmware or any combination thereof. In addition, the data processing circuit can be a single independent processing module, or any one of the other elements combined into the computer terminal 10 (or mobile device) in whole or in part. As referred to in the embodiments of the present application, the data processing circuit is a processor control (for example, the selection of the variable resistance terminal path connected with the interface).

[0044] The memory 104 can be used to store software programs and modules of application software, such as program instructions / data storage devices corresponding to the risk assessment method for the financial service in the embodiments of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, implements the above-mentioned risk assessment method for the financial service. The memory 104 can include a high-speed random access memory, and can also include a non-volatile memory, such as one or more magnetic storage devices, flash memory or other non-volatile solid state memory. In some examples, the memory 104 can further include a memory remotely arranged with respect to the processor 102, which can be connected to the computer terminal 10 through a network. Examples of the above-mentioned network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network and a combination thereof.

[0045] The transmission device 106 is configured to receive or send data via a network. The network can include a wireless network provided by a communication provider of the computer terminal 10. In one example, the transmission device 106 includes a network interface controller (NIC) that can be connected to other network devices through a base station to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module configured to communicate with the Internet wirelessly.

[0046] The display can be a liquid crystal display (LCD) that is touch screen, for example, which can enable a user to interact with a user interface of the computer terminal 10 (or mobile device).

[0047] In the above operating environment, the present application provides a risk assessment method for financial services as shown in Figure 2 The embodiment of the present application is a risk assessment method for financial services, and the implementation subject of the method is a financial service risk intelligent assessment system. The method is used in the automatic risk assessment scene of enterprise financial services, and is particularly used to solve the problem of efficient risk analysis of different data sources and different structured business data. The method includes the following implementation steps: automatic information collection, multi-modal feature deep extraction, feature fusion, quantitative risk assessment, and risk warning, so as to improve the efficiency and accuracy of the risk assessment of financial services.

[0048] The embodiment of the present application will be described in detail in combination with each specific step.

[0049] Figure 2 The embodiment of the present application is a risk assessment method for financial services, and the implementation subject of the method is a financial service risk intelligent assessment system. The method is used in the automatic risk assessment scene of enterprise financial services, and is particularly used to solve the problem of efficient risk analysis of different data sources and different structured business data. The method includes the following implementation steps: automatic information collection, multi-modal feature deep extraction, feature fusion, quantitative risk assessment, and risk warning, so as to improve the efficiency and accuracy of the risk assessment of financial services. Figure 2

[0050] In step S201, after receiving a financial service application submitted by a target service subject, information of the target service subject is collected from N heterogeneous data sources to obtain service subject information, wherein N is a positive integer.

[0051] Specifically, the target service subject refers to a company or an individual who applies for a financial service (such as a loan, financing, a letter of credit, etc.) to a financial institution. In the embodiment of the present application, an enterprise entity is mainly concerned, and the target service subject is the direct object of risk assessment. The application materials provided contain necessary information for risk assessment.

[0052] ​Financial service application refers to the formal application submitted by an enterprise or individual to a financial institution, which includes enterprise profile, financial statements, financial service history, business plan and other materials. In this stage, the financial institution needs to evaluate the credit status, repayment ability and risk level of the applicant based on these materials to decide whether to approve the application and under what conditions to provide financial support.

[0053] Heterogeneous data sources refer to various types and formats of information sources involved in the financial service application process, including but not limited to: financial service application documents submitted by the enterprise (such as PDF, Word format); financial statements of the enterprise (Excel table, PDF, etc.); transaction flow data of the enterprise's financial accounts; external data sources such as public financial data, industry analysis reports, market news, social media information. Each data source provides different types and structures of information, which can include structured data (such as financial data), semi-structured data (such as tables and reports) or unstructured data (such as text and pictures on social media).

[0054] Business subject information refers to the information directly related to the target business subject automatically collected from the above-mentioned heterogeneous data sources, including not only the content of the financial service application materials submitted by the enterprise, but also the supplementary information obtained from external data sources. The collection process of business subject information needs to handle diversified data formats, such as: parsing and extracting key indicators from financial statements; extracting text descriptions and metadata from financial service application documents; identifying fund flow and transaction patterns from transaction flow data; gathering and analyzing auxiliary information from public databases and the Internet, such as industry trends, competitor conditions, market dynamics, etc.

[0055] The embodiment of the present application can efficiently and accurately extract key information related to risk assessment from a large amount of data through automated information collection, significantly improving the ability of financial institutions to handle complex business applications, and helping to quickly identify and assess the financial service risk of the target business subject.

[0056] Optionally, in the financial service risk assessment method provided by the embodiment of the present application, the step of collecting information of the target business subject from N heterogeneous data sources to obtain business subject information includes: collecting the application materials submitted by the target business subject, wherein the application materials contain text form business description information and table form financial data; collecting the fund flow records of the target business subject from the financial system database, wherein the fund flow records contain time-ordered transaction flow data; collecting credit enhancement data of the target business subject from third-party institutions, wherein the credit enhancement data contains asset ownership proof information related to the target business subject; integrating the application materials, fund flow records and credit enhancement data according to the preset association rules to obtain structured business subject information.

[0057] In a specific implementation scenario, collecting the application materials submitted by the target business subject involves obtaining information from financial business application materials directly submitted by the enterprise. The application materials include a series of documents covering the basic information, operating status, financial statements, credit history and other aspects of the enterprise. The business description information mainly exists in the form of text, which elaborates on the development history, market positioning, business model, strategic planning and other content of the enterprise. The financial data is mainly presented in the form of tables, such as balance sheet, profit and loss statement and cash flow statement, which provides direct evidence of the financial health of the enterprise.

[0058] Further, the collection of fund flow records focuses on the transaction flow information between the target business subject and the financial institution, including deposit, withdrawal, transfer, loan and other historical records. These time series data can reflect the trend of enterprise fund flow, the change of account balance and the efficiency of fund use, and are an important basis for assessing the credit and financial stability of the enterprise. By analyzing the transaction flow, abnormal transaction patterns such as frequent large amount of fund in and out and non-normal time period transactions can be identified, thereby warning potential financial risks.

[0059] Obtaining credit enhancement data involves cooperation with external credit evaluation agencies, legal service agencies, asset registration offices and the like, which can supplement the information collected by the financial institution itself. Asset ownership proof information (such as detailed records of real estate, equipment, intellectual property rights, etc.) is used to verify the authenticity of the enterprise's reported assets. The above-mentioned credit enhancement information helps the financial institution to more comprehensively assess the solvency of the business subject and reduce information asymmetry in credit decision-making.

[0060] The collected application materials, fund flow records and credit enhancement data are integrated through pre-set association rules to generate structured business subject information. The association rules can associate information of different sources and formats together based on enterprise ID, account number, timestamp and other key fields to form a unified and structured data set, which not only improves the readability and analyzability of the information, but also facilitates subsequent feature extraction and risk assessment.

[0061] The embodiments of the present application improve the accuracy and comprehensiveness of risk assessment by collecting multi-source data to obtain comprehensive information of the business subject. The heterogeneous data is integrated into structured information, which facilitates subsequent automated analysis and processing, greatly improving the data processing efficiency. Real-time collection and processing of data enable timely risk warning to respond to market and enterprise dynamics, providing immediate risk management basis for financial institutions.

