Credit risk detection methods and devices, electronic equipment, and software products

Abstract

This invention discloses a credit risk detection method, device, electronic equipment, and program product, relating to the field of financial technology. The method includes: responding to credit business requests, collecting structured financial data and unstructured news text data from customers; filtering the preprocessed financial data to output a set of financial indicators; performing topic characterization on the preprocessed news text data to output the text topic probability distribution data for each news document; fusing the filtered set of financial indicators with the topic-characterized text topic probability distribution data to generate a detection information set; inputting the detection information set into a credit risk detection model; and outputting a risk assessment report and credit business approval recommendations. This invention solves the technical problem in related technologies where financial institutions rely on a single data source for risk detection of customer credit business, resulting in incomplete information collection, which affects credit risk detection and can easily lead to financial institutions' financial losses.

Description

Technical Field

[0001] This invention relates to the field of financial technology or other related fields, and more specifically, to a credit risk detection method and apparatus, electronic device, and program product. Background Technology

[0002] In today's financial markets, lending is one of the core services of financial institutions, and its risk control capabilities directly affect the sound operation of these institutions and the healthy development of the financial market. Accurate credit risk assessment is a necessary step before financial institutions issue loans, involving consideration of various factors such as the borrower's financial situation, credit history, and market environment. Traditional credit risk assessment mainly relies on manual analysis and structured data from financial statements. Existing credit risk detection models typically depend on a single data source, using structured data such as corporate financial statements as input, lacking the use of unstructured data such as text involved in traditional credit risk assessment. This leads to incomplete and inaccurate information, ultimately affecting the effectiveness of risk detection.

[0003] In related technologies, financial institutions still make relatively low use of unstructured data in credit risk assessment, mainly due to the following technical problems: 1. Insufficient information utilization: Traditional credit risk assessment models are mainly based on numerical information in financial statements, ignoring the textual descriptions and detailed explanations in the notes to the financial statements. These notes may contain important clues about the company's operating conditions and financial health. 2. Limited data processing capabilities: Unstructured data, such as news texts, is usually large in volume and complex. Existing text analysis methods are often inefficient in processing this type of data and perform poorly in processing short texts and sparse information.

[0004] There is currently no effective solution to the above problems. Summary of the Invention

[0005] This invention provides a credit risk detection method, device, electronic equipment, and program product to at least solve the technical problem in the related art where financial institutions rely on a single data source and collect incomplete information when conducting risk detection on customer credit business, which affects credit risk detection and can easily cause financial institutions to suffer property losses.

[0006] To achieve the above objectives, according to one aspect of this application, a credit risk detection method is provided, comprising: upon receiving a credit business request, collecting structured financial data and unstructured news text data associated with the target customer in the credit business request, and preprocessing the structured financial data and the news text data; filtering the preprocessed financial data to output a set of financial indicators for credit risk detection, and using a Dirichlet multinomial mixture model to perform topic representation on the preprocessed news text data, outputting text topic probability distribution data for each news document; fusing the filtered set of financial indicators with the topic-represented text topic probability distribution data to generate a detection information set containing structured financial information and unstructured text information; and inputting the detection information set into a credit risk detection model to output a risk assessment report and credit business approval recommendations for the current credit business initiated by the target customer.

[0007] Optionally, the step of collecting structured financial data and unstructured news text data associated with the target customer in the credit business request includes: collecting financial reports involving the target customer, and filtering out the balance sheet, income statement, and cash flow statement from the financial reports to obtain a financial data table; extracting multiple financial indicators and the corresponding financial data from the financial data table to obtain the structured financial data, wherein the multiple financial indicators include at least one of the following: current ratio, quick ratio, total asset growth rate, gross profit margin, net profit margin, period expense ratio, cash cost ratio, and financial expense cash ratio; collecting news reports associated with the target customer from financial news websites and financial engine databases, wherein the types of the news reports include at least one of the following: corporate profit announcements, goodwill warnings, penalties for violations, and industry news; and extracting text data from the news reports to obtain the unstructured news text data.

[0008] Optionally, the preprocessing steps for the structured financial data and the news text data include: cleaning and handling missing values ​​in the structured financial data, wherein the data cleaning includes deleting duplicate values ​​and deleting outliers, and the missing value handling includes filling in missing data; cleaning the news text data, wherein the cleaning includes removing stop words, removing punctuation marks, removing numbers and special characters; and segmenting the cleaned news text to obtain a set of text words and a set of text sentences.

[0009] Optionally, the step of filtering the preprocessed financial data and outputting a set of financial indicators for credit risk detection includes: calculating the IV value of each financial indicator in the structured financial data, filtering out financial indicators with IV values ​​greater than or equal to a preset threshold to obtain financial indicators associated with credit risk detection; discretizing the continuous financial data in the structured financial data, using a decision tree binning strategy to convert continuous variables into discrete variables; and using a chi-square test strategy to perform feature filtering on the discretized financial data, removing features with a correlation degree lower than a preset correlation degree threshold for credit risk detection, and outputting the set of financial indicators for credit risk detection.

[0010] Optionally, the step of using a Dirichlet multinomial mixture model to perform topic representation on the preprocessed news text data includes: using a pre-trained Word2Vec model to perform word vectorization on the preprocessed news text to construct a word vector matrix; and using the Dirichlet multinomial mixture model to extract topics from the word vector matrix to obtain the probability distribution of each news document on each topic.

[0011] Optionally, the Dirichlet multinomial mixture model is trained as follows: initializing model parameters, including the number of topics and the hyperparameters of the Dirichlet distribution; acquiring preprocessed historical news text data and randomly assigning each historical news document to a topic as the initial state for model training; a topic probability calculation step, iteratively calculating the probability of each historical news document belonging to each topic, wherein, when calculating the probability of each historical news document belonging to each topic, a Gibbs sampling algorithm is used for probability estimation based on the topic features of each document and the model parameters; a probability redistribution step, redistributing each historical news document to the target topic according to the calculated probability; repeating the iterative calculation of topic probability and probability redistribution steps until the rate of change of document topic allocation is lower than a preset threshold or the maximum number of iterations is reached, completing the topic model convergence, and obtaining the trained Dirichlet multinomial mixture model.

