Artificial intelligence-based data evaluation method and device, computer device and medium

Abstract

The application belongs to the technical field of artificial intelligence, and relates to a data evaluation method and device based on artificial intelligence, a computer device and a medium, which comprises the following steps: collecting target data of a user from a plurality of preset dimensions; performing feature construction processing on the target data to obtain corresponding multi-dimensional feature data; performing standardization and fusion processing on the multi-dimensional feature data to obtain corresponding fusion features; performing feature mapping processing on the fusion features based on a preset data evaluation model to obtain a potential risk embedding representation corresponding to the fusion features; performing score mapping processing on the potential risk embedding representation based on a preset scoring strategy to obtain corresponding score data; performing risk evaluation processing on the user based on the score data to obtain a corresponding risk evaluation result; and outputting the risk evaluation result. The application can be applied to a data risk evaluation scene in the field of financial technology, can realize dynamic evaluation processing on credit risks of a user, and improves the accuracy of risk evaluation.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology and can be applied to the financial technology field, particularly to data evaluation methods, devices, computer equipment and storage media based on artificial intelligence. Background Technology

[0002] In traditional financial credit assessment systems, credit risk assessment primarily relies on linear statistical models, such as logistic regression and linear discriminant analysis (LDA). These models typically assess risk based solely on structured financial data (such as income, liabilities, credit records, and loan history), offering advantages like simple calculations and high interpretability. However, their limited feature dimensions make it difficult to comprehensively depict a customer's true risk profile. Specifically, traditional models often neglect unstructured data such as customer behavioral characteristics, failing to identify potential risks from customers with similar structured data but significantly different behavioral patterns. This limitation results in low accuracy in credit risk assessment, making it difficult to adapt to the dynamically changing financial market environment.

[0003] For example, in credit guarantee insurance within the financial insurance sector, traditional models may assess underwriting risk solely based on a client's financial data and historical credit history, without fully considering dynamic factors such as the client's business practices and industry trends. If systemic risks emerge in the client's industry or their business model undergoes significant changes, traditional models may fail to adjust their risk assessment results in a timely manner, leading to unexpected payouts after the insurance company has underwritten the policy. This assessment bias not only affects the insurance company's profitability but may also weaken its ability to withstand risks in the market.

[0004] Therefore, there is an urgent need to provide a dynamic credit risk assessment method based on [the relevant technology] to improve the accuracy of risk assessment, enhance the risk management capabilities of financial institutions, and improve the overall quality of financial services. Summary of the Invention

[0005] The purpose of this application is to propose a data assessment method, apparatus, computer equipment, and storage medium based on artificial intelligence, in order to solve the technical problem that existing data assessment methods for credit risk mainly rely on linear statistical models, resulting in low accuracy of risk assessment.

[0006] Firstly, an artificial intelligence-based data evaluation method is provided, including: The system collects target data from users across multiple preset dimensions; wherein the target data includes at least financial structured data, behavioral data, and social data. The target data is subjected to feature construction processing to obtain corresponding multidimensional feature data; The multidimensional feature data is standardized and fused to obtain the corresponding fused features; Based on a preset data evaluation model, the fused features are subjected to feature mapping processing to obtain a potential risk embedding representation corresponding to the fused features; The potential risk embedding representation is subjected to a scoring mapping process based on a preset scoring strategy to obtain the corresponding scoring data; Based on the rating data, a risk assessment is performed on the user to obtain the corresponding risk assessment result; The risk assessment results are then processed for output.

[0007] Secondly, an artificial intelligence-based data evaluation device is provided, comprising: The data collection module is used to collect target data from users from multiple preset dimensions; wherein, the target data includes at least financial structured data, behavioral data, and social data. The construction module is used to perform feature construction processing on the target data to obtain corresponding multidimensional feature data; The processing module is used to standardize and fuse the multidimensional feature data to obtain the corresponding fused features; The first mapping module is used to perform feature mapping processing on the fusion feature based on a preset data evaluation model to obtain a potential risk embedding representation corresponding to the fusion feature; The second mapping module is used to perform scoring mapping processing on the potential risk embedding representation based on a preset scoring strategy to obtain the corresponding scoring data. The assessment module is used to perform risk assessment on the user based on the scoring data and obtain the corresponding risk assessment result. The output module is used to process the risk assessment results.

[0008] Thirdly, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described artificial intelligence-based data evaluation method.

[0009] Fourthly, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the steps of the aforementioned artificial intelligence-based data evaluation method.

[0010] The aforementioned scheme implemented using artificial intelligence-based data evaluation methods, devices, computer equipment, and storage media first collects user target data from multiple preset dimensions; wherein, the target data includes at least financial structured data, behavioral data, and social data; then, feature construction processing is performed on the target data to obtain corresponding multi-dimensional feature data; subsequently, the multi-dimensional feature data is standardized and fused to obtain corresponding fused features; subsequently, feature mapping processing is performed on the fused features based on a preset data evaluation model to obtain a potential risk embedding representation corresponding to the fused features; further, scoring mapping processing is performed on the potential risk embedding representation based on a preset scoring strategy to obtain corresponding scoring data; in a next step, risk assessment processing is performed on the user based on the scoring data to obtain corresponding risk assessment results; finally, the risk assessment results are output. Based on the above automated processing flow, this application constructs multidimensional feature data from user target data collected from multiple dimensions, then standardizes and fuses this multidimensional feature data to obtain fused features. Subsequently, based on a data evaluation model, feature mapping is performed on the fused features to obtain a potential risk embedding representation corresponding to the fused features. Then, based on a scoring strategy, a scoring mapping is performed on the potential risk embedding representation to obtain scoring data. Finally, based on the scoring data, user risk assessment is performed to obtain and output the risk assessment result. Thus, this application, by using a data evaluation model to perform deep feature interaction on fused features combining financial structured data, behavioral data, and social data, and by using a scoring strategy to evaluate the potential risk embedding representation output by the model, can automatically and accurately complete the dynamic assessment of user credit risk, effectively improving the accuracy of risk assessment. Attached Figure Description

[0011] To more clearly illustrate the solutions in this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is an exemplary system architecture diagram to which this application can be applied; Figure 2 This is a flowchart of an embodiment of the AI-based data evaluation method according to this application; Figure 3 This is a schematic diagram of a structure of an embodiment of the artificial intelligence-based data evaluation device according to this application; Figure 4 This is a schematic diagram of the structure of one embodiment of the computer device according to this application. Detailed Implementation

[0013] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application, are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0014] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0015] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0016] like Figure 1 As shown, system architecture 100 may include terminal device 101, network 102, and server 103. Terminal device 101 may be a laptop 1011, tablet 1012, or mobile phone 1013. Network 102 is used as a medium to provide a communication link between terminal device 101 and server 103. Network 102 may include various connection types, such as wired, wireless communication links, or fiber optic cables, etc.

[0017] Users can use terminal device 101 to interact with server 103 via network 102 to receive or send messages, etc. Various communication client applications can be installed on terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social media platform software, etc.

[0018] Terminal device 101 can be various electronic devices with a display screen and support web browsing. In addition to laptops 1011, tablets 1012, or mobile phones 1013, terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 player (Moving Picture Experts Group Audio Layer IV), a laptop computer, and a desktop computer, etc.