[0062] On the other hand, risk assessment based on comprehensive and structured information can significantly reduce uncertainty in credit decision-making, reduce decision-making risks caused by insufficient information or information asymmetry; automated data collection and information integration simplifies the review process of financial credit business applications, improves the business processing capacity and customer satisfaction of financial institutions; not only solves the technical problems in credit risk assessment, but also brings additional benefits such as improving the risk management efficiency of financial institutions, optimizing the credit process, and enhancing market adaptability.

[0063] In step S202, multi-modal feature extraction processing is performed on the business subject information to obtain business subject features, wherein the multi-modal feature extraction includes text feature extraction and structured data feature extraction.

[0064] Specifically, the multi-modal feature extraction processing refers to automatically identifying and extracting multiple types of features that are helpful for risk assessment from the business subject information. Unlike traditional single data source analysis, multi-modal feature extraction integrates the analysis of multiple types of data such as text information, structured data, and image information, which can more comprehensively and deeply understand the business subject's operating status and potential risks.

[0065] Among them, text feature extraction focuses on unstructured text data such as financial business application documents, market reports, and financial statements. Text feature extraction includes natural language processing, sentiment analysis, keyword extraction, and text classification.

[0066] Specifically, natural language processing refers to using part-of-speech tagging, named entity recognition, and semantic parsing techniques to understand key information in the text, such as business strategies, financial condition descriptions, and industry status. Sentiment analysis refers to identifying potential optimism or pessimism by analyzing the tone of the text when a company describes its financial condition and industry prospects. Keyword extraction and text classification refer to automatically identifying keywords related to risk, such as "debt", "loss", and "legal proceedings", and classifying text content, such as operational risk, financial risk, and legal risk.

[0067] Another need to be explained, structured data feature extraction mainly targets structured or semi-structured data such as enterprise financial statements and transaction records. Structured data feature extraction involves financial indicator extraction, transaction data analysis, and standardization processing.

[0068] Among them, financial indicator extraction includes but is not limited to total assets, total liabilities, operating income, net profit, and cash flow, which can be used to analyze the financial health of the enterprise; transaction data analysis refers to using time series analysis and other statistical methods to identify transaction patterns and abnormal transaction records, and to assess the stability and compliance of the enterprise's cash flow; standardization processing refers to normalizing the extracted financial data to eliminate dimensional effects, facilitating cross-enterprise and cross-time comparative analysis.

[0069] Further, the business subject feature refers to the information set obtained after the multi-modal feature extraction process, which can comprehensively reflect the multi-aspect attributes of the target enterprise, such as financial status, operating risk, market position and compliance, etc., including semantic features, emotional features, financial indicators, transaction behavior features and compliance features.

[0070] Specifically, the semantic feature refers to the semantic representation extracted from the text by natural language processing technology, such as the semantic vector describing the financial status of the enterprise; the emotional feature refers to the emotional score of the enterprise obtained by sentiment analysis, which is used to indicate the subjective cognition of the enterprise on its own status; the financial indicator refers to the financial data vector after standardization processing, which is used to quantify the financial status of the enterprise; the transaction behavior feature refers to the vector converted from the transaction flow data, which is used to reflect the fund flow situation and transaction mode of the enterprise; the compliance feature refers to the compliance information extracted from the financial business application materials, which is used to evaluate the compliance of the enterprise.

[0071] Through deep learning and analysis of the business subject feature, the embodiment of the present application combines text understanding and financial data analysis to depict a multi-dimensional enterprise risk portrait, and takes the business subject feature as the input of the subsequent risk quantification and early warning mechanism, which can realize accurate identification and quantification of potential risks, provide scientific financial business decision basis for financial institutions, and help financial institutions to comprehensively evaluate the credit status of enterprises and reduce the uncertainty in financial business.

[0072] Optionally, in the financial business risk assessment method provided by the embodiment of the present application, the step of performing multi-modal feature extraction processing on the business subject information to obtain the business subject feature includes: performing text feature extraction on the business description information and the asset ownership proof information to obtain a text feature set; performing structured data feature extraction on the financial data and the transaction flow data to obtain a structured feature set; and generating a business subject feature containing the text feature set and the structured feature set.

[0073] It should be noted that the text feature extraction is a process of deep analysis on the two types of unstructured text data, i.e., the business description information and the asset ownership proof information. The specific implementation steps include natural language processing, keyword and entity recognition, and sentiment analysis.

[0074] Among them, natural language processing (NLP) is to convert text into a computer processable form by applying techniques such as stem extraction, word segmentation, and syntax analysis, and key word and entity recognition is to filter out words and entities closely related to risk assessment through key word matching and entity recognition algorithms, such as "liabilities", "litigation", "property rights", etc., to build a risk-related vocabulary library; sentiment analysis is to analyze the tone and emotion in the text with the help of deep learning models to identify the positive or negative tendency when the enterprise describes its financial and asset conditions, providing emotional insights into potential risks.

[0075] The text feature set is a set of all text features obtained after extraction, including but not limited to key word frequency, sentiment polarity value, entity relationship graph, etc., used to reflect the authenticity and stability of the business subject's operating status and asset ownership.

[0076] Another point to note is that structured data feature extraction is aimed at structured information such as financial data and transaction stream data, focusing on the following aspects: time series analysis, financial indicator calculation, and compliance check.

[0077] Among them, time series analysis refers to the use of statistical and machine learning models to analyze transaction stream data to predict and identify abnormal changes in cash flow, such as sudden large expenditures, unusual sources of income, etc.; financial indicator calculation refers to calculating key financial ratios and indicators based on balance sheet, income statement, etc., such as liquidity ratio, debt equity ratio, operating profit ratio, etc., to assess the solvency and profitability of the enterprise; compliance check refers to verifying the compliance of financial data through database queries and rule matching to check whether there are violations of accounting standards or laws and regulations.

[0078] Correspondingly, the structured feature set obtained after extraction contains all the features extracted from financial and transaction data, such as financial indicator values, time series anomaly scores, and compliance scores.

[0079] Further, the last step is to fuse the text feature set and the structured feature set to generate a business subject feature containing multi-modal information. This fusion step can be achieved through deep learning models (such as multi-modal Transformer) or feature engineering methods (such as weight-based feature combination), which integrate text semantics, emotional tendencies, financial ratios, and transaction patterns into a feature vector, i.e. the business subject feature, as the data input for the risk assessment model.

[0080] In the embodiments of the present application, the multi-modal feature extraction not only considers the hard indicators of financial data, but also combines the sentiment and intention of the text description, realizing the deep understanding of the risk status of the business subject; the business subject features of the fusion of text and structured data enable the risk assessment model to make predictions based on more comprehensive and diverse information, significantly improving the accuracy and reliability of the predictions; automated feature extraction reduces the need for human intervention, speeds up information processing, shortens the time for risk assessment and credit decision-making, and improves the operational efficiency of financial institutions.