[0012] Optionally, the step of fusing the filtered set of financial indicators with the text topic probability distribution data after topic representation to generate a detection information set containing structured financial information and unstructured text information includes: merging the set of financial indicators of the target customer and the text topic probability distribution data into the detection information set based on the data table connection strategy, wherein, in the detection information set, independent columns are assigned to each financial indicator in the set of financial indicators and each text subject in the text topic probability distribution data.

[0013] Optionally, the credit risk detection model is trained as follows: A historical detection information set containing historical structured financial information and historical unstructured text information is obtained, and the historical detection information set is divided into a training set and a test set; an initial risk detection model is constructed using the Adaboost ensemble learning algorithm; a decision tree is selected as a weak classifier, and each training sample in the training set is assigned equal weight; the classifier weight distribution adjustment step is as follows: following the iterative process of the Adaboost algorithm, a weak classifier is selected in each round, and risk detection training is performed based on the weight distribution of the training samples, wherein the weights of the weak classifier... The weights are adjusted based on the classification accuracy. The sample weight update step involves updating the weights of the training samples after each round of weak classifier training. When updating the weights of the training samples, the weights of misclassified training samples are increased, and the weights of correctly classified training samples are decreased. This process of adjusting the classifier weight distribution and updating the sample weights is repeated until a preset number of weak classifiers is reached, resulting in a strong classifier composed of multiple weak classifiers. The credit risk detection model is then evaluated using the test set. If the performance evaluation results meet the evaluation accuracy, the training is confirmed to be complete, and the credit risk detection model is obtained.

[0014] According to another aspect of the present invention, a credit risk detection device is also provided, comprising: a data collection unit, configured to, upon receiving a credit business request, collect structured financial data and unstructured news text data associated with a target customer in the credit business request, and preprocess the structured financial data and the news text data; a data filtering unit, configured to filter the preprocessed financial data, output a set of financial indicators for credit risk detection, and employ a Dirichlet multinomial mixture model to perform topic representation on the preprocessed news text data, outputting text topic probability distribution data for each news document; a data fusion unit, configured to fuse the filtered set of financial indicators with the topic-represented text topic probability distribution data to generate a detection information set containing structured financial information and unstructured text information; and a risk assessment unit, configured to input the detection information set into a credit risk detection model, and output a risk assessment report and credit business approval recommendations for the current credit business initiated by the target customer.

[0015] Optionally, the data collection unit includes: a financial report collection module, used to collect financial reports related to the target customer, and filter out the balance sheet, income statement and cash flow statement from the financial reports to obtain a financial data table; a financial data extraction module, used to extract multiple financial indicators and the financial data corresponding to each financial indicator from the financial data table to obtain the structured financial data, wherein the multiple financial indicators include at least one of the following: current ratio, quick ratio, total asset growth rate, gross profit margin, net profit margin, period expense ratio, cash cost ratio and financial expense cash ratio; a news report collection module, used to collect news reports related to the target customer from financial news websites and financial engine databases, wherein the types of news reports include at least one of the following: corporate profit announcements, goodwill warnings, violation penalties and industry news; and a news text extraction module, used to extract text data from the news reports to obtain unstructured news text data.

[0016] Optionally, the data collection unit includes: a data preprocessing module for cleaning and handling missing values ​​in the structured financial data, wherein the data cleaning includes deleting duplicate values ​​and deleting outliers, and the missing value handling includes filling in missing data; cleaning the news text data, wherein the cleaning includes removing stop words, removing punctuation marks, and removing numbers and special characters; and a word segmentation module for segmenting the cleaned news text to obtain a set of text words and a set of text sentences.

[0017] Optionally, the data filtering unit includes: an IV value calculation module, used to calculate the IV value of each financial indicator in the structured financial data, and filter out financial indicators with IV values ​​greater than or equal to a preset threshold to obtain financial indicators associated with credit risk detection; a discretization module, used to discretize the continuous financial data in the structured financial data, and use a decision tree binning strategy to convert continuous variables into discrete variables; and a feature filtering module, used to use a chi-square test strategy to filter features of the discretized financial data, remove features with a correlation degree lower than a preset correlation degree threshold for credit risk detection, and output the set of financial indicators used for credit risk detection.

[0018] Optionally, the data filtering unit further includes: a word vectorization processing module, used to perform word vectorization processing on the pre-processed news text using a pre-trained Word2Vec model to construct a word vector matrix; and a topic extraction module, used to extract topics from the word vector matrix using the Dirichlet multinomial mixture model to obtain the probability distribution of each news document on each topic.

[0019] Optionally, the Dirichlet multinomial mixture model is trained as follows: initializing model parameters, including the number of topics and the hyperparameters of the Dirichlet distribution; acquiring preprocessed historical news text data and randomly assigning each historical news document to a topic as the initial state for model training; a topic probability calculation step, iteratively calculating the probability of each historical news document belonging to each topic, wherein, when calculating the probability of each historical news document belonging to each topic, a Gibbs sampling algorithm is used for probability estimation based on the topic features of each document and the model parameters; a probability redistribution step, redistributing each historical news document to the target topic according to the calculated probability; repeating the iterative calculation of topic probability and probability redistribution steps until the rate of change of document topic allocation is lower than a preset threshold or the maximum number of iterations is reached, completing the topic model convergence, and obtaining the trained Dirichlet multinomial mixture model.

[0020] Optionally, the data fusion unit includes: a data merging module, used to merge the set of financial indicators of the target customer and the text topic probability distribution data into the detection information set based on the data table connection strategy, wherein, in the detection information set, independent columns are assigned to each financial indicator in the set of financial indicators and each text subject in the text topic probability distribution data.