[0019] Server 103 can be a server that provides various services, such as a backend server that provides support for the pages displayed on terminal device 101.

[0020] It should be noted that the AI-based data evaluation method provided in this application is generally executed by a server / terminal device, and correspondingly, the AI-based data evaluation device is generally located in the server / terminal device.

[0021] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0022] Continue to refer to Figure 2 This document illustrates a flowchart of an embodiment of the AI-based data evaluation method according to this application. The order of steps in the flowchart can be changed, and some steps can be omitted, depending on different needs. The AI-based data evaluation method provided in this application can be applied to any scenario requiring data evaluation, and thus can be applied to products in these scenarios, such as data evaluation products in the financial and insurance fields. The AI-based data evaluation method includes the following steps: Step S201: Collect target data of users from multiple preset dimensions; wherein, the target data includes at least financial structured data, behavioral data and social data.

[0023] In this embodiment, the data evaluation method based on artificial intelligence runs on an electronic device (e.g., Figure 1The server / terminal device shown can collect user target data via wired or wireless connection. It should be noted that the aforementioned wireless connection methods may include, but are not limited to, 3G / 4G / 5G connections, WiFi connections, Bluetooth connections, WiMAX connections, Zigbee connections, UWB (ultra-wideband) connections, and other currently known or future-developed wireless connection methods. The implementing entity of this application is specifically a data evaluation system, which can be simply referred to as the system. This application can be applied to credit risk assessment scenarios in the financial and insurance fields. In credit risk assessment, data from a single source often cannot comprehensively and accurately reflect a customer's credit status. Therefore, collecting data from multiple dimensions and using it to subsequently construct multi-dimensional feature vectors can more comprehensively characterize a user's credit features, improving the accuracy and reliability of credit risk assessment. Financial structured data reflects a user's economic strength and repayment ability; behavioral data reflects a user's consumption and usage habits; and social data provides additional credit information from the perspective of social relationships and the social environment.

[0024] The process of collecting structured financial data includes: Data source identification: Interfacing with financial institutions' business systems, such as banks' core business systems and credit management systems. These systems record users' basic information (e.g., name, age, occupation), income (salary, investment returns), debt information (loan balance, credit card debt), and repayment records (number of on-time payments, number of overdue payments, number of overdue days, etc.). Data collection methods: A combination of periodic batch collection and real-time collection is used. Periodic batch collection extracts historical data from business systems according to a certain time period (e.g., daily, weekly); real-time collection uses technologies such as message queues to obtain the latest repayment records and other real-time data promptly. For example, for repayment records, repayment information can be sent to a message queue immediately after a user makes a payment, and the system retrieves and updates the data warehouse from the message queue in real time.

[0025] The process of collecting behavioral data includes: Determining the data source: Behavioral data primarily originates from user interaction channels with financial institutions, such as mobile banking apps, online banking, and POS machines. Consumption frequency can be calculated using user transaction records; device login patterns can record the device information (such as device model, operating system, etc.) and login location used by users when logging into mobile banking apps or online banking; transaction time distribution can analyze the time periods in which user transactions occurred. Data collection methods: Data collection modules are embedded in various interaction channels to record user behavioral data in real time. For example, in mobile banking apps, every user operation, including login, transactions, and queries, is recorded using event tracking technology, and this data is sent to a data collection server.

[0026] The process of collecting social data includes: Determining data sources: Social data can be obtained through various channels, such as social media platforms and professional social networks. Relationship networks can reflect a user's social relationships with others, including the number of friends and interaction frequency; professional circles can provide information about a user's professional background and professional circles; risk exposure can be assessed by analyzing risk information (such as fraudulent information, negative credit information, etc.) encountered by the user on social networks. Data collection methods: Collaborating with social media platforms and professional social networks to obtain relevant social data through API interfaces. Simultaneously, web scraping technology can also be used to collect data from publicly available social network pages, provided it is legal and compliant. For example, by scraping a user's social media profile, information such as their friend list and posted content can be obtained.

[0027] Step S202: Perform feature construction processing on the target data to obtain corresponding multidimensional feature data.

[0028] In this embodiment, the specific implementation process of performing feature construction processing on the target data to obtain the corresponding multidimensional feature data will be further described in detail in subsequent specific embodiments of this application, and will not be elaborated on here.

[0029] Step S203: Standardize and fuse the multidimensional feature data to obtain the corresponding fused features.

[0030] In this embodiment, the specific implementation process of standardizing and fusing the multidimensional feature data to obtain the corresponding fused features will be further described in detail in subsequent specific embodiments of this application, and will not be elaborated on here.

[0031] Step S204: Based on a preset data evaluation model, perform feature mapping processing on the fused features to obtain the potential risk embedding representation corresponding to the fused features.

[0032] In this embodiment, the fusion features can be used. The input is fed into a pre-trained deep learning network data evaluation model. This model has learned the complex nonlinear relationships between features in a large amount of sample data and can automatically discover the correlation between high-order feature combinations and potential risks. The feature mapping process mentioned above refers to the nonlinear interactive learning process. The specific implementation includes: in the deep learning network, the fused features are first processed by the first layer of the neural network. The first layer of the neural network performs a linear transformation on the input features, as shown in the formula... ,in, It is the weight matrix of the first layer of the neural network. It is the bias vector. Then it is passed through the activation function. (such as the ReLU function, The linear transformation (x) = max(0,x) is applied to the result of the linear transformation with a nonlinear activation to obtain the output of the first layer. Next, the output of the first layer is used as the input to the second layer of the neural network, undergoing similar linear transformations and nonlinear activation operations, as shown in the formula: ,in, It is the weight matrix of the second layer of the neural network. It is the bias vector. Finally, it passes through the second activation function. The processing yields a potential risk embedding representation corresponding to the user: This potential risk is embedded in the representation. It is a low-dimensional vector that integrates all the information from the original features and, through non-linear interactive learning of the data evaluation model, more accurately reflects the user's credit risk status. For example, the data evaluation model can identify the relationship between high-order feature combinations learned during the training phase, such as "high consumption frequency + abnormal login region," and potential fraud risk, and incorporate this relationship into the potential risk embedding representation.

[0033] Furthermore, the training and generation process of the aforementioned data evaluation model will be described in more detail in subsequent specific embodiments of this application, and will not be elaborated upon here.

[0034] Step S205: Perform a scoring mapping process on the potential risk embedding representation based on a preset scoring strategy to obtain the corresponding scoring data.

[0035] In this embodiment, the specific implementation process of performing scoring mapping on the potential risk embedding representation based on the preset scoring strategy to obtain the corresponding scoring data will be further described in detail in subsequent specific embodiments of this application, and will not be elaborated on here.

[0036] Step S206: Perform risk assessment on the user based on the rating data to obtain the corresponding risk assessment result.

[0037] In this embodiment, the specific implementation process of performing risk assessment on the user based on the scoring data to obtain the corresponding risk assessment result will be further described in detail in subsequent specific embodiments of this application, and will not be elaborated on here.

[0038] Step S207: Output the risk assessment results.

[0039] In this embodiment, the generated risk assessment results can be sent to users and relevant business personnel via email, text message, or interface display, thereby completing the output processing of the risk assessment results.