[0081] Optionally, in the financial business risk assessment method provided by the embodiments of the present application, the step of performing text feature extraction on the business description information and the asset ownership proof information to obtain a text feature set comprises: performing paragraph-level semantic analysis on the business description information to obtain first text features, wherein the first text features are used to describe business attributes; performing text analysis on the asset ownership proof information to obtain second text features, wherein the second text features are used to describe ownership statements; and merging the first text features and the second text features according to semantic similarity to obtain the text feature set.

[0082] It should be noted that the paragraph-level semantic analysis is a deep text processing step performed on the business description information. This step uses natural language processing techniques such as word embedding, syntax analysis, and semantic role labeling to understand and analyze the text content in detail. Specifically, the text information in the credit application materials is divided by paragraphs, and then each paragraph is semantically analyzed to identify key entities, events, and semantic roles, thereby extracting the first text features describing business attributes. For example, events such as "income growth" and "new market expansion" are identified, as well as the relationships between these events and entities such as "company" and "market".

[0083] Further, the text analysis of the asset ownership proof information is used to extract clear and accurate asset details from unstructured text. Specifically, by using keyword matching and entity linking techniques, the asset type (such as real estate, vehicles, intellectual property), ownership status (ownership, use rights, lease), and value assessment information in the text are identified to construct the second text features, which are specifically used to describe ownership statements, helping financial institutions verify the asset situation of enterprises and assess the authenticity and value of their collateral.

[0084] After obtaining the first and second text features, the features related to the same or similar business attributes are merged by calculating semantic similarity to construct a comprehensive text feature set. This merging process uses a word vector model to calculate the similarity score between features, and features with a score higher than a preset threshold are considered to be the same topic or domain and are merged. For example, features describing "income growth" and "income stability" can be merged into a comprehensive feature related to "income status" if the similarity score is high.

[0085] The embodiment of the present application can accurately locate risk points such as potential operating challenges of enterprises and ownership disputes of assets through paragraph-level semantic analysis and ownership declaration recognition, thereby improving the accuracy of risk identification; the use of a deep learning model to analyze the text enhances the understanding of the semantic of the text and enables the identification of potential risk signals in complex contexts; feature merging based on semantic similarity reduces feature redundancy and optimizes the feature set, thereby making the subsequent risk assessment model training more efficient.

[0086] Through the above steps, the embodiment of the present application significantly improves the risk identification and assessment ability of financial institutions in credit business, not only solves the problems of information processing efficiency and accuracy in traditional risk assessment, but also realizes the deep understanding of information, optimization of features and comprehensiveness of decision information, thereby providing financial institutions with a more intelligent and forward-looking risk management tool.

[0087] Optionally, in the financial business risk assessment method provided by the embodiment of the present application, the step of extracting structured data features from the financial data and the transaction flow data to obtain a structured feature set includes: performing table parsing on the financial data to obtain first structured features, wherein the first structured features are used to represent the financial health; performing time series analysis on the transaction flow data to obtain second structured features, wherein the second structured features are used to represent the fund behavior mode; and combining the first structured features and the second structured features in a standardized manner to obtain the structured feature set.

[0088] It should be noted that the table parsing is an automatic processing process for the financial data, which is used to extract key financial indicators from complex table structures to form the first structured features. The specific steps include: cell recognition and content extraction, financial indicator standardization, and financial ratio calculation.

[0089] Among them, cell recognition and content extraction refers to using OCR (Optical Character Recognition) technology to identify the cell boundaries in the financial statements, and then extracting the numerical and textual information in the cells; financial indicator standardization refers to standardizing the extracted financial information, such as unifying different currency units and removing non-numeric characters in the text, to ensure the comparability between indicators; financial ratio calculation refers to calculating various financial ratios based on the standardized financial data, such as liquidity ratio, asset-liability ratio, net profit ratio, etc., which can directly reflect the financial health of the enterprise.

[0090] The first type of structured features is obtained by the above steps, mainly including standardized financial data and calculated financial ratios, which are important basis for evaluating the solvency and profitability of enterprises.

[0091] Another point to note is that time series analysis and fund behavior pattern feature extraction refer to the time series analysis of transaction flow data, aiming to identify and understand the rules and patterns of enterprise fund inflow and outflow, and then obtain the second type of structured features, including the following steps: transaction frequency analysis, amount distribution statistics, time series modeling.

[0092] Among them, transaction frequency analysis refers to the statistics of the number of transactions within a certain period of time, which is used to analyze the activity of transaction behavior; amount distribution statistics refers to analyzing the distribution of transaction amounts to identify large transactions, abnormal transactions, etc., and evaluating the stability of fund flow; time series modeling refers to using time series analysis techniques to predict future fund flow trends to identify potential risk signals.

[0093] The second type of structured features embodies the characteristics of enterprise fund behavior patterns, such as transaction frequency, amount distribution, and predicted flow trends, which are crucial for identifying the health of enterprise fund flow and potential risks.

[0094] After obtaining the first and second types of structured features, the features need to be standardized and combined to build a comprehensive structured feature set. Standardization includes but is not limited to: data cleaning, i.e., removing outliers and incorrect data; feature scaling, i.e., scaling all features to the same order of magnitude range to avoid a feature dominating the risk assessment results due to a large order of magnitude difference; feature selection, i.e., based on the importance analysis of features, selecting the most representative and significantly contributing features to risk assessment to build a structured feature set. It should be noted that the structured feature set is a collection of standardized features such as financial ratios and fund behavior patterns.

[0095] The embodiment of the present application realizes the automatic and efficient processing of financial data and transaction flow data through table analysis and time series analysis, significantly improves the information processing speed and accuracy; the construction of the structured feature set enables the risk assessment process to be based on quantified data, enhancing the objectivity and scientificity of risk identification and assessment; through time series analysis technology, the regularity and pattern of fund flow can be identified, future possible fund risks can be predicted, and the forward-looking and accuracy of risk early warning are improved.

[0096] On the other hand, the feature extraction of financial health degree and fund behavior pattern enables the risk assessment to cover the financial status and operation risk of the enterprise more finely, providing a multi-dimensional evaluation perspective; the structured feature set provides clear and quantified data for the financial institutions, which helps decision makers quickly understand the risk points and provides strong assistance for the credit approval process; the time series analysis based on transaction flow can realize dynamic monitoring of risks, timely discovery of abnormalities in fund flow, and support for real-time or regular risk assessment update.

[0097] Through the above steps, the embodiment of the present application provides a structured data feature extraction method, which not only solves the technical problems of processing financial and transaction data in traditional risk assessment, but also realizes the efficiency of data processing, the refinement of risk assessment and the dynamic monitoring capability.

[0098] Step S203, inputting the business subject feature into the risk assessment model to output a risk assessment result, wherein the risk assessment model comprises a feature fusion layer and a risk quantification layer, and the risk assessment model is used to calculate the evaluation scores of M risk dimensions through the cooperation of the feature fusion layer and the risk quantification layer, and M is a positive integer.

[0099] Specifically, the risk assessment model combines the feature fusion capability of deep learning and the risk quantification capability of statistical learning, and is used to assess the financial business risk of the target business subject. The model structure mainly includes a feature fusion layer and a risk quantification layer, which work together to calculate a comprehensive risk assessment result.