[0021] Optionally, the credit risk detection model is trained as follows: A historical detection information set containing historical structured financial information and historical unstructured text information is obtained, and the historical detection information set is divided into a training set and a test set; an initial risk detection model is constructed using the Adaboost ensemble learning algorithm; a decision tree is selected as a weak classifier, and each training sample in the training set is assigned equal weight; the classifier weight distribution adjustment step is as follows: following the iterative process of the Adaboost algorithm, a weak classifier is selected in each round, and risk detection training is performed based on the weight distribution of the training samples, wherein the weights of the weak classifier... The weights are adjusted based on the classification accuracy. The sample weight update step involves updating the weights of the training samples after each round of weak classifier training. When updating the weights of the training samples, the weights of misclassified training samples are increased, and the weights of correctly classified training samples are decreased. This process is repeated until a preset number of weak classifiers is reached, resulting in a strong classifier composed of multiple weak classifiers. The credit risk detection model is then evaluated using the test set. If the performance evaluation results meet the evaluation accuracy, the training is considered complete, and the credit risk detection model is obtained.

[0022] According to another aspect of the present invention, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the credit risk detection method described above.

[0023] According to another aspect of the present invention, an electronic device is also provided, including one or more processors and a memory, the memory being 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 cause the one or more processors to implement the credit risk detection method described above.

[0024] According to another aspect of the present invention, a computer program product is also provided, including a computer program that, when executed by a processor, implements the steps of the credit risk detection method described in any one of the above embodiments.

[0025] In this disclosure, upon receiving a credit business request, structured financial data and unstructured news text data associated with the target customer in the credit business request can be collected. The structured financial data and news text data are preprocessed, the preprocessed financial data is filtered, and a set of financial indicators for credit risk detection is output. A Dirichlet multinomial mixture model is used to perform topic representation on the preprocessed news text data, and the text topic probability distribution data of each news document is output. The filtered set of financial indicators is fused with the topic representation text topic probability distribution data to generate a detection information set containing structured financial information and unstructured text information. The detection information set is input into the credit risk detection model, and a risk assessment report and credit business approval recommendations for the current credit business initiated by the target customer are output.

[0026] Based on the aforementioned publicly available information, unstructured news text data can be introduced on top of traditional structured financial data to analyze customer credit risk from multiple information dimensions. This enables more comprehensive and accurate detection of corporate credit risk. Furthermore, in representing unstructured text data, the advantages of pre-trained models and topic models are integrated, while also considering the vectorized representation of text and topic probability distribution. This enriches textual information, improves the accuracy of credit risk detection, enhances the risk control capabilities and market competitiveness of financial institutions, reduces their financial losses, and addresses the technical problem in related technologies where financial institutions rely on a single data source for risk detection in customer credit business, resulting in incomplete information collection, which affects credit risk detection and can easily lead to financial losses for financial institutions. Attached Figure Description

[0027] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0028] Figure 1 A hardware structure block diagram of a computer terminal (or mobile device) for implementing a credit risk detection method is shown.

[0029] Figure 2 This is a flowchart of an optional credit risk detection method according to an embodiment of the present invention;

[0030] Figure 3 This is a flowchart of a credit risk detection model based on multidimensional information according to an embodiment of the present invention;

[0031] Figure 4 This is a schematic diagram of a probabilistic graphical model of a Dirichlet polynomial mixture model according to an embodiment of the present invention;

[0032] Figure 5 This is a schematic diagram of an optional credit risk detection device according to an embodiment of the present invention;

[0033] Figure 6 This is a structural block diagram of an electronic device according to an embodiment of this application. Detailed Implementation

[0034] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0035] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0036] To facilitate understanding of the present invention by those skilled in the art, some terms or nouns involved in the various embodiments of the present invention are explained below:

[0037] The word2vec model is a deep learning model that converts words in text into vector representations.

[0038] IV (Information Value) is a statistical measure used to quantify the strength of the relationship between a feature variable and a target variable, primarily used for feature selection. In the field of credit risk, it measures the predictive power of a financial indicator on credit risk; a higher V value indicates a stronger correlation between the feature variable and the target variable, and a greater contribution to the model.

[0039] The chi-square test is a statistical test used to determine whether a significant association exists between two categorical variables. In feature selection, the chi-square test can be used to examine whether a feature variable is associated with a target variable (such as "default" or "no default").

[0040] The Dirichlet Multinomial Mixture Model (DMM) is a nonparametric Bayesian model for text topic modeling. Unlike traditional LDA models, DMMs assume that terms within a document follow a multinomial distribution, while the prior distribution of document topics follows a Dirichlet distribution. In this invention, during model training, the model parameters are iteratively optimized using methods such as Gibbs sampling or variational inference to obtain the document's topic distribution.

[0041] AdaBoost, or Adaptive Boosting, is an ensemble learning algorithm for classification that combines multiple weak classifiers into a single strong classifier. In each training round, AdaBoost adjusts the weights of samples based on the error rate of the previous classifier. Samples that were misclassified are given higher weights in the next round, forcing subsequent classifiers to focus more on these difficult samples. In this way, a series of weak classifiers gradually adjust and focus on the most difficult samples to classify. Finally, the outputs of these classifiers are combined to form a more powerful model capable of more accurate classification predictions. In credit risk detection, AdaBoost effectively improves the model's sensitivity and accuracy in identifying high-risk customers.

[0042] It should be noted that the credit risk detection method and apparatus disclosed herein can be used in the fintech field to detect risks in customer credit business based on multi-dimensional customer association data, and can also be used in any field other than fintech to detect risks in customer credit business based on multi-dimensional customer association data. This disclosure does not limit the application areas of the credit risk detection method and apparatus.

[0043] It should be noted that the information collected in this public disclosure (including but not limited to customer identity information, customer financial account information, customer news reports, customer personal information, etc.) and data (including but not limited to data used for analysis, stored data, and displayed data) are information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of this data all comply with the relevant laws, regulations, and standards of the relevant regions, and necessary confidentiality measures have been taken. This does not violate public order and good morals, and corresponding operation entry points are provided for users to choose to authorize or refuse. For example, this system has interfaces with relevant users or institutions. Before obtaining relevant information, a request to obtain the information needs to be sent to the aforementioned user or institution through the interface, and the relevant information is obtained only after receiving consent from the aforementioned user or institution.