[0040] This application first collects target data from users from multiple preset dimensions; wherein, the target data includes at least financial structured data, behavioral data, and social data; then, it performs feature construction processing on the target data to obtain corresponding multidimensional feature data; subsequently, it performs standardization and fusion processing on the multidimensional feature data to obtain corresponding fused features; next, it performs feature mapping processing on the fused features based on a preset data evaluation model to obtain a potential risk embedding representation corresponding to the fused features; further, it performs scoring mapping processing on the potential risk embedding representation based on a preset scoring strategy to obtain corresponding scoring data; then, it performs risk assessment processing on the user based on the scoring data to obtain corresponding risk assessment results; finally, it outputs the risk assessment results. Based on the above automated processing flow, this application constructs multidimensional feature data from user target data collected from multiple dimensions, then standardizes and fuses this multidimensional feature data to obtain fused features. Subsequently, based on a data evaluation model, feature mapping is performed on the fused features to obtain a potential risk embedding representation corresponding to the fused features. Then, based on a scoring strategy, a scoring mapping is performed on the potential risk embedding representation to obtain scoring data. Finally, based on the scoring data, user risk assessment is performed to obtain and output the risk assessment result. Thus, this application, by using a data evaluation model to perform deep feature interaction on fused features combining financial structured data, behavioral data, and social data, and by using a scoring strategy to evaluate the potential risk embedding representation output by the model, can automatically and accurately complete the dynamic assessment of user credit risk, effectively improving the accuracy of risk assessment.

[0041] In some alternative implementations, step S202 includes the following steps: The target data is preprocessed to obtain the corresponding processed data.

[0042] In this embodiment, the original target data collected often has problems such as missing values, outliers, and duplicate values, which need to be preprocessed to improve data quality.

[0043] Specifically, preprocessing includes: handling missing values: for data with few missing values, methods such as mean, median, and mode can be used to fill in the missing values; for data with many missing values, if the feature has a significant impact on credit assessment, it is advisable to re-collect or delete the feature. For example, for a small number of missing values ​​in income data, the average income of users in the same industry or region can be used to fill in the missing values.

[0044] Outlier Handling: Outliers are identified using statistical methods (such as box plots and the 3σ principle) and handled according to the actual situation. Outliers that are clearly erroneous or inconsistent with reality can be deleted directly; reasonable outliers can be retained or appropriately adjusted. For example, if a user's spending far exceeds their normal income level and cannot be reasonably explained, it can be considered an outlier and handled accordingly.

[0045] Data transformation and encoding: For categorical features, such as gender and occupation, one-hot encoding or label encoding are used to convert them into numerical form so that the model can process them. For numerical features, normalization or standardization is performed as needed to eliminate differences in units and make different features comparable. For example, numerical features such as age and income are normalized to map their values ​​to the interval [0, 1].

[0046] Feature extraction is performed on the processed data to obtain initial features related to risk assessment.

[0047] In this embodiment, the feature extraction includes: extracting credit risk-related features from financial structured data, behavioral data, and social data respectively, and integrating them to obtain corresponding initial features. Specifically, features such as income-to-debt ratio, number of overdue payments, and credit limit utilization rate are extracted from financial structured data; features such as consumption frequency fluctuation, device login frequency, and transaction time entropy are extracted from behavioral data; and features such as social network centrality, occupational circle stability, and risk exposure index are extracted from social data.

[0048] Among these features, the following can be extracted from structured financial data such as bank statements, credit records, and balance sheets: income-to-debt ratio, number of overdue payments, and credit limit utilization rate. Specifically, the calculation logic for the income-to-debt ratio includes: Income: The sum of stable income such as salary and investment returns over the past 12 months (noise removal is required to exclude abnormally large income amounts). Debt: The sum of loan balances, outstanding credit card payments, and overdue penalties. Formula: Income-to-debt ratio = Average monthly income / Average monthly debt. Significance: Reflects a user's long-term debt repayment ability; a low ratio may indicate high risk. The statistical method for the number of overdue payments includes: differentiating overdue levels (e.g., 1-30 days, 30-90 days, over 90 days), and calculating the total number of overdue payments within different time windows (the past 3 / 6 / 12 months). Weighted processing: Severe overdue payments (e.g., >90 days) are given higher weight. The calculation logic for the credit limit utilization rate includes: Numerator: Average daily balance of the credit card over the past 6 months; Denominator: Credit limit. Formula: Credit limit utilization rate = (Average balance / Credit limit) × 100%.

[0049] Additionally, the following features can be extracted from behavioral data such as transaction records, device logs, and APP usage records: consumption frequency fluctuation, device login frequency, and transaction time entropy. Specifically, the calculation logic for consumption frequency fluctuation includes: counting the number of transactions weekly / monthly and calculating the standard deviation or coefficient of variation (standard deviation / mean). Example: User A's monthly consumption frequency is [10, 12, 8, 15], and large fluctuations may indicate unstable consumption habits. Business significance: High-frequency and stable consumption is usually associated with low risk. The statistical method for device login frequency includes: counting the number of times a user logs into the financial APP using different devices (mobile phone, computer) in the past 30 days. Anomaly detection: If the login frequency of a certain device is much higher than the historical average, it may indicate account risk. The calculation logic for transaction time entropy includes: dividing a 24-hour day into 6 time periods (e.g., 0-4 am is "night") and counting the distribution ratio of user transactions in each time period. Calculating transaction time entropy: Entropy = -Σ(p_i × log(p_i)), where p_i is the proportion of transactions in the i-th time period. Business significance: A high transaction time entropy value indicates that the transaction time is scattered (such as part-time or night workers), and needs to be judged in conjunction with occupational characteristics; a low value may be due to regular office workers.

[0050] In addition, the following features can be extracted from social data, including address books, social platform relationships, and professional information: social network centrality, professional circle stability, and risk exposure index. Specifically, the calculation logic for social network centrality includes: constructing a user's social relationship graph (e.g., the number of contacts in the address book, interaction frequency). Calculating "degree centrality": the percentage of high-credit contacts directly associated with the user. Business significance: Users with high centrality may be more socially constrained, resulting in higher costs for default. The statistical methods for professional circle stability include: judging professional stability based on the proportion of colleagues / peers in the address book, the frequency of changes in work email addresses, etc. Example: A stable professional circle is defined as one where the percentage of newly added colleagues in the past year is less than 10%. The calculation logic for the risk exposure index includes: statistically analyzing the proportion of high-risk individuals in the user's social circle (e.g., users with overdue payments, multiple borrowers). Weighting rule: the higher the frequency of interaction with risky users, the higher the index.

[0051] The initial features are processed using a preset mining strategy to obtain the corresponding generated features.

[0052] In this embodiment, the specific implementation process of performing feature mining processing on the initial features based on the preset mining strategy to obtain the corresponding generated features will be further described in detail in subsequent specific embodiments of this application, and will not be elaborated on here.

[0053] The initial features and the generated features are integrated to obtain the corresponding integrated features.

[0054] In this embodiment, all the extracted and derived features are combined into a multi-dimensional feature vector in a certain order: , where n represents the feature dimension, which can exceed thousands depending on the richness of the data.