[0100] The feature fusion layer is used to integrate the multi-modal information in the business subject feature into a unified representation form. Deep learning technology (such as the combination of bidirectional long short-term memory network and pre-trained language model) is used to fuse the features extracted from different sources such as text, structured data and images, to generate a comprehensive feature vector. The feature fusion layer can capture the correlation between different modal information, providing more comprehensive input for subsequent risk quantification.

[0101] The risk quantification layer uses machine learning algorithms to calculate evaluation scores for multiple risk dimensions based on the fused business subject features. The output of this layer is a quantitative risk score, each score corresponding to a specific risk dimension such as operational stability, financial health, market competitiveness, compliance, etc. The risk quantification layer can use models such as support vector machines, random forests, or neural networks to establish risk assessment models through training data sets to identify and quantify different risk factors.

[0102] The risk assessment result is the information reflecting the degree of financial business risk of the target business subject output by the model, including evaluation scores for multiple risk dimensions and an overall risk score. In one specific embodiment, each score ranges from 0 to 1, with higher values indicating greater risk in the corresponding dimension. The overall risk score is a weighted average based on all dimensions, reflecting the overall risk level of the target business subject. The assessment result can also include an explanation of each risk dimension to help financial institutions understand the source and nature of the risk.

[0103] Another point to note is that risk dimensions refer to the classification of various risks considered in financial business risk assessment, which can include operational risk, financial risk, compliance risk, market risk, credit risk, and legal risk. Operational risk is related to enterprise operational stability and market competitiveness; financial risk focuses on enterprise assets, liabilities, cash flow, and profitability; compliance risk assesses the extent to which the enterprise complies with regulations and industry standards; market risk considers the potential impact of industry fluctuations and economic cycles on the enterprise; credit risk is based on historical financial business records and analysis of default probability; legal risk involves potential legal lawsuits and disputes faced by the enterprise. Each risk dimension corresponds to a specific set of features and evaluation indicators, which are key components of the risk assessment model.

[0104] The evaluation score is a numerical representation obtained by the risk assessment model after quantitative analysis of each risk dimension, reflecting the importance and current status of individual risk dimensions and facilitating subsequent comprehensive analysis and comparison. The calculation of the evaluation score is based on the weight parameters and feature values defined during model training, which can be dynamically adjusted to adapt to changes in the financial market and new types of risks.

[0105] The embodiments of the present application can generate comprehensive and quantitative risk assessment results, providing financial institutions with accurate financial business risk early warning and decision support, thereby effectively reducing uncertainty and potential losses in financial business. Not only does it enhance the understanding of complex risk factors, but also improves the accuracy and efficiency of risk assessment through machine learning technology.

[0106] Optionally, in the financial service risk assessment method provided by the embodiment of the application, the step of inputting the service subject characteristics into the risk assessment model and outputting the risk assessment result comprises: performing cross-modal correlation analysis on the input text feature vector and the structured feature vector through a feature fusion layer to generate a multi-dimensional feature vector, wherein the cross-modal correlation analysis comprises: dimension alignment and weighted fusion; performing collaborative calculation of M risk dimensions based on the multi-dimensional feature vector through a risk quantification layer to obtain M evaluation scores, wherein the collaborative calculation comprises: independent score calculation of each risk dimension and associated score weight adjustment between risk dimensions; and outputting a risk assessment result containing the evaluation scores of the M risk dimensions.

[0107] Need to be specified is that the main task of the feature fusion layer is to perform cross-modal correlation analysis on the obtained text feature vector and the structured feature vector to generate a multi-dimensional feature vector, ensuring the comprehensiveness and depth of information. The cross-modal correlation analysis specifically comprises: dimension alignment and weighted fusion.

[0108] Specifically, since the original dimensions of the text features and the structured features may be different, the purpose of the dimension alignment is to adjust the feature dimensions of the two to the same or compatible degree through vector mapping or feature expansion, facilitating subsequent fusion calculation. For example, a pre-trained multi-modal model (such as a multi-modal Transformer) is used to re-encode the text feature vector to match the dimension of the structured feature vector.

[0109] On the basis of the dimension alignment, the feature fusion layer combines the text feature vector and the structured feature vector into a multi-dimensional feature vector through weighted fusion. The setting of the weight is usually based on the importance of the features and can be adjusted through machine learning techniques (such as feature importance ranking) or expert experience. The fused multi-dimensional feature vector not only contains the original text and structured information, but also embodies the interaction between different modal features, providing a more comprehensive perspective for risk assessment.

[0110] Further, the risk quantification layer performs collaborative calculation of multiple risk dimensions based on the generated multi-dimensional feature vector to obtain the evaluation scores of each dimension, and this process includes independent score calculation, associated score weight adjustment, and output of a risk assessment result.

[0111] Specifically, for each risk dimension (such as financial risk, market risk, legal risk, etc.), the risk quantification layer uses a specially designed scoring model (such as a deep learning-based classifier) to perform independent score calculation, and each model is trained for the risk features of a specific dimension, which can accurately assess the risk status under that dimension.

[0112] On the basis of independent scoring, the risk quantification layer also considers the mutual influence between different risk dimensions, realizes the collaborative calculation of multiple dimensions by adjusting the correlation score weight between risk dimensions. The weight adjustment can be realized in the following ways: expert system, that is, manually setting the weight relationship between different dimensions according to the experience of field experts; machine learning, that is, using historical data to train the model to automatically learn the correlation and influence degree between different risk dimensions and dynamically adjust the weight; network analysis, that is, constructing a risk factor network diagram, and adjusting the weight by quantifying the importance of each dimension risk in the network through graph algorithm.

[0113] Finally, the risk quantification layer outputs the risk assessment result containing the evaluation scores of multiple risk dimensions, and the evaluation score of each dimension reflects the quantification degree of the risk of the dimension, providing a quantitative basis for credit decision-making, making the risk assessment more objective and accurate.

[0114] In the embodiment of the application, the text features and structured features are deeply integrated through cross-modal correlation analysis, the input information quality of the risk assessment model is improved, and the prediction ability of the model is enhanced; the collaborative calculation of multiple risk dimensions overcomes the limitations of traditional single dimension evaluation, provides a comprehensive risk assessment perspective, and helps to identify and manage various types of risks early; the correlation score weight adjustment mechanism between risk dimensions makes the risk assessment more adaptable to changing market environment and enterprise conditions, improves the flexibility and accuracy of the evaluation.

[0115] Through the above steps, the risk assessment method in the embodiment of the application not only solves the problems of incomplete information processing and single dimension risk identification in traditional credit risk assessment, but also realizes the depth of information fusion, multi-dimensional risk quantification and enhancement of market adaptability and other beneficial effects.

[0116] Step S204, triggering a risk warning signal based on the risk assessment result.

[0117] Specifically, the risk warning signal is a warning mechanism triggered based on the risk assessment result, which is used to remind the financial institution to pay attention to the potential risks of the target business subject in time, so as to ensure that the risks can be identified and handled early to prevent or reduce possible economic losses.

[0118] It should be noted that the triggering of the risk warning signal is based on the evaluation scores of multiple risk dimensions output by the risk assessment model, and the warning signal is automatically generated when the evaluation score of one or more dimensions exceeds the preset threshold, which can be static or dynamically adjusted according to market environment, industry characteristics and historical data. In addition, the warning signal can also be assigned different priorities according to the high and low of the comprehensive risk score.