[0044] It should be noted that in this disclosure, customer information is collected and analyzed, and users are provided with corresponding operation entry points to choose whether to agree to or reject the automated decision results; if the user chooses to reject, the process will proceed to the expert decision-making process.

[0045] The following embodiments of the present invention can be applied to various credit risk detection systems / applications / devices. Based on traditional structured financial data, this invention focuses on the notes to the client's financial statements and introduces unstructured news text data to analyze corporate credit risk from multiple information dimensions, enabling more comprehensive and accurate corporate credit risk detection compared to existing technologies. Furthermore, in representing unstructured text data, it integrates the advantages of pre-trained models and topic models, while considering the vectorized representation of text and topic probability distribution, thus enriching the text information.

[0046] The present invention will now be described in detail with reference to various embodiments.

[0047] Example 1

[0048] According to an embodiment of the present invention, an embodiment of a credit risk detection method is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0049] The credit risk detection method embodiment provided in Embodiment 1 of this application can be executed on a mobile terminal, computer terminal, or similar computing device. Figure 1 A hardware block diagram of a computer terminal (or mobile device) for implementing a credit risk detection method is shown. Figure 1 As shown, computer terminal 10 (or mobile device) may include one or more ( Figure 1 (Illustrated as 102a, 102b, ..., 102n) Processor 102 (processor 102 may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.), memory 104 for storing data, and transmission device 106 for communication functions. In addition, it may include: a display, input / output interface (I / O interface), Universal Serial Bus (USB) port (which may be included as one of the ports of a BUS bus), network interface, power supply, and / or camera. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, computer terminal 10 may also include... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0050] It should be noted that the aforementioned one or more processors 102 and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be integrated, in whole or in part, into any other element within the computer terminal 10 (or mobile device). As involved in the embodiments of this application, the data processing circuits serve as a processor control mechanism (e.g., selection of a variable resistor termination path connected to an interface).

[0051] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the credit risk detection method in this embodiment. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, thereby realizing the aforementioned credit risk detection method. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the computer terminal 10 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0052] The transmission device 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the computer terminal 10. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 may be a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0053] The display can be, for example, a touchscreen liquid crystal display (LCD), which allows the user to interact with the user interface of the computer terminal 10 (or mobile device).

[0054] It should be noted that the credit risk detection method provided in this invention can solve the challenges faced by financial institutions in assessing the credit risk of customers (such as enterprises and factories) before issuing loans. It is applicable to various financial institutions in scenarios such as enterprise credit assessment, risk management, and investment decision-making. By integrating structured financial data and unstructured text data, it improves the accuracy, timeliness, and comprehensiveness of credit risk assessment, thereby effectively enhancing the risk control capabilities and market competitiveness of financial institutions. Specific application scenarios may include: 1. Enterprise loan approval: When an enterprise applies for a loan from a bank or other financial institution, the credit risk detection model of this application can quickly and accurately assess the applicant enterprise's financial situation and potential risks. By integrating structured data from enterprise financial statements with unstructured data such as news reports, the model can provide a more comprehensive risk assessment, helping financial institutions make more informed lending decisions. 2. Credit portfolio management: For loans already issued, by continuously monitoring changes in the enterprise's financial indicators and related news reports, the credit risk detection model of this application can periodically assess the overall risk level of the credit portfolio. This dynamic monitoring capability is crucial for timely identification and handling of potential credit risks and for adjusting credit strategies and portfolio structures. 3. Risk Early Warning System: Combining real-time news feeds and updated financial data, the credit risk detection model in this application can serve as part of a risk early warning system. It can automatically identify borrowers with deteriorating financial conditions or negative news reports, providing early warning signals to help financial institutions take proactive measures to avoid or reduce potential losses. 4. Credit Rating Updates: Credit rating agencies or financial institutions can use the model in this application to update corporate credit ratings. Compared to traditional rating methods, this credit risk detection model considers more dimensions of information, such as the company's operating environment, industry trends, and unforeseen events, ensuring the accuracy and timeliness of the rating. 5. Risk Preference Management: Financial institutions can adjust parameters in the model according to their risk preferences, such as the screening thresholds for financial indicators and the parameters of thematic models, to adapt to different credit policies and market environments.

[0055] Under the aforementioned operating environment, this application provides the following: Figure 2 The credit risk detection method shown. Figure 2 This is a flowchart of an optional credit risk detection method according to an embodiment of the present invention, such as... Figure 2 As shown, the method includes the following steps:

[0056] Step S201: After receiving a credit business request, collect structured financial data and unstructured news text data associated with the target customer in the credit business request, and preprocess the structured financial data and news text data.

[0057] Upon receiving a credit request, the system parses the request to obtain the target customer's identifier, associated financial accounts, and the specific credit product and term. The next step is to collect structured financial data and unstructured news text data associated with the target customer. Structured data primarily comes from the customer's financial reports, including but not limited to balance sheets, income statements, and cash flow statements. This data is obtained through the bank's internal systems or third-party credit assessment platforms and reflects the company's financial health. Unstructured data mainly comes from financial news websites, industry reports, social media, and notes to company financial statements. This type of data provides multi-dimensional information on the company's operating status, market reputation, industry dynamics, and unforeseen events, playing a crucial supplementary role in credit risk assessment.

[0058] Optionally, the step of collecting structured financial data and unstructured news text data associated with target customers in credit business requests includes: collecting financial reports involving target customers, and filtering out balance sheets, income statements, and cash flow statements from the financial reports to obtain financial data tables; extracting multiple financial indicators and the corresponding financial data for each financial indicator from the financial data tables to obtain structured financial data, wherein the multiple financial indicators include at least one of the following: current ratio, quick ratio, total asset growth rate, gross profit margin, net profit margin, period expense ratio, cash cost ratio, and financial expense cash ratio; collecting news reports associated with target customers from financial news websites and financial search engine databases, wherein the types of news reports include at least one of the following: corporate profit announcements, goodwill warnings, penalties for violations, and industry news; and extracting text data from the news reports to obtain unstructured news text data.