[0055] Specifically, the above integration process may include: 1) Feature alignment and standardization, including: Time window alignment: ensuring all features are based on the same time range (e.g., the last 12 months). Missing value handling: Numerical features: filled with median or business common sense (e.g., missing "device login frequency" is set to 0). Categorical features: set to "unknown" category separately. Standardization: Z-score standardization or Min-Max normalization is performed on continuous features (e.g., income-to-debt ratio). One-hot encoding is performed on categorical features (e.g., occupation type). 2) Using ranking rules to sort and combine all aligned and standardized features to obtain a multidimensional feature vector. The ranking rules include: grouping by feature type (financial → behavioral → social). Ranking within the same type by business importance (e.g., "number of overdue payments" takes precedence over "credit limit utilization rate").

[0056] The integrated features are used as the multidimensional feature data.

[0057] Based on the above processing flow, this application preprocesses the target data to obtain processed data, extracts features from the processed data to obtain initial features related to risk assessment, and then intelligently performs feature mining processing on the initial features based on the use of mining strategies to obtain generated features. Subsequently, the initial features and generated features are integrated and processed, and the generated integrated features are used as corresponding multidimensional feature data. This enables efficient and accurate completion of feature construction processing of the target data, ensuring the accuracy, standardization, and diversity of the generated multidimensional feature data.

[0058] In some optional implementations of this embodiment, the step of performing feature mining processing on the initial features based on a preset mining strategy to obtain corresponding generated features includes the following steps: The initial features are processed by a preset feature combination method to obtain the corresponding combined features.

[0059] In this embodiment, the aforementioned feature combination method refers to a way to generate new combined features by combining extracted features in order to mine deeper information from the data. Specifically, the income-to-debt ratio from the financial dimension is combined with the consumption frequency fluctuation from the behavioral dimension to generate the "income-to-debt ratio - consumption frequency fluctuation" feature. This feature may better reflect the relationship between users' economic pressure and consumption behavior. Based on the business assumption that users with high debt and large consumption fluctuations may face greater economic pressure and higher default risk, a new combined feature can be generated by multiplying the income-to-debt ratio from the financial dimension with the consumption frequency fluctuation from the behavioral dimension.

[0060] The initial features are processed using a preset feature derivation method to obtain corresponding derived features.

[0061] In this embodiment, the aforementioned feature derivation method refers to generating new features by deriving from extracted features in order to mine deeper information from the data. Specifically, the derived features include a first derived feature of "transaction time entropy + device login frequency" and a second derived feature of "social network centrality + risk exposure index". The first derived feature can be obtained by standardizing "transaction time entropy" and "device login frequency" and then weighting and summing them (e.g., weights 0.6:0.4). This first derived feature can comprehensively reflect the regularity of user behavior and help identify abnormal patterns (e.g., high-frequency trading at night + new device login). In addition, the generation logic of the second derived feature includes: combining "social network centrality" and "risk exposure index": Score = Centrality × (1 - Risk Exposure Index). Business explanation: Users with high centrality but few exposures to risky groups score higher.

[0062] The combined features and the derived features are integrated to obtain the corresponding processed features.

[0063] In this embodiment, the generated combined features and derived features can be integrated and processed, and the resulting processed features can be used as the corresponding generated features.

[0064] The processing features are used as the generated features.

[0065] Based on the above processing flow, this application uses a feature combination method to process the initial features to obtain combined features, and uses a feature derivation method to process the initial features to obtain derived features. Then, the generated combined features and derived features are integrated and processed, and the processed features are used as the corresponding generated features. This achieves automatic and intelligent feature mining processing of the initial features, and improves the richness and multidimensionality of the generated features.

[0066] In some alternative implementations, step S203 includes the following steps: Obtain the preset standardization strategy and the preset weight allocation strategy.

[0067] In this embodiment, since the data units and value ranges of different data sources may vary significantly, in order to eliminate the impact of these differences on the model evaluation results, it is necessary to standardize the feature data of new customers according to the standards and methods determined during the training phase of the aforementioned data evaluation model. Specifically, the standardization strategy can employ the standard deviation standardization method (Z-score standardization), with the following formula: .in, This represents the original value of the j-th feature of the i-th customer. Let represent the mean of the j-th feature. Let represent the standard deviation of the j-th feature. Standardization transforms the values ​​of each feature into a distribution with a mean of 0 and a standard deviation of 1, making different features comparable.

[0068] In addition, the weight allocation strategy described above adopts the weight allocation method (such as the Analytic Hierarchy Process (AHP) or the entropy weight method) determined during the training phase of the data evaluation model to assign weights to the various characteristics of new customers.

[0069] Specifically, the Analytic Hierarchy Process (AHP) is a weighting method that combines qualitative and quantitative approaches. First, a hierarchical model is constructed, decomposing the credit assessment problem into an objective layer, a criterion layer, and a solution layer. Then, the relative importance of elements in each layer is determined through pairwise comparisons, constructing a judgment matrix. Next, the judgment matrix undergoes a consistency check to ensure the rationality and consistency of the judgments. Finally, the weight vector of each element is calculated. For example, in credit assessment, financial structured data, behavioral data, and social data can be used as the criterion layer. Pairwise comparisons are performed on each feature under the criterion layer to determine their relative importance, and then the weight of each feature is calculated.

[0070] Furthermore, the feature weight generation process based on the entropy weight method includes: firstly, calculating the information entropy value of each standardized feature. Entropy reflects the amount of information contained in a feature; the greater the information content, the smaller the entropy. Then, the weight of each feature is calculated based on the entropy value. (The weight of the j-th feature is in the range [0,1], and the sum of all feature weights is 1), formula .in, The entropy values ​​of all features are summed to form the normalized denominator, where k is the total number of features. Through weight allocation, the relative importance of different features in credit assessment is determined, enabling the model to more rationally utilize the information from each feature for evaluation.

[0071] The multidimensional feature data is standardized based on the standardization strategy to obtain the corresponding first feature data.

[0072] In this embodiment, the multidimensional feature data can be standardized based on the above-mentioned standardization strategy, namely the selected standard deviation standardization method, to obtain the processed first feature data.

[0073] The first feature data is processed by weight allocation based on the weight allocation strategy to obtain the corresponding weight data.

[0074] In this embodiment, the first feature data can be weighted based on the weight allocation strategy, i.e., the selected weight allocation method, to obtain the weighted data after allocation.

[0075] The first feature data and the weighted data are fused based on a preset weighted fusion strategy to obtain the corresponding second feature data.

[0076] In this embodiment, based on the calculated weighted data, the standardized features (first feature data) of the new customer (i.e., user) are weighted and fused to obtain the corresponding second feature data, i.e., the fused feature. Feature fusion integrates information from various features and takes into account the importance of different features, providing a more reasonable and effective input for subsequent deep feature interactions.

[0077] The second feature data is used as the fused feature.

[0078] Based on the above processing flow, this application standardizes multidimensional feature data using a standardization strategy to obtain first feature data, and then assigns weights to the first feature data using a weight allocation strategy to obtain weighted data. Finally, it fuses the first feature data and weighted data using a weighted fusion strategy, and uses the resulting second feature data as the corresponding fused feature. This allows for efficient and accurate standardization and fusion of multidimensional feature data. Standardization converts data of different dimensions into a unified standard, making the data comparable; weighted fusion assigns reasonable weights to each feature based on the importance of different data sources, increasing the model's focus on important features. Furthermore, subsequent feature mapping processing of the fused features using a data evaluation model effectively improves the accuracy and stability of the model processing.