[0119] In a specific embodiment, the content of the risk warning signal can include: risk dimension and score, risk factor, recommended measures, etc. Among them, the risk dimension and score clearly indicate the risk dimension triggering the warning and the corresponding evaluation score, facilitating understanding of the specific type and severity of the risk; the risk factor lists the specific factors leading to the increase of the risk score in detail, such as abnormal financial indicators, improper business strategy, increase of legal litigation, etc.; the recommended measures are based on the risk type, and the system can provide preliminary coping strategies or suggestions, such as requiring additional collateral, adjusting financial business conditions, strengthening post-loan monitoring, etc.

[0120] After receiving the risk warning signal, the financial institution can adopt the following processing methods according to the severity and urgency of the warning: 1. Further review, i.e. conducting more in-depth investigation and analysis on the business subject involved in the warning, including but not limited to on-site audit, third-party assessment, communication with enterprise executives, etc.; 2. Adjusting financial business conditions, i.e. adjusting the loan interest rate, term, limit or requiring additional guarantee measures according to the risk assessment results to reduce the risk of financial business; 3. Risk monitoring, i.e. strengthening the risk monitoring of the target business subject, re-evaluating the risk regularly or irregularly to ensure the safety of financial business assets.

[0121] The embodiment of the present application can realize real-time monitoring and warning of financial business risks, provide decision support for financial institutions, help financial institutions find a balance point between risk control and business development, reduce the subjectivity and lag of manual review, and improve the sensitivity and response speed of financial institutions to market changes.

[0122] Optionally, in the financial business risk assessment method provided by the embodiment of the present application, the step of triggering a risk warning signal based on the risk assessment result comprises: performing risk state classification determination based on the evaluation scores of the M risk dimensions recorded in the risk assessment result to obtain a determination result; generating a risk warning signal based on the risk level in the determination result, and generating a risk disposal scheme corresponding to the risk level.

[0123] It should be noted that the risk state classification determination is based on the evaluation scores of multiple risk dimensions in the risk assessment result, aiming to divide the risk state of the business subject into different levels through quantitative indicators. This determination process usually follows a pre-set risk scoring standard, for example, a specific implementation scenario includes:

[0124] Setting risk level thresholds: defining the scoring thresholds for low risk, medium risk and high risk, such as low risk: <30 points; medium risk: 30-60 points; high risk: >60 points;

[0125] Risk dimension comprehensive score: combining the evaluation scores of multiple dimensions, the weighted average or accumulation of the calculation results of the risk quantification layer is performed to obtain the comprehensive risk score of the business subject;

[0126] Risk level determination: determine the risk level of the business subject according to the comparison of the comprehensive risk score with the preset threshold, such as the score is higher than the high risk threshold, it is determined as high risk state.

[0127] Further, generate a corresponding risk warning signal based on the risk level in the determination result, and formulate a risk disposal scheme for different risk levels, specifically including: risk warning signal generation and risk disposal scheme generation.

[0128] Specifically, for the business subject of medium and high risk level, automatically trigger the risk warning signal, including but not limited to: sending an email, a short message or generating a warning notification in the system, reminding the relevant departments or personnel to pay attention and take corresponding measures; and automatically generate disposal scheme suggestions according to the risk level and risk type (such as financial risk, credit risk, market risk, etc.), for example, for enterprises with high financial risk, require additional financial reports, increase collateral, adjust loan conditions or limit loan amount, etc.

[0129] In the embodiment of the present application, the risk quantification and grading are realized through risk scoring and level determination, which provides clear and quantitative risk management basis for financial institutions; the automatic triggering of the warning signal based on the risk level ensures the timeliness and systematicness of the risk warning, and reduces human negligence or delay; the automatic generation of risk disposal scheme combining risk level and type provides intelligent suggestions for credit approval and risk management, and enhances the pertinence and effectiveness of the disposal scheme.

[0130] Through the above steps, the risk warning mechanism in the embodiment of the present application not only solves the timeliness and accuracy of the warning signal generation in credit risk assessment, but also realizes the risk quantification, intelligent disposal scheme generation and optimization of credit decision-making process, etc., so that the risk assessment not only stays at the theoretical analysis level, but also can be transformed into actual warning and management action, which significantly enhances the risk response ability of financial institutions.

[0131] Through the above steps S201 to S204, the target business subject can first receive the financial business application submitted by the target business subject, then collect information of the target business subject from N heterogeneous data sources to obtain business subject information, wherein N is a positive integer, then perform multi-modal feature extraction processing on the business subject information to obtain business subject features, wherein the multi-modal feature extraction includes text feature extraction and structured data feature extraction, then input the business subject features into the risk assessment model to output the risk assessment result, wherein the risk assessment model includes a feature fusion layer and a risk quantification layer, the risk assessment model is used to calculate the evaluation scores of M risk dimensions through the feature fusion layer and the risk quantification layer, M is a positive integer, and finally trigger the risk warning signal based on the risk assessment result.

[0132] In the embodiment of the present application, the combination of automation and intelligence is adopted, and through the means of integrated data acquisition, multi-modal feature extraction, deep feature fusion and risk quantification evaluation, the purpose of efficiently and accurately mining the potential risk information of the target business subject from multiple heterogeneous data sources is achieved, thereby realizing the technical effects of rapid response and accurate risk assessment of the financial service application. Specifically, the embodiment of the present application receives the financial service application submitted by the target business subject, and then starts the automatic information acquisition process, and uses OCR, natural language processing and database access technology to perform all-round information acquisition from multiple data sources, ensuring the data comprehensiveness and real-time performance; then the multi-modal feature extraction processing is performed on the business subject information, the semantic information of the text content and the quantitative indicators of the structured data are fused, the feature fusion layer under the deep learning framework is used for deep integration, and the business subject feature representation is generated; then the feature representation is input into the risk assessment model, and the risk quantification layer built in the model can cooperatively calculate the evaluation scores of multiple risk dimensions, realizing the multi-dimensional fine-grained evaluation of the risk; finally, the risk warning signal is rapidly triggered based on the risk assessment result, thereby solving the technical problem of low efficiency of risk analysis on different data sources and different structured business data in the related art.

[0133] The present application will be described in detail below in conjunction with another specific embodiment.

[0134] In a specific implementation scenario, based on the problems of low efficiency and insufficient risk identification capability in the semi-structured information retrieval and risk mining process of enterprise credit risk audit materials in the prior art, an automatic information processing scheme is proposed, which automatically extracts and processes the key information in the massive reports through data acquisition, natural language processing and data structuring technology, uses deep learning and knowledge graph technology to perform deep analysis and classification on the risk factors, and automatically identifies the potential risks.

[0135] Figure 3 is a functional module diagram of an optional information processing and risk assessment system according to the embodiment of the present application, as shown in Figure 3 The system adopts a hierarchical architecture design and includes the following core functional modules: a data acquisition layer, a data preprocessing layer, an information extraction layer and a risk analysis layer. The specific structure of each core functional module and the data flow relationship will be described below.