[0059] This embodiment collects data including structured financial data and unstructured news text data. Here, we describe the collection and screening of structured financial data. First, financial institutions collect all financial reports related to target clients through various legal channels, such as financial information submitted by enterprises, publicly available financial reports, or data from third-party credit rating agencies. These reports form the basis for assessing the financial health and credit risk of enterprises. From the collected financial reports, the balance sheet, income statement, and cash flow statement are the key screening items. These three statements are the core reflection of an enterprise's financial condition, providing detailed information on assets, liabilities, revenue, profit, and cash flow, and are indispensable data sources for credit risk assessment. Several key indicators are extracted from the financial statements, including but not limited to the current ratio, quick ratio, total asset growth rate, gross profit margin, net profit margin, period expense ratio, cash cost ratio, and financial expense cash ratio. These indicators are directly related to an enterprise's solvency, profitability, cost control, and cash flow status, and have a direct impact on credit risk assessment. The extracted financial indicators and their corresponding values ​​are converted into structured data tables to facilitate subsequent model processing and analysis. This step ensures data standardization and format consistency.

[0060] For the collection and preprocessing of unstructured news text data, news reports relevant to the target customers can be collected through channels such as financial news websites, financial database engines, or social media. These reports cover a wide range of information, including corporate profits, goodwill changes, legal proceedings, penalties for violations, and industry trends, providing important clues beyond financial statements. Information directly related to credit risk is filtered from these news reports, including corporate profit announcements, goodwill warnings, penalties for violations, and industry news. This type of information reveals the real-time dynamics of a company's operations, helping to adjust risk assessments promptly. Text data is extracted from the filtered news reports, removing irrelevant formatting, advertising information, stop words, and punctuation, retaining only text content with potential value for risk assessment. Text cleaning is a necessary step before text analysis, improving data quality and analytical efficiency. The extracted text data is preprocessed, including word segmentation, part-of-speech tagging, and topic modeling, converting unstructured news text into quantifiable features such as word vectors and topic probability distributions, providing rich information input for subsequent credit risk assessment models.

[0061] After obtaining structured financial data and unstructured news text data, both types of data need to be preprocessed. Optionally, the preprocessing steps for structured financial data and news text data include: data cleaning and missing value handling for structured financial data, where data cleaning includes: removing duplicate values ​​and outliers, and missing value handling includes: filling in missing data; cleaning the news text data, where cleaning includes removing stop words, removing punctuation marks, and removing numbers and special characters; and performing word segmentation on the cleaned news text to obtain a set of text words and a set of text sentences.

[0062] In the preprocessing stage, structured financial data needs to be merged, cleaned, and transformed, including filling in missing values, deleting outliers, and standardizing monetary variables to ensure data consistency and reliability. For unstructured text data, text cleaning is performed first to remove irrelevant symbols, numbers, and stop words. Then, word segmentation tools are used to divide the text into words or phrases to prepare for subsequent vectorization and topic extraction.

[0063] Specifically, the data cleaning described above can include: removing duplicate values ​​and outliers. Removing duplicate values ​​refers to the situation where the same data appears repeatedly in different records within the collected financial data. This could be due to human error during data entry or inconsistencies in data sources. Removing duplicate values ​​ensures that each piece of financial information is unique, avoiding bias in model training. Removing outliers refers to the occasional significant deviations from the normal range in the financial data. These values ​​may be due to data entry errors, unusual financial activities, or special circumstances (such as changes in accounting policies). Identifying and removing these outliers is crucial for maintaining the accuracy and reliability of the dataset. Optionally, the outlier detection methods used in this embodiment include using box plots, Z-scores, and IQR (interquartile range) to determine the degree of numerical anomalies.

[0064] Furthermore, in structured financial data, certain fields may be missing due to various reasons (such as reporting delays or incomplete data). There are several methods for handling missing values, including deleting records containing missing values, filling with the mean or median, or using regression models for prediction. In this application, considering the continuity and importance of financial data, we may choose to use predictive models or fill with data based on historical trends to maintain data integrity and information content.

[0065] Further, it is necessary to explain the preprocessing of news text data. Among them, the cleaning process includes: removing stop words. Stop words refer to words that appear frequently in the text but contribute little to the text theme, such as "of", "is", "and", etc. Removing stop words can reduce the redundancy of text data and improve the efficiency and accuracy of subsequent analysis. Removing punctuation marks, numbers, and special characters: Punctuation marks, numbers, and special characters usually do not have a direct impact on the text theme or sentiment analysis, but instead increase the complexity of the data. Removing these elements helps to simplify the text and make it more suitable for analysis by natural language processing algorithms.

[0066] After that, it is necessary to perform word segmentation on the news text to obtain a set of text words and a set of text clauses. Here, word segmentation is the process of splitting continuous text into independent words or phrases, which is a key step in text preprocessing. For news text data, word segmentation helps to convert the text into a form that can be understood by machines. By word segmentation, a set of text words is obtained, that is, the set of all words in the text.

[0067] Cleaning and handling missing values for structured financial data can eliminate inconsistencies and incompleteness in the data and improve the accuracy of the data. For news text data, removing stop words, punctuation marks, numbers, and special characters, as well as performing word segmentation, can convert unstructured information into structured features for further analysis.

[0068] Step S202: Screen the preprocessed financial data to output a set of financial indicators for credit risk detection, and use the Dirichlet polynomial mixture model to perform topic representation on the preprocessed news text data to output the text topic probability distribution data of each news document.

[0069] The preprocessed financial data is screened through IV value calculation and chi-square test to select financial indicators with a high contribution to credit risk prediction and construct a set of financial indicators. The IV value measures the discrimination ability of variables for credit risk prediction, and the chi-square test helps to screen variables significantly related to credit risk, thereby ensuring the efficiency and accuracy of the model. Optionally, the step of screening the preprocessed financial data to output a set of financial indicators for credit risk detection includes: calculating the IV value of each financial indicator in the structured financial data, screening out financial indicators with an IV value greater than or equal to a preset threshold to obtain financial indicators related to credit risk detection; discretizing the continuous financial data in the structured financial data, and using the decision tree binning strategy to convert continuous variables into discrete variables; using the chi-square test strategy to perform feature screening on the discretized financial data, removing features with a correlation degree lower than the preset correlation degree threshold for credit risk detection, and outputting a set of financial indicators for credit risk detection.