[0079] In some optional implementations, the training and generation process of the aforementioned data assessment model involves collecting structured financial data (such as income, liabilities, and repayment records), behavioral data (such as consumption frequency, device login patterns, and transaction time distribution), and social data (such as relationship networks, professional circles, and risk exposure) from a large amount of sample data, and constructing multi-dimensional feature vectors. Next, the different data sources are standardized, and weights are assigned using methods such as the Analytic Hierarchy Process (AHP) or entropy weighting to complete the standardization and weight fusion of multi-source data. Then, a deep learning network is used to perform non-linear interactive learning on the multi-dimensional features. By continuously adjusting the neural network weights to minimize prediction errors, a well-trained data assessment model is finally obtained. This model can accurately learn the relationship between features and credit risk and has the ability to map input features into potential risk embedding representations.

[0080] Deep feature interaction and model training include: 1. Select a Deep Learning Network Model. Based on the characteristics of the credit assessment problem and the data features, choose a suitable deep learning network model. Commonly used models include DeepFM and the Wide & Deep Model. DeepFM: This model combines the advantages of Factorization Machines (FM) and Deep Neural Networks (DNNs), enabling it to learn both low-order and high-order feature combinations simultaneously. The FM part captures second-order interactions between features, while the DNN part automatically learns higher-order nonlinear interactions, thus more comprehensively mining information from the data. Wide & Deep Model: This model consists of a Wide part and a Deep part. The Wide part is a linear model used to learn and memorize simple rules and patterns between features; the Deep part is a deep neural network used to learn and discover complex nonlinear relationships between features. By combining the Wide and Deep parts, the model can improve generalization ability while maintaining prediction accuracy.

[0081] 2. Model Training and Optimization. Using fused features as input and credit risk labels (such as whether a payment is overdue, credit rating, etc.) as output, the selected deep learning network model is trained. 1) Dividing the data into training, validation, and test sets: The collected sample data is divided into training, validation, and test sets according to a certain ratio (e.g., 7:2:1). The training set is used for model training and learning, the validation set is used to adjust the model's hyperparameters and evaluate its performance, and the test set is used to finally evaluate the model's generalization ability. 2) Defining the loss function and optimization algorithm: Based on the characteristics of the credit assessment problem, a suitable loss function is selected, such as the cross-entropy loss function (for classification problems) or the mean squared error loss function (for regression problems). Simultaneously, a suitable optimization algorithm, such as stochastic gradient descent (SGD) or Adam, is selected to adjust the model's parameters to minimize the loss function. 3) Training the model: The training data is input into the model. The model's output is calculated through forward propagation, and then the prediction error is calculated based on the loss function. Finally, the neural network weights of the model are adjusted through the backpropagation algorithm to minimize the prediction error. Repeat this process until the model's performance on the validation set stabilizes or meets preset requirements. 4) Model tuning: During training, optimize model performance by adjusting hyperparameters (such as learning rate, batch size, number of network layers, number of neurons, etc.). Grid search, random search, and other methods can be used for hyperparameter tuning to find the optimal combination of hyperparameters.

[0082] 3. Model Evaluation and Validation. 1) Evaluation Metric Selection. Choose appropriate evaluation metrics to assess model performance. Commonly used metrics include accuracy, precision, recall, F1 score, and AUC. 2) Model Validation. Validate the trained model using test set data and calculate the evaluation metric values ​​on the test set. If the model performs well on the test set, it indicates good generalization ability and can be used for actual credit risk assessment, serving as the final data evaluation model. If the model performs poorly on the test set, further analysis of the reasons is needed, and the model should be optimized and improved.

[0083] Through the complete and rigorous training process described above, a well-trained data evaluation model is finally obtained. This model can accurately learn the relationship between features and credit risk and has the ability to map input features into potential risk embedding representations, providing a solid foundation for subsequent credit risk assessment of users.

[0084] In some alternative implementations, step S205 includes the following steps: Call the preset scoring function.

[0085] In this embodiment, the scoring function is specifically as follows: .in, This represents the embedded representation of potential risk. The scoring function described above is designed based on the principle of logistic regression. The logistic regression function has the characteristic that its output range is between (0,1). By multiplying it by 1000, it is mapped to the interval [0, 1000], which better meets the actual needs of credit scoring.

[0086] The potential risk embedding representation is calculated based on the scoring function to obtain the corresponding calculation result.

[0087] In this embodiment, the potential risk embedding representation can be substituted into the scoring function for calculation, and the calculation result can be used as the corresponding scoring data.

[0088] The calculation results are used as the scoring data.

[0089] This application calls a preset scoring function; then, based on the scoring function, it calculates and processes the potential risk embedding representation to obtain the corresponding calculation result; subsequently, the calculation result is used as the scoring data. Based on the above processing flow, this application calculates and processes the potential risk embedding representation using the scoring function, and uses the calculated scoring data as the corresponding scoring data, thereby achieving efficient and accurate scoring mapping processing of the potential risk embedding representation and ensuring the accuracy of the generated scoring data.

[0090] In some optional implementations of this embodiment, step S206 includes the following steps: Call the preset risk level classification table.

[0091] In this embodiment, the aforementioned risk level classification table is a pre-constructed data table based on actual credit assessment needs, showing a one-to-one correspondence between scoring ranges and customer risk levels. Specifically, the scoring range is set to [300, 1000], where, for the scoring data... , Customers with a credit score >800 are considered low-risk; these customers typically have high credit ratings and a low risk of default. Customers with a credit score <600... Customers with a credit score <800 are considered medium-risk; their creditworthiness is generally average, and they carry a certain risk of default. Customers with a credit score <300 are considered medium-risk. Customers with a credit score of less than 600 are considered high-risk customers, as they have lower credit ratings and a higher risk of default.

[0092] The risk level classification table is queried based on the scoring data to determine the target scoring range that matches the scoring data.

[0093] In this embodiment, the risk level classification table can be queried based on the use of the scoring data to find the target scoring range that matches the scoring data. For example, if the value of the scoring data is 500, then the target scoring range that matches the scoring data is [300, 600]. If the value of the scoring data is 700, then the target scoring range that matches the scoring data is [600, 800].

[0094] Obtain the target risk level corresponding to the target score range from the risk level classification table.

[0095] In this embodiment, after determining the target rating interval that matches the aforementioned rating data, the target risk level corresponding to the target rating interval can be further queried from the aforementioned risk level classification table, and this target risk level is used as the user's risk assessment result. For example, if the rating data value is 500, then the target rating interval matched by the rating data is [300, 600], and the corresponding target risk level can be queried from the aforementioned risk level classification table as a high-risk customer. Conversely, if the rating data value is 700, then the target rating interval matched by the rating data is [600, 800], and the corresponding target risk level can be queried from the aforementioned risk level classification table as a medium-risk customer.

[0096] The target risk level is used as the risk assessment result.