[0136] The data acquisition layer includes a structured data acquisition module and a semi-structured information acquisition module, wherein the structured data acquisition module is responsible for acquiring standardized data with clear field definition; the semi-structured information acquisition module is used to process document data with partial structural characteristics.

[0137] In the data preprocessing layer, a text processing unit, a table parsing unit and a metadata processing unit are included. The text processing unit is configured to perform title-level segmentation, paragraph-level segmentation and text cleaning. The table parsing unit is configured to extract table data and establish a field mapping relationship. The metadata processing unit is configured to generate data traceability identification and version control information.

[0138] In the information extraction layer, a text feature extraction engine, a numerical feature extraction engine, a time series feature extraction engine and a multi-modal fusion module are included. The text feature extraction engine is configured to extract semantic features based on a deep learning model. The numerical feature extraction engine is configured to generate an index vector through quantitative analysis. The time series feature extraction engine is configured to capture data dynamic change rules. The multi-modal fusion module is configured to perform cross-modal feature correlation and dimension reduction.

[0139] In the risk analysis layer, a risk factor extraction module, an anomaly detection module, a sentiment analysis module and a comprehensive evaluation engine are included. The risk factor extraction module is configured to identify key risk indicator factors. The anomaly detection module is configured to discover data deviations based on a statistical model. The sentiment analysis module is configured to evaluate risk tendencies in text expressions. The comprehensive evaluation engine is configured to output risk evaluation conclusions with confidence.

[0140] The application will be described in detail below in combination with another alternative embodiment.

[0141] Embodiment two

[0142] The application also provides a financial service risk assessment device. The financial service risk assessment device includes a plurality of implementation units and can be used to execute the financial service risk assessment method provided in Embodiment One. Each implementation unit corresponds to each implementation step in Embodiment One.

[0143] Figure 4 is a schematic diagram of an alternative financial service risk assessment device according to an embodiment of the application, as shown in the figure, the device can include: a collection unit 41, an extraction unit 42, an input unit 43, and a triggering unit 44. Figure 4

[0144] The collection unit 41 is configured to receive a financial service application submitted by a target service subject and collect information of the target service subject from N heterogeneous data sources to obtain service subject information, where N is a positive integer.

[0145] The extraction unit 42 is configured to perform multi-modal feature extraction processing on the service subject information to obtain service subject features, where the multi-modal feature extraction includes text feature extraction and structured data feature extraction.

[0146] ​The input unit 43 is configured to input the service subject feature into the risk assessment model, and output a risk assessment result, wherein the risk assessment model comprises a feature fusion layer and a risk quantification layer, and the risk assessment model is configured to cooperatively calculate evaluation scores of M risk dimensions through the feature fusion layer and the risk quantification layer, wherein M is a positive integer.

[0147] The triggering unit 44 is configured to trigger a risk warning signal based on the risk assessment result.

[0148] The risk assessment device for the financial service can first receive a financial service application submitted by a target service subject through the collecting unit 41, collect information of the target service subject from N heterogeneous data sources, and obtain service subject information, wherein N is a positive integer. Then, the service subject information is subjected to multi-modal feature extraction processing through the extracting unit 42, and service subject features are obtained, wherein the multi-modal feature extraction includes text feature extraction and structured data feature extraction. Then, the service subject features are input into the risk assessment model through the input unit 43, and a risk assessment result is output, wherein the risk assessment model comprises a feature fusion layer and a risk quantification layer, and the risk assessment model is configured to cooperatively calculate evaluation scores of M risk dimensions through the feature fusion layer and the risk quantification layer, wherein M is a positive integer. Finally, the triggering unit 44 triggers a risk warning signal based on the risk assessment result.

[0149] In the embodiment of the present application, an automatic and intelligent combination is adopted, and data collection, multi-modal feature extraction, deep feature fusion and risk quantification evaluation are integrated, so that the potential risk information of the target service subject can be efficiently and accurately mined from multiple heterogeneous data sources, thereby achieving the technical effects of rapid response to the financial service application and precise risk assessment. Specifically, the financial service application submitted by the target service subject is received, and then an automatic information collection process is started. Information is collected from multiple data sources by using OCR, natural language processing and database access technology, so as to ensure the comprehensiveness and real-time performance of the data. Then, multi-modal feature extraction processing is performed on the service subject information, the semantic information of the text content and the quantitative indicators of the structured data are fused, the feature fusion layer under the deep learning framework is used for deep integration, and the service subject feature representation is generated. Then, the feature representation is input into the risk assessment model, the risk quantification layer built in the model can cooperatively calculate the evaluation scores of multiple risk dimensions, and the multi-dimensional fine risk assessment is realized. Finally, a risk warning signal is triggered based on the risk assessment result, thereby solving the technical problem of low efficiency of risk analysis of different data sources and different structured service data in the related art.

[0150] Further, the collecting unit comprises: a first collecting module configured to collect application materials submitted by the target business subject, wherein the application materials comprise business description information in text form and financial data in table form; a second collecting module configured to collect fund flow records of the target business subject from a financial system database, wherein the fund flow records comprise transaction flow data sorted by time; a third collecting module configured to collect credit enhancement data of the target business subject from a third-party institution, wherein the credit enhancement data comprise asset ownership proof information related to the target business subject; and an integration module configured to integrate the application materials, the fund flow records and the credit enhancement data according to a preset association rule to obtain structured business subject information.

[0151] Further, the extracting unit comprises: a first extracting module configured to perform text feature extraction on the business description information and the asset ownership proof information to obtain a text feature set; a second extracting module configured to perform structured data feature extraction on the financial data and the transaction flow data to obtain a structured feature set; and a first generating module configured to generate business subject features comprising the text feature set and the structured feature set.

[0152] Further, the first extracting module comprises: a first analyzing submodule configured to perform paragraph-level semantic analysis on the business description information to obtain first text features, wherein the first text features are used to describe business attributes; a first parsing submodule configured to perform text parsing on the asset ownership proof information to obtain second text features, wherein the second text features are used to describe ownership statements; and a merging submodule configured to merge the first text features and the second text features according to semantic similarity to obtain the text feature set.

[0153] Further, the second extracting module comprises: a second parsing submodule configured to perform table parsing on the financial data to obtain first structured features, wherein the first structured features are used to represent financial health; a second analyzing submodule configured to perform time series analysis on the transaction flow data to obtain second structured features, wherein the second structured features are used to represent fund behavior patterns; and a combination submodule configured to combine the first structured features and the second structured features according to a standardization rule to obtain the structured feature set.

[0154] Furthermore, the input unit includes: an analysis module, which is used to perform cross-modal correlation analysis on the input text feature vector and structured feature vector through a feature fusion layer to generate a multidimensional feature vector, wherein the cross-modal correlation analysis includes: dimension alignment and weighted fusion; a calculation module, which is used to perform collaborative calculation of M risk dimensions based on the multidimensional feature vector through a risk quantification layer to obtain M evaluation scores, wherein the collaborative calculation includes: independent score calculation of each risk dimension and adjustment of the correlation score weights between risk dimensions; an output module, which is used to output a risk assessment result containing the evaluation scores of M risk dimensions.