[0070] First, the IV value for each financial indicator in the structured financial data is calculated to identify those indicators that significantly contribute to credit risk detection. The formula for calculating the IV value is as follows:

[0071] [IV=\sum(GoodPercent-BadPercent)*Weight];

[0072] Here, (GoodPercent) and (BadPercent) represent the proportions of risk-free customers and credit-risk customers in this financial indicator segment, respectively, while (Weight) is the proportion of customers in this segment to the total number of customers. By calculating the IV value, the ability of each financial indicator to distinguish credit risk can be measured, and financial indicators with IV values ​​greater than or equal to a preset threshold can be selected as input for subsequent risk assessment models. Indicators with IV values ​​greater than 0.3 are considered to have strong predictive features, while those less than 0.02 are considered almost useless and should be excluded.

[0073] For example, in this embodiment, 176 financial indicators were extracted from the balance sheet, income statement, and cash flow statement, as well as the notes to the financial statements, and included in the initial feature pool. The data was then merged, cleaned, and transformed. Considering the correlation between the initial features, the contribution of variables to the target prediction was measured by calculating the variable IV value. Variables with higher IV values ​​were selected. For discrete variables, the IV values ​​and categorization results are shown in Table 1 below:

[0074] Table 1

[0075] Class IV value range Useless <0.02 Weak [0.02,0.1) Medium [0.1,0.3) Strong [0.3,0.5) Suspicious >＝0.5

[0076] After calculation, all fields with IV values ​​at the Useless level are removed. For continuous variables, they are first discretized using decision tree binning, then their IV values ​​are calculated, and features with IV values ​​less than 0.02 are removed. A chi-square test is then used for feature selection to assess the correlation between the discretized financial data and credit risk. Specifically, a significance level for the chi-square test is set (e.g., p-value less than 0.01), and the chi-square value for each discretized indicator is calculated. Features with a correlation to credit risk detection below a preset threshold are excluded to reduce model complexity, avoid overfitting, and ensure the model's predictive accuracy and efficiency.

[0077] Among the selected financial indicators, continuous financial data such as current ratio and gross profit margin are discretized using a decision tree binning strategy. Discretization is the process of converting continuous values ​​into categorical values, which simplifies model complexity and improves predictive performance. The decision tree binning strategy automatically finds the optimal split point for the data by constructing a decision tree, converting continuous variables into discrete variables. This facilitates subsequent statistical analysis and model training. The discretized financial data requires further feature selection to ensure that the features input to the model actually contribute to credit risk detection.

[0078] For unstructured news text data, a pre-trained word2vec model is used for word embedding to convert the text into vector representations. Simultaneously, this embodiment innovatively uses a Dirichlet Multinomial Mixture Model (DMM) for topic representation of the news text. The DMM model, tailored to the characteristics of short texts and sparse information, can more accurately extract news topics. During model training, each news document is assigned to different topics, obtaining text topic probability distribution data, which provides a rich source of unstructured information for subsequent risk detection.

[0079] Optionally, the step of using a Dirichlet multinomial mixture model to represent the topics of the preprocessed news text data includes: using a pre-trained Word2Vec model to perform word vectorization on the preprocessed news text to construct a word vector matrix; and using a Dirichlet multinomial mixture model to extract topics from the word vector matrix to obtain the probability distribution of each news document on each topic.

[0080] In this embodiment, a pre-trained Word2Vec model is used to perform word vectorization on the pre-processed news text, constructing a word vector matrix. Word2Vec is a neural network-based word embedding technique that converts words in text into fixed-size vectors. These vectors can capture the semantic relationships between words while reducing the dimensionality of the processed text data. In this implementation step, the pre-processed news text (with stop words, punctuation marks, etc. removed) is input into the Word2Vec model to generate a word vector matrix.

[0081] It should be noted that the Dirichlet Multinomial Mixture Model (DPMM) mentioned in this embodiment is trained as follows: Initialize model parameters, including the number of topics and hyperparameters of the Dirichlet distribution; acquire preprocessed historical news text data and randomly assign each historical news document to a topic as the initial state for model training; topic probability calculation step, iteratively calculate the probability of each historical news document belonging to each topic, wherein, when calculating the probability of each historical news document belonging to each topic, a Gibbs sampling algorithm is used for probability estimation based on the topic features of each document and the model parameters; probability redistribution step, redistribute each historical news document to the target topic according to the calculated probability; repeat the iterative calculation of topic probability step and probability redistribution step until the rate of change of document topic allocation is lower than a preset threshold or the maximum number of iterations is reached, the topic model converges, and the trained Dirichlet Multinomial Mixture Model is obtained.

[0082] During model training, in the initial stage, it's necessary to set the number of topics (K) (although DPMM is adaptive in terms of topic number, an initial value must be set at the start of training), the hyperparameters of the Dirichlet distribution (α) (controlling the concentration of topic distribution), and (β) (affecting the distribution of words within topics). Then, preprocessed historical news text data is collected. This data is used for model training to help the model learn different topics and their corresponding word distributions. Each historical news document is randomly assigned to a topic as the initial state for model training. During iteration, for each historical news document, based on the topic features of each document and the current model parameters, the Gibbs sampling algorithm is used to estimate the probability that the document belongs to each topic. Gibbs sampling is a sampling method used to handle complex probabilistic models. It can progressively update model parameters based on existing information until a stable state is reached. Based on the probability that each document belongs to each topic, each historical news document is reassigned to the most probable topic. This process optimizes document topic allocation, concentrating documents with the same topic as much as possible and distributing documents with different topics as much as possible. The iterative calculation of topic probabilities and probability redistribution steps are repeated until the rate of change in document topic allocation falls below a preset threshold (indicating the model has reached a stable state) or the maximum number of iterations is reached (to prevent infinite iteration). Model convergence marks the completion of the training process; at this point, the model parameters and topic allocation can effectively represent the topic structure of the news text. Through this process, DPMM can perform topic representation on news text data, obtaining the probability distribution of each news document across various topics. This probability distribution data, along with structured financial data, will be used to build a credit risk detection model, improving the model's predictive ability.