[0097] Based on the above processing flow, this application uses the scoring data to query the risk level classification table to determine the target scoring range that matches the scoring data. Then, it obtains the target risk level corresponding to the target scoring range from the risk level classification table and uses the obtained target risk level as the user's risk assessment result. Thus, it can automatically and accurately complete the risk assessment of the user based on the use of scoring data, improve the processing efficiency of risk assessment, and ensure the accuracy of the obtained risk assessment results.

[0098] In some optional implementations of this embodiment, after step S207, the electronic device may further perform the following steps: Collect feedback data related to risk assessment from pre-defined data sources.

[0099] In this embodiment, the process of collecting the aforementioned feedback data includes: Data source determination: Clearly defining the source of real-time feedback data, including internal business systems of financial institutions (such as loan management systems, repayment reminder systems, etc.), user feedback channels (such as online customer service, complaint platforms, etc.), and external credit information sharing platforms. For example, a loan management system can provide users' repayment records, including information on on-time repayment, delayed repayment, and default; online customer service can collect user feedback and complaints regarding credit assessment results. Data collection method: Employing real-time data collection technologies, such as message queues or streaming data processing frameworks, to ensure timely acquisition of the latest feedback data. For example, when a user makes a repayment on the repayment date, the loan management system immediately sends the repayment information to the message queue, and the model can retrieve and process this data from the message queue in real time. Data preprocessing: Cleaning and preprocessing the collected real-time feedback data to remove noise and outliers, ensuring the quality of the final feedback data. For example, verifying and correcting abnormal amounts or times in repayment records; filtering and organizing invalid information in complaint data.

[0100] Get the preset reward function.

[0101] In this embodiment, the definition of the reward function includes: 1) Setting a target default rate: Based on the financial institution's risk appetite and business objectives, a reasonable target default rate is set. The target default rate can be a fixed value or dynamically adjusted according to different business scenarios and user groups. For example, for high-risk businesses, the target default rate can be set relatively high; for low-risk businesses, the target default rate can be set relatively low.

[0102] 2) Reward Function Construction: Construct a reward function based on the set target default rate. The reward function measures the model's decision-making performance at the current time step. If the actual default rate is lower than the target default rate, the model's decision-making performance is good, and the reward function value is positive; conversely, if the actual default rate is higher than the target default rate, the model's decision-making performance is poor, and the reward function value is negative. For example, assuming the target default rate is 5% and the actual default rate at the current time step is 3%, the reward function value is -(3%-5%)=2%; if the actual default rate is 7%, the reward function value is -(7%-5%)=-2%.

[0103] Invoke the preset reinforcement learning algorithm.

[0104] In this embodiment, the reinforcement learning algorithm described above can specifically employ the policy gradient method, and the parameter updates of the model follow the policy gradient method. Specifically, parameter updates based on the policy gradient method include: policy gradient calculation: The policy gradient method is a gradient-based reinforcement learning algorithm used to update the model's parameters. First, the policy gradient needs to be calculated. ,in Indicating in strategy Lower reward function The expected value. Methods for calculating the policy gradient typically involve adjusting the model parameters. The derivative can be obtained using automatic differentiation tools (such as TensorFlow, PyTorch, etc.) to simplify the computation process. For example, in neural network models, the policy gradient can be calculated using the backpropagation algorithm.

[0105] Learning rate determination: Learning rate The learning rate is an important hyperparameter that controls the step size for updating model parameters. An excessively large learning rate may cause the model parameters to update too quickly, preventing convergence to the optimal solution; an excessively small learning rate may cause the model parameters to update too slowly, resulting in excessively long training times. A suitable learning rate can be determined through experimentation and tuning, for example, using methods such as grid search or random search.

[0106] Application of the parameter update formula: Based on the calculated policy gradient and the determined learning rate, apply the parameter update formula. To update the model parameters. For example, if the current model parameters are... The policy gradient is The learning rate is The updated model parameters are .

[0107] Based on the reward function, the reinforcement learning algorithm and the feedback data are used to optimize the model parameters of the data evaluation model.

[0108] In this embodiment, the optimization of the model parameters refers to the dynamic adjustment of the feature sensitivity of the data evaluation model (hereinafter referred to as the model) to achieve real-time self-learning of credit scoring. The specific implementation process includes: 1) Feature importance analysis: During the model parameter update process, the dependence of the model on different features is analyzed to understand the influence of each feature on the credit score. Feature importance assessment methods, such as gradient-based feature importance assessment methods, can be used to calculate the contribution of each feature in the model prediction. For example, for a neural network model, its importance can be assessed by calculating the gradient of the neuron corresponding to each feature. 2) Sensitivity adjustment strategy formulation: Based on the feature importance analysis results, a feature sensitivity adjustment strategy is formulated. If the importance of a feature is high, it means that the feature has a greater impact on the credit score, and the sensitivity of the model to that feature can be appropriately increased; conversely, if the importance of a feature is low, it means that the feature has a smaller impact on the credit score, and the sensitivity of the model to that feature can be appropriately decreased. For example, if it is found that the user's income feature has a greater impact on the credit score, the model parameters can be adjusted to make the model pay more attention to changes in the user's income. 3) Real-time Adjustment: During model operation, the model's sensitivity to various features is adjusted in real time based on feedback data and parameter updates. By continuously optimizing model parameters, the model can better adapt to changes in different features, improving the accuracy and reliability of credit scoring. For example, when a user's consumption behavior changes significantly, the model can promptly adjust its sensitivity to consumption characteristics to reflect changes in the user's credit risk.

[0109] This application collects risk assessment-related feedback data from a pre-defined data source; then obtains a pre-defined reward function; subsequently, it calls a pre-defined reinforcement learning algorithm; and finally, based on the reward function, it uses the reinforcement learning algorithm and the feedback data to optimize the model parameters of the data assessment model. Based on this process, in credit risk assessment scenarios, market environments, user behavior, and other factors are constantly changing. Static credit scoring models are difficult to adapt to these dynamic changes, which may lead to inaccurate assessment results. Therefore, this application introduces a dynamic model update and reinforcement learning self-optimization mechanism. By continuously receiving real-time feedback data and automatically adjusting model parameters using a reinforcement learning algorithm, the model can adapt to new data and situations in real time, maintaining high assessment accuracy.

[0110] In some optional implementations of this embodiment, the system also has the functions of interpretability of results and blockchain audit records. The specific implementation process includes: 1. Generation of Scoring Results and Key Feature Contribution Values. Scoring Result Calculation: Using a trained credit scoring model, the user's credit is assessed, generating the current user's credit score. Credit scores can be specific numerical values ​​or classification levels (such as excellent, good, average, poor, etc.). For example, in a neural network model, a user's feature vector is input into the model, and the user's credit score is calculated through forward propagation. Key feature contribution calculation: Analyze the degree of contribution of each feature to the credit score to determine the contribution value of key features. Feature importance assessment methods, such as SHAP (SHapley Additive exPlanations) values, can be used to calculate the marginal contribution of each feature to the credit score. SHAP values ​​consider the influence of all feature combinations and can more accurately reflect the importance of each feature. For example, for features such as a user's income, debt, and spending frequency, their SHAP values ​​can be calculated separately to determine their contribution to the credit score.