[0155] Furthermore, the trigger unit includes: a judgment module, which is used to perform risk status classification judgment based on the assessment scores of M risk dimensions recorded in the risk assessment results to obtain a judgment result; a second generation module, which is used to generate a risk warning signal based on the risk level in the judgment result, and generate a risk disposal plan corresponding to the risk level.

[0156] It should be noted that the above-mentioned acquisition unit 41, extraction unit 42, input unit 43, and trigger unit 44 correspond to steps S201 to S204 in the first embodiment. The examples and application scenarios implemented by the above-mentioned units and corresponding steps are the same, but are not limited to the contents disclosed in the above-mentioned first embodiment. It should be noted that the above-mentioned modules or units can be hardware components or software components stored in a memory (e.g., memory 104) and processed by one or more processors (e.g., processors 102a, 102b, ..., 102n). The above-mentioned modules or units can also be part of the device and can be run in the computer terminal 10 provided in the first embodiment.

[0157] The present invention is described below in conjunction with another optional embodiment.

[0158] Example 3

[0159] An embodiment of the present invention may further provide an electronic device, Figure 5 FIG. 1 is a structural block diagram of an electronic device for executing a risk assessment method for financial services according to an embodiment of the present invention. Figure 5 As shown, the electronic device may include: one or more ( Figure 5 Only one is shown) processor 502, memory 504, storage controller, and peripheral interface, wherein the peripheral interface is connected to the radio frequency module, audio module and display.

[0160] Among them, the memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the risk assessment method and device for financial services in the embodiments of the present application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, that is, implementing the risk assessment method for financial services described above. The memory may include a high-speed random access memory and may also include a 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 may further include a memory remotely located relative to the processor, and these remote memories may be connected to the terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0161] The processor can call the information and application programs stored in the memory through the transmission device to perform the following steps: after receiving the financial business application submitted by the target business entity, collect information about the target business entity from N heterogeneous data sources to obtain business entity information, where N is a positive integer; perform multimodal feature extraction on the business entity information to obtain business entity features, where multimodal feature extraction includes text feature extraction and structured data feature extraction; input the business entity features into the risk assessment model and output the risk assessment results, where the risk assessment model includes a feature fusion layer and a risk quantification layer, and the risk assessment model is used to collaboratively calculate the assessment scores of M risk dimensions through the feature fusion layer and the risk quantification layer, where M is a positive integer; trigger a risk warning signal based on the risk assessment results.

[0162] The processor can also call the information and application programs stored in the memory through the transmission device to perform the following steps: collecting application materials submitted by the target business entity, wherein the application materials include business description information in text form and financial data in tabular form; collecting the target business entity's fund transaction records from the financial system database, wherein the fund transaction records include transaction flow data sorted by time; collecting the target business entity's credit enhancement data from a third-party institution, wherein the credit enhancement data includes asset ownership certificate information related to the target business entity; integrating the application materials, fund transaction records and credit enhancement data according to preset association rules to obtain structured business entity information.

[0163] The processor can also call the information and applications stored in the memory through the transmission device to perform the following steps: extract text features from business description information and asset ownership proof information to obtain a text feature set; extract structured data features from financial data and transaction flow data to obtain a structured feature set; generate business subject features that include a text feature set and a structured feature set.

[0164] The processor can further call information and application programs stored in the memory through the transmission device to perform the following steps: performing paragraph-level semantic analysis on the business description information to obtain first text features, wherein the first text features are used to describe business attributes; performing text analysis on the asset ownership certificate information to obtain second text features, wherein the second text features are used to describe ownership statements; and merging the first text features and the second text features according to semantic similarity to obtain a text feature set.

[0165] The processor can further call information and application programs stored in the memory through the transmission device to perform the following steps: performing table analysis on the financial data to obtain first structured features, wherein the first structured features are used to represent financial health; performing time series analysis on the transaction log data to obtain second structured features, wherein the second structured features are used to represent fund behavior patterns; and performing standardized combination of the first structured features and the second structured features to obtain a structured feature set.

[0166] The processor can further call information and application programs stored in the memory through the transmission device to perform the following steps: performing cross-modal correlation analysis on the input text feature vector and the structured feature vector through a feature fusion layer to generate a multi-dimensional feature vector, wherein the cross-modal correlation analysis includes dimension alignment and weighted fusion; performing collaborative calculation of M risk dimensions based on the multi-dimensional feature vector through a risk quantification layer to obtain M evaluation scores, wherein the collaborative calculation includes independent score calculation of each risk dimension and associated score weight adjustment between risk dimensions; and outputting a risk assessment result containing the evaluation scores of the M risk dimensions.

[0167] The processor can further call information and application programs stored in the memory through the transmission device to perform the following steps: performing risk state classification determination based on the evaluation scores of the M risk dimensions recorded in the risk assessment result to obtain a determination result; generating a risk warning signal based on the risk level in the determination result, and generating a risk disposal scheme corresponding to the risk level.

[0168] The embodiment of the present application provides a risk assessment scheme for a financial service. In a manner combining automation and intelligence, through integrated data acquisition, multi-modal feature extraction, deep feature fusion and risk quantification evaluation, the potential risk information of a target service subject is efficiently and accurately mined from multiple heterogeneous data sources, so that the technical effects of rapid response and accurate risk assessment of the financial service application are achieved. Specifically, the financial service application submitted by the target service subject is received, and then an automatic information acquisition process is started, and information is collected from multiple data sources by using OCR, natural language processing and database access technology, so that the data comprehensiveness and real-time performance are ensured; then, the service subject information is subjected to multi-modal feature extraction processing, the semantic information of the text content and the quantitative indicators of the structured data are fused, the feature fusion layer under the deep learning framework is used for deep integration, and the service subject feature representation is generated; then, the feature representation is input into the risk assessment model, and the risk quantification layer built in the model can cooperatively calculate the evaluation scores of multiple risk dimensions, so that the multi-dimensional fine risk assessment is realized; finally, the risk warning signal is rapidly triggered based on the risk assessment result, and the technical problem of low efficiency of risk analysis on different data sources and different structured service data in the related art is solved.

[0169] Those skilled in the art can understand that, Figure 5 The structure shown is only schematic, and the electronic device can also be a smart phone, a tablet computer, a palm computer, a Mobile Internet Device (MID), a PAD terminal device, etc. Figure 5 It does not limit the structure of the above-mentioned electronic device. For example, the electronic device can further include more or less components (such as a network interface, a display device, etc.) than those shown in the figure, or have a different configuration from that shown in the figure. Figure 5 It does not limit the structure of the above-mentioned electronic device. For example, the electronic device can further include more or less components (such as a network interface, a display device, etc.) than those shown in the figure, or have a different configuration from that shown in the figure. Figure 5 It does not limit the structure of the above-mentioned electronic device. For example, the electronic device can further include more or less components (such as a network interface, a display device, etc.) than those shown in the figure, or have a different configuration from that shown in the figure.

[0170] Those skilled in the art can understand that all or part of the steps in the various methods of the above-mentioned embodiments can be completed by instructing the related hardware of the terminal device through a program, and the program can be stored in a computer readable storage medium, which can include a flash disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, etc.

[0171] The present application will be described below in combination with another alternative embodiment.