[0083] Step S203: The selected set of financial indicators is fused with the text topic probability distribution data after topic representation to generate a set of detection information containing structured financial information and unstructured text information.

[0084] Optionally, step S203 includes: merging the target customer's financial indicator set and text topic probability distribution data into a detection information set based on the data table connection strategy, wherein an independent column is assigned to each financial indicator in the financial indicator set and each text subject in the text topic probability distribution data in the detection information set.

[0085] During data fusion, a table join strategy is employed to merge the topic probability distributions of structured financial data with those of unstructured text data. This typically involves using the customer ID as the join key to ensure that each financial data record can be associated with its corresponding text data record. Each target customer has a unique identifier in both the financial indicator set and the text topic probability distribution data; this identifier is used for accurate data joining.

[0086] By employing a connection strategy, each financial indicator record of the target customer is merged with its corresponding text topic probability distribution record into a unified data table, which serves as the detection information set. Within this set, the probability distribution of each financial indicator and each text topic is assigned to an independent column, forming a multidimensional data table where each row represents a customer and each column contains information from different dimensions.

[0087] It's important to note that during the merging process, special attention was paid to the independent column allocation for each financial indicator and text topic probability distribution to ensure a clear and distinguishable data structure. For financial indicators, each column will contain a value for a specific indicator (such as current ratio, quick ratio, etc.), reflecting the customer's financial health. For text topics, each column will contain a probability value for a specific topic, quantifying the degree of relevance between the news text and that topic. To ensure that the financial indicators and text topic probability distribution data can work effectively together in the detection information set, data standardization may be necessary. This includes converting financial numerical indicators into a form comparable to the probability distribution data, or adjusting the range of the text topic probability distribution to match the magnitude of the financial data. Standardization ensures that different types of features are treated equally during model training, preventing differences in numerical magnitude from affecting the model's decisions. The final detection information set is a dataset that integrates structured financial information and unstructured text information, providing a comprehensive analytical foundation for the credit risk detection model. Through this multi-dimensional data integration, the model will be able to conduct credit risk assessments based on richer and more diverse information sources, thereby improving the accuracy and reliability of detection.

[0088] Step S204: Input the detection information set into the credit risk detection model, and output a risk assessment report and credit approval suggestions for the current credit business initiated by the target customer.

[0089] The detected information set is input into the credit risk detection model constructed in this embodiment. The model is based on the Adaboost ensemble learning algorithm, which, through the collaboration of multiple weak classifiers, forms a strong classifier that is highly sensitive to credit risk. After training, for a current credit transaction initiated by a target customer, the model outputs a risk assessment report containing information such as the customer's default probability, credit rating, and potential risk factors. Based on the risk assessment report, the model also provides approval suggestions for the credit transaction, such as approving the loan, requiring additional collateral, or rejecting the loan, providing a scientific basis for credit decisions.

[0090] Through the above steps, after receiving a credit business request, structured financial data and unstructured news text data associated with the target customer in the credit business request can be collected. The structured financial data and news text data are preprocessed, the preprocessed financial data is filtered, and a set of financial indicators for credit risk detection is output. The preprocessed news text data is then represented by a Dirichlet multinomial mixture model, and the text topic probability distribution data of each news document is output. The filtered set of financial indicators is fused with the text topic probability distribution data after topic representation to generate a detection information set containing structured financial information and unstructured text information. The detection information set is then input into the credit risk detection model, and a risk assessment report and credit approval recommendations for the current credit business initiated by the target customer are output. In this embodiment, unstructured news text data can be introduced on the basis of traditional structured financial data to analyze customer credit risk from multiple information dimensions. This enables more comprehensive and accurate detection of corporate credit risk. At the same time, in the representation of unstructured text data, the advantages of pre-trained models and topic models are combined, while considering the vectorized representation of text and topic probability distribution, enriching text information, improving the accuracy of credit risk detection, enhancing the risk control capabilities and market competitiveness of financial institutions, reducing the property losses of financial institutions, and solving the technical problem in related technologies where financial institutions rely on a relatively single data source when conducting risk detection of customer credit business, resulting in incomplete information collection, affecting credit risk detection, and easily causing property losses for financial institutions.

[0091] Optionally, the credit risk detection model is trained as follows: A historical detection information set containing historical structured financial information and historical unstructured text information is obtained, and this set is divided into a training set and a test set; an initial risk detection model is constructed using the Adaboost ensemble learning algorithm; a decision tree is selected as a weak classifier, and each training sample in the training set is assigned an equal weight; the classifier weight distribution adjustment step: following the iterative process of the Adaboost algorithm, a weak classifier is selected in each round, and risk detection training is performed based on the weight distribution of the training samples, wherein the weights of the weak classifiers are adjusted according to the classification accuracy; the sample weight update step: after each round of weak classifier training, the weights of the training samples are updated, wherein, when updating the weights of the training samples, the weights of misclassified training samples are increased, and the weights of correctly classified training samples are decreased; the classifier weight distribution adjustment step and the sample weight update step are repeated until a preset number of weak classifiers is reached, resulting in a strong classifier composed of multiple weak classifiers; the performance of the credit risk detection model is evaluated using a test set, and if the performance evaluation results meet the evaluation accuracy, the training is confirmed to be complete, and the credit risk detection model is obtained.

[0092] In training the credit risk detection model, the first step is to obtain a set of detection information from historical data. This set includes structured financial information and unstructured text information. To evaluate the model's performance and avoid overfitting, the historical detection information set is divided into a training set and a test set. The training set is used for model training, and the test set is used for evaluating model performance.