[0111] 2. Hash signature generation. Data concatenation: Combine the current user's credit score... Embedded representation of potential customer risks and timestamp Concatenate the strings to form a single string. For example, concatenate the strings to form a single string. , and Concatenate them into a long string according to a certain format, such as " Hash Algorithm Selection: Choose a suitable hash algorithm, such as Sha256, to perform hash calculations on the concatenated string. Sha256 is a commonly used hash algorithm with high security and uniqueness, and it can generate fixed-length hash values. For example, using the Sha256 algorithm to calculate the hash value of the concatenated string yields a 256-bit hash value. , .

[0112] 3. Blockchain On-Chain Recording. Blockchain Platform Selection: Choose a suitable blockchain platform, such as Ethereum or Hyperledger Fabric, for recording hash signatures and audit logs. Different blockchain platforms have different characteristics and applicable scenarios, requiring selection based on actual needs. For example, Ethereum is a public blockchain platform with high openness and transparency; Hyperledger Fabric is a consortium blockchain platform suitable for enterprise-level applications. Smart Contract Development: Develop smart contracts to record hash signatures and audit logs on the blockchain. A smart contract is an automatically executing computer program that can implement specific business logic on the blockchain. For example, develop a smart contract that, upon receiving a hash signature and audit log, stores them on the blockchain and ensures the data is immutable. On-Chain Operation Execution: Execute the generated hash signature... Audit logs are recorded on the blockchain via smart contracts. This on-chain operation can be achieved through blockchain client tools or API interfaces. For example, using Ethereum's Web3.js library, a function in the smart contract can be called to send the hash signature and audit logs to the blockchain.

[0113] 4. Audit Chain Formation and Regulatory Review. Audit Chain Construction: Over time, new hash signatures and audit logs are continuously recorded on the chain, forming a verifiable audit chain. The audit chain records all strategy evolution processes of the credit scoring model, including model parameter updates and feature sensitivity adjustments. For example, after each model update, a corresponding hash signature and audit log are generated and recorded on the chain, forming a complete audit chain. Regulatory Review Mechanism Establishment: Provide regulators with interfaces and tools to access the blockchain, enabling them to review historical risk records at any time. Regulators can query and verify data on the audit chain through blockchain explorers or dedicated regulatory platforms. For example, regulators can enter a user ID or time range to query a specific user's credit scoring history and model strategy evolution process. Public Trust Guarantee: The immutability and transparency of the blockchain ensure the credibility of risk control results. Because the data on the audit chain cannot be tampered with, regulators and users can trust the evaluation results and strategy evolution process of the credit scoring model. For example, if a user disagrees with the credit scoring result, the regulator can query relevant data and records through the audit chain to make a fair ruling.

[0114] In some alternative implementations, the user information obtained is subject to user consent and complies with relevant laws and policies.

[0115] Furthermore, any software tools or components not belonging to our company that appear in the embodiments of this application are merely illustrative examples and do not represent actual use.

[0116] Furthermore, this application represents a leap from "static scoring" to "dynamic evolutionary credit assessment" on the basis of traditional credit scoring models. Through multi-dimensional feature fusion and deep feature interaction, the system can more comprehensively depict customer risk profiles; the reinforcement learning module enables the model to continuously adjust its parameters based on real-time data, thereby continuously optimizing the scoring results and significantly improving the sensitivity and accuracy of risk identification.

[0117] Meanwhile, the introduction of blockchain technology to record the evolution of each score and model parameters ensures the transparency, compliance, and auditability of the credit scoring system. The fusion and interpretation mechanism of multi-source heterogeneous data enables the model to provide reliable assessments even when facing new customers (without historical credit data), enhancing the financial inclusion capabilities of financial institutions.

[0118] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0119] It should be emphasized that, to further ensure the privacy and security of the above risk assessment results, the risk assessment results can also be stored in a blockchain node.

[0120] The blockchain referred to in this application is a novel application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms. Essentially, a blockchain is a decentralized database, a chain of data blocks linked together using cryptographic methods. Each data block contains information about a batch of network transactions, used to verify the validity of the information (anti-counterfeiting) and generate the next block. A blockchain can include an underlying blockchain platform, a platform product service layer, and an application service layer.

[0121] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results. Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0122] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing related hardware through computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, optical disk, or read-only memory (ROM), or random access memory (RAM).

[0123] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0124] Further reference Figure 3 As a response to the above Figure 2 To implement the method shown, this application provides an embodiment of an artificial intelligence-based data evaluation device, which is similar to... Figure 2 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0125] like Figure 3 As shown, the AI-based data evaluation device 300 described in this embodiment includes: a data acquisition module 301, a construction module 302, a processing module 303, a first mapping module 304, a second mapping module 305, an evaluation module 306, and an output module 307. Wherein: The data collection module 301 is used to collect target data of users from multiple preset dimensions; wherein, the target data includes at least financial structured data, behavioral data and social data; The construction module 302 is used to perform feature construction processing on the target data to obtain corresponding multidimensional feature data; Processing module 303 is used to standardize and fuse the multidimensional feature data to obtain the corresponding fused features; The first mapping module 304 is used to perform feature mapping processing on the fusion feature based on a preset data evaluation model to obtain a potential risk embedding representation corresponding to the fusion feature; The second mapping module 305 is used to perform a scoring mapping process on the potential risk embedding representation based on a preset scoring strategy to obtain the corresponding scoring data. Assessment module 306 is used to perform risk assessment processing on the user based on the scoring data to obtain the corresponding risk assessment result; The output module 307 is used to output the risk assessment results.

[0126] In some optional implementations of this embodiment, the construction module 302 includes: The preprocessing submodule is used to preprocess the target data to obtain the corresponding processed data; An extraction submodule is used to extract features from the processed data to obtain initial features related to risk assessment. The mining submodule is used to perform feature mining processing on the initial features based on a preset mining strategy to obtain the corresponding generated features; The integration submodule is used to integrate the initial features and the generated features to obtain the corresponding integrated features; The first determining submodule is used to use the integrated features as the multidimensional feature data.

[0127] In some optional implementations of this embodiment, the mining submodule includes: The first processing unit is used to perform feature combination processing on the initial features based on a preset feature combination method to obtain the corresponding combined features; The second processing unit is used to perform feature derivation processing on the initial feature based on a preset feature derivation method to obtain the corresponding derived feature; An integration unit is used to integrate the combined features and the derived features to obtain corresponding processed features; A determining unit is used to use the processed features as the generated features.

[0128] In some optional implementations of this embodiment, the processing module 303 includes: The first acquisition submodule is used to acquire the preset standardization strategy and the preset weight allocation strategy; The processing submodule is used to perform standardization processing on the multidimensional feature data based on the standardization strategy to obtain the corresponding first feature data. The allocation submodule is used to perform weight allocation processing on the first feature data based on the weight allocation strategy to obtain the corresponding weight data. The fusion submodule is used to fuse the first feature data and the weight data based on a preset weighted fusion strategy to obtain the corresponding second feature data. The second determining submodule is used to use the second feature data as the fused feature.

[0129] In some optional implementations of this embodiment, the second mapping module 305 includes: The first submodule is used to call the preset scoring function; The calculation submodule is used to perform calculation processing on the potential risk embedding representation based on the scoring function to obtain the corresponding calculation result; The third determining submodule is used to use the calculation result as the scoring data.