[0172] Embodiment Four

[0173] The embodiment of the present application further provides a computer readable storage medium. Optionally, in the embodiment of the present application, the computer readable storage medium can be used to save the program code executed by the risk assessment method of the financial service provided in the first embodiment.

[0174] Optionally, in the embodiment of the present application, the storage medium can be located in any one of the computer terminals in the computer terminal group in the computer network, or in any one of the mobile terminals in the mobile terminal group.

[0175] The embodiment of the present application further provides a computer program product, when executed on a data processing device, is adapted to execute the steps of the risk assessment method of the financial service: after receiving a financial service application submitted by a target service subject, collecting information of the target service subject from N heterogeneous data sources to obtain service subject information, wherein N is a positive integer; performing multi-modal feature extraction processing on the service subject information to obtain service subject features, wherein the multi-modal feature extraction includes text feature extraction and structured data feature extraction; inputting the service subject features into a risk assessment model to output a risk assessment result, wherein the risk assessment model comprises a feature fusion layer and a risk quantification layer, and the risk assessment model is used to cooperatively calculate evaluation scores of M risk dimensions through the feature fusion layer and the risk quantification layer, and M is a positive integer; triggering a risk warning signal based on the risk assessment result.

[0176] The above embodiment numbers of the present application are only for description, and do not represent the advantages and disadvantages of the embodiments.

[0177] In the above embodiments of the present application, the description of each embodiment has its own focus, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments.

[0178] In the several embodiments provided by the present application, it should be understood that the disclosed technology can be implemented in other ways. Of course, the unit embodiment described above is only schematic. For example, the division of the units is only a logical function division. There can be another division for actual implementation. For example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed coupling or direct coupling or communication connection between the units can be indirect coupling or communication connection through some interface, and can be electrical or other form.

[0179] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, that is, they can be located in one place, or distributed on a plurality of network units. According to actual needs, part or all of the units can be selected to achieve the purpose of the embodiment scheme.

[0180] In addition, each function unit in each embodiment of the present application can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software function unit.

[0181] When the integrated unit is realized in the form of a software function unit and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application, essentially or in part, or all or part of the technical solutions, can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment of the present application. The foregoing storage medium includes: a U disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a mobile hard disk, a magnetic disk or an optical disk, and various other media that can store program codes.

[0182] The above is only the preferred embodiment of the present application. It should be pointed out that, for those skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, which should be considered as the protection scope of the present application.

Claims

1. A risk assessment method for financial services, characterized in that: include: After receiving a financial business application submitted by a target business entity, information about the target business entity is collected from N heterogeneous data sources to obtain business entity information, where N is a positive integer; Performing multimodal feature extraction on the business subject information to obtain business subject features, wherein the multimodal feature extraction includes text feature extraction and structured data feature extraction; Inputting the business entity features into a risk assessment model and outputting a risk assessment result, wherein the risk assessment model includes a feature fusion layer and a risk quantification layer, and the risk assessment model is used to collaboratively calculate assessment scores of M risk dimensions through the feature fusion layer and the risk quantification layer, where M is a positive integer; A risk warning signal is triggered based on the risk assessment result.

2. The risk assessment method according to claim 1, characterized in that: The step of collecting information about the target business entity from N heterogeneous data sources to obtain the business entity information includes: Collecting application materials submitted by the target business entity, wherein the application materials include business description information in text form and financial data in tabular form; Collecting fund transaction records of the target business entity from a financial system database, wherein the fund transaction records include transaction flow data sorted by time; Collecting credit enhancement data of the target business entity from a third-party institution, wherein the credit enhancement data includes asset ownership certification information related to the target business entity; The application materials, the fund transaction records and the credit enhancement data are integrated according to preset association rules to obtain structured business entity information.

3. The risk assessment method according to claim 2, characterized in that: The step of performing multimodal feature extraction processing on the business subject information to obtain business subject features includes: Performing text feature extraction on the business description information and asset ownership certification information to obtain a text feature set; Performing structured data feature extraction on the financial data and transaction flow data to obtain a structured feature set; The business subject feature is generated, which includes the text feature set and the structured feature set.

4. The risk assessment method according to claim 3, characterized in that: The step of extracting text features from the business description information and the asset ownership certification information to obtain a text feature set includes: Performing paragraph-level semantic analysis on the service description information to obtain first-category text features, wherein the first-category text features are used to describe service attributes; Performing text parsing on the asset ownership certification information to obtain a second type of text feature, wherein the second type of text feature is used to describe the ownership statement; The first category of text features and the second category of text features are merged according to semantic similarity to obtain the text feature set.

5. The risk assessment method according to claim 3, characterized in that: The step of extracting structured data features from the financial data and transaction flow data to obtain a structured feature set includes: Performing table parsing on the financial data to obtain a first type of structured features, wherein the first type of structured features is used to characterize financial health; Performing a time series analysis on the transaction flow data to obtain a second type of structured features, wherein the second type of structured features is used to characterize fund behavior patterns; The first type of structured features and the second type of structured features are standardized and combined to obtain the structured feature set.

6. The risk assessment method according to claim 1, characterized in that: The step of inputting the business entity characteristics into a risk assessment model and outputting a risk assessment result includes: Performing cross-modal correlation analysis on the input text feature vector and the structured feature vector through the feature fusion layer to generate a multi-dimensional feature vector, wherein the cross-modal correlation analysis includes: dimension alignment and weighted fusion; Performing collaborative calculation of the M risk dimensions based on the multidimensional feature vector by the risk quantification layer to obtain M assessment scores, wherein the collaborative calculation includes: independent score calculation of each risk dimension and adjustment of associated score weights between the risk dimensions; Output the risk assessment result including the assessment scores of the M risk dimensions.

7. The risk assessment method according to claim 1, characterized in that: The step of triggering a risk warning signal based on the risk assessment result includes: Performing risk status classification based on the assessment scores of the M risk dimensions recorded in the risk assessment result to obtain a determination result; A risk warning signal is generated based on the risk level in the determination result, and a risk handling plan corresponding to the risk level is generated.

8. A risk assessment device for financial services, characterized in that: include: a collection unit configured to, upon receiving a financial service application submitted by a target business entity, collect information about the target business entity from N heterogeneous data sources to obtain business entity information, where N is a positive integer; An extraction unit, configured to perform multimodal feature extraction processing on the business subject information to obtain business subject features, wherein the multimodal feature extraction includes text feature extraction and structured data feature extraction; an input unit, configured to input the business entity features into a risk assessment model and output a risk assessment result, wherein the risk assessment model comprises a feature fusion layer and a risk quantification layer, and the risk assessment model is configured to collaboratively calculate assessment scores for M risk dimensions through the feature fusion layer and the risk quantification layer, where M is a positive integer; A triggering unit is used to trigger a risk warning signal based on the risk assessment result.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the risk assessment method for financial services according to any one of claims 1 to 7.

10. An electronic device, characterized in that: It includes one or more processors and a memory, the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the risk assessment method for financial services described in any one of claims 1 to 7.

11. A computer program product, characterized in that The method comprises computer instructions, wherein when the computer instructions are executed by a processor, the steps of the risk assessment method for financial services according to any one of claims 1 to 7 are implemented.

Images