[0130] In some optional implementations of this embodiment, the evaluation module 306 includes: The second submodule is used to invoke the preset risk level classification table; The query submodule is used to query the risk level classification table based on the scoring data in order to determine the target scoring range that matches the scoring data. The second acquisition submodule is used to obtain the target risk level corresponding to the target scoring interval from the risk level classification table; The fourth determination submodule is used to take the target risk level as the risk assessment result.

[0131] In some optional implementations of this embodiment, the artificial intelligence-based data evaluation device further includes: The collection module is used to collect feedback data related to risk assessment from preset data sources; The acquisition module is used to acquire a preset reward function; The calling module is used to invoke preset reinforcement learning algorithms; An optimization module is used to optimize the model parameters of the data evaluation model based on the reward function, using the reinforcement learning algorithm and the feedback data.

[0132] To address the aforementioned technical problems, embodiments of this application also provide a computer device. Please refer to [link / reference needed] for details. Figure 4 , Figure 4 This is a basic structural block diagram of the computer device in this embodiment.

[0133] The computer device 4 includes a memory 41, a processor 42, and a network interface 43 that are interconnected via a system bus. It should be noted that only the computer device 4 with components 41-43 is shown in the figure; however, it should be understood that it is not required to implement all the shown components, and more or fewer components can be implemented alternatively. Those skilled in the art will understand that the computer device described here is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0134] The computer device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device can interact with the user via a keyboard, mouse, remote control, touchpad, or voice control.

[0135] The memory 41 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 41 may be an internal storage unit of the computer device 4, such as the hard disk or memory of the computer device 4. In other embodiments, the memory 41 may also be an external storage device of the computer device 4, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 4. Of course, the memory 41 may also include both the internal storage unit and its external storage device of the computer device 4. In this embodiment, the memory 41 is typically used to store the operating system and various application software installed on the computer device 4, such as computer-readable instructions for data evaluation methods based on artificial intelligence. In addition, the memory 41 can also be used to temporarily store various types of data that have been output or will be output.

[0136] In some embodiments, the processor 42 may be a central processing unit (CPU), controller, microcontroller, microprocessor, or other data processing chip. The processor 42 is typically used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is used to execute computer-readable instructions stored in the memory 41 or to process data, for example, to execute computer-readable instructions for the artificial intelligence-based data evaluation method.

[0137] The network interface 43 may include a wireless network interface or a wired network interface, which is typically used to establish communication connections between the computer device 4 and other electronic devices.

[0138] This application also provides another embodiment, namely, providing a computer-readable storage medium storing computer-readable instructions that can be executed by at least one processor to cause the at least one processor to perform the steps of the artificial intelligence-based data evaluation method described above.

[0139] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0140] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application can be implemented in many different forms; rather, the purpose of providing these embodiments is to provide a more thorough and comprehensive understanding of the disclosure of this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application.

Claims

1. A data evaluation method based on artificial intelligence, characterized in that, Includes the following steps: The system collects target data from users across multiple preset dimensions; wherein the target data includes at least financial structured data, behavioral data, and social data. The target data is subjected to feature construction processing to obtain corresponding multidimensional feature data; The multidimensional feature data is standardized and fused to obtain the corresponding fused features; Based on a preset data evaluation model, the fused features are subjected to feature mapping processing to obtain a potential risk embedding representation corresponding to the fused features; The potential risk embedding representation is subjected to a scoring mapping process based on a preset scoring strategy to obtain the corresponding scoring data; Based on the rating data, a risk assessment is performed on the user to obtain the corresponding risk assessment result; The risk assessment results are then processed for output.

2. The data evaluation method based on artificial intelligence according to claim 1, characterized in that, The step of performing feature construction processing on the target data to obtain corresponding multidimensional feature data specifically includes: The target data is preprocessed to obtain the corresponding processed data; Feature extraction is performed on the processed data to obtain initial features related to risk assessment; Based on a preset mining strategy, the initial features are subjected to feature mining processing to obtain the corresponding generated features; The initial features and the generated features are integrated to obtain the corresponding integrated features; The integrated features are used as the multidimensional feature data.

3. The data evaluation method based on artificial intelligence according to claim 2, characterized in that, The step of performing feature mining processing on the initial features based on a preset mining strategy to obtain the corresponding generated features specifically includes: The initial features are combined using a preset feature combination method to obtain the corresponding combined features. The initial features are subjected to feature derivation processing based on a preset feature derivation method to obtain corresponding derived features; The combined features and the derived features are integrated to obtain the corresponding processed features; The processing features are used as the generated features.

4. The data evaluation method based on artificial intelligence according to claim 1, characterized in that, The step of standardizing and fusing the multidimensional feature data to obtain the corresponding fused features specifically includes: Obtain the preset standardization strategy and the preset weight allocation strategy; The multidimensional feature data is standardized based on the standardization strategy to obtain the corresponding first feature data. The first feature data is weighted according to the weight allocation strategy to obtain the corresponding weight data. The first feature data and the weight data are fused based on a preset weighted fusion strategy to obtain the corresponding second feature data. The second feature data is used as the fused feature.

5. The data evaluation method based on artificial intelligence according to claim 1, characterized in that, The step of performing a scoring mapping process on the potential risk embedding representation based on a preset scoring strategy to obtain the corresponding scoring data specifically includes: Call the preset scoring function; The potential risk embedding representation is calculated based on the scoring function to obtain the corresponding calculation result; The calculation results are used as the scoring data.

6. The data evaluation method based on artificial intelligence according to claim 1, characterized in that, The step of performing risk assessment on the user based on the scoring data to obtain the corresponding risk assessment result specifically includes: Call the preset risk level classification table; Based on the scoring data, the risk level classification table is queried to determine the target scoring range that matches the scoring data. Obtain the target risk level corresponding to the target scoring interval from the risk level classification table; The target risk level is used as the risk assessment result.

7. The data evaluation method based on artificial intelligence according to claim 1, characterized in that, Following the step of outputting the risk assessment results, the method further includes: Collect feedback data related to risk assessment from pre-defined data sources; Obtain the preset reward function; Invoke the preset reinforcement learning algorithm; Based on the reward function, the reinforcement learning algorithm and the feedback data are used to optimize the model parameters of the data evaluation model.

8. A data evaluation device based on artificial intelligence, characterized in that, include: The data collection module is used to collect target data from users from multiple preset dimensions; wherein, the target data includes at least financial structured data, behavioral data, and social data. The construction module is used to perform feature construction processing on the target data to obtain corresponding multidimensional feature data; The processing module is used to standardize and fuse the multidimensional feature data to obtain the corresponding fused features; The first mapping module is used to perform feature mapping processing on the fusion feature based on a preset data evaluation model to obtain a potential risk embedding representation corresponding to the fusion feature; The second mapping module is used to perform scoring mapping processing on the potential risk embedding representation based on a preset scoring strategy to obtain the corresponding scoring data. The assessment module is used to perform risk assessment on the user based on the scoring data and obtain the corresponding risk assessment result. The output module is used to process the risk assessment results.

9. A computer device, characterized in that, It includes a memory and a processor, wherein the memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement the steps of the data evaluation method based on artificial intelligence as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the data evaluation method based on artificial intelligence as described in any one of claims 1 to 7.

Images