[0093] AdaBoost is an iterative ensemble learning method designed to build a strong classifier from multiple weak classifiers. This paper uses the AdaBoost algorithm as a foundation to construct an initial risk detection model. In the initialization phase of AdaBoost, a decision tree is selected as a weak classifier, and each training sample in the training set is assigned an equal weight of 1 / N, where N is the total number of training samples. During the iterative training process of AdaBoost, a weak classifier (decision tree) is selected in each round, and risk detection training is performed based on the current sample weight distribution. The weights of the weak classifiers are adjusted according to their classification accuracy; that is, the weights of weak classifiers with high classification accuracy are increased, and the weights of weak classifiers with low classification accuracy are decreased. The adjustment of classifier weights is achieved by calculating the error rate of each weak classifier and applying AdaBoost's weight update rules. After each round of weak classifier training, AdaBoost updates the weight distribution of the training samples. Specifically, the weights of samples misclassified by the current weak classifier are increased, while the weights of correctly classified samples are decreased. This weight update mechanism ensures that the model gradually focuses on samples that are difficult to classify correctly during training, thereby improving the overall classification performance. The Adaboost algorithm generates a series of weak classifiers with different weights through an iterative process. When a preset number of weak classifiers is reached, these classifiers are combined into a strong classifier for final credit risk detection.

[0094] After the strong classifier is built, the performance of the credit risk detection model is evaluated using the previously partitioned test set. Key evaluation metrics include accuracy, recall, and F1 score. If the model's performance on the test set meets a preset accuracy threshold (e.g., accuracy greater than 90%), the model is considered successfully trained, and the final credit risk detection model is confirmed.

[0095] This model can effectively integrate structured financial information and unstructured text information, providing financial institutions with comprehensive credit risk assessment capabilities.

[0096] The following describes in detail another optional implementation method.

[0097] Figure 3 This is a flowchart illustrating a credit risk detection model based on multidimensional information according to an embodiment of the present invention, such as... Figure 3 As shown, including:

[0098] Step 1: Data collection.

[0099] like Figure 3As shown, the first step is to collect corporate financial data, which mainly includes data tables such as the balance sheet, income statement, and cash flow statement. Select financial indicators with more information in the financial statements, including the current ratio, quick ratio, and total asset growth rate in the balance sheet; the gross profit margin, net profit margin, and period expense ratio in the income statement; and the cash cost ratio and financial expense cash ratio in the cash flow statement.

[0100] Considering that corporate financial data is updated regularly and the published data has a certain lag, making it impossible to keep up with the company's operating conditions in a timely manner, this embodiment also considers information in unstructured data in addition to structured financial data. Corporate news reports on financial websites often contain many clues. At the same time, news reports are timely and objective. Introducing news text features into the risk detection model can detect corporate credit risks more promptly.

[0101] Step 2: Data preprocessing and feature selection.

[0102] For structured financial data, this embodiment extracts 176 financial indicators from three financial statements (balance sheet, income statement, and cash flow statement) and their notes, and incorporates them into the initial feature pool. The data is then merged, cleaned, and transformed, including missing value imputation, outlier removal, and removal of outliers from monetary variables. Considering the correlation between the initial features, the contribution of variables to the target prediction is measured by calculating their indicative value (IV), and variables with higher IV values ​​are selected. After calculation, all fields with IV values ​​at the Useless level are removed. For continuous variables, they are first discretized using decision tree binning, and then their IV values ​​are calculated, removing features with IV values ​​less than 0.02.

[0103] Chi-square tests are used for feature selection to assess the correlation between discretized financial data and credit risk. Specifically, a significance level is set (e.g., p-value less than 0.01), and the chi-square value is calculated for each discretized indicator. Features with a correlation lower than a preset threshold for credit risk detection are excluded to reduce model complexity, avoid overfitting, and ensure predictive accuracy and efficiency. A smaller p-value indicates a lower probability that the hypothesis H0 is true under current observations, meaning the current field has a significant impact on the sample distribution; variables with p-values ​​less than 0.01 are removed. For the remaining fields, the Pearson coefficient between each pair is calculated. If the coefficient is greater than 0.95, the field with the lower IV value is removed. For example, 89 financial indicators are selected for credit risk detection.

[0104] For unstructured text data, a word2vec pre-trained model is first used to vectorize the text. Word embedding is then performed on the word vectors by setting the window size to achieve dimensionality reduction. Simultaneously, the context window in this model considers the fact that words with similar distributions of adjacent words often have similar meanings, thus including word meaning in the model.

[0105] Step 3: Topic model construction and training.

[0106] Keywords in news reports can summarize the main content of the news. For example, when a report repeatedly uses keywords such as "year-on-year," "growth," "net profit," and "hundred million yuan," the report is generally related to the company's profitability. Conversely, when a report repeatedly uses words such as "goodwill," "claims," ​​"fines," "warnings," and "violations," it is often negative news about the company. Companies whose news reports frequently disclose profitability tend to have lower creditworthiness, while companies whose news reports are mostly negative tend to have higher credit risk. Therefore, based on the vectorization of text, this paper further uses a topic model to extract text keyword representations. Building upon traditional topic models, it proposes a Dirichlet multinomial mixture model for short text topic representation and explores more effective methods for solving the Dirichlet multinomial mixture model in short text topic representation, aiming to improve the accuracy and adaptability of short text topic representation.

[0107] This embodiment uses the Dirichlet multinomial mixture model for short text topic representation. This model is a nonparametric Bayesian model that can well balance the sparsity and high dimensionality problems in short text topic representation.

[0108] Traditional LDA topic models assume that each document in a document set is distributed across topics. When generating words in a document, a topic is first generated according to the topic distribution of the document, and then a word is generated based on the distribution of that topic across words. Dirichlet multinomial mixture models assume that documents are generated by a mixture model, with a one-to-one correspondence between mixture components and clusters. When generating a document, a topic is first selected from K topics based on the weights p(z=k) of each topic, let's say it's the Kth topic. Then, document w is generated based on the distribution p(u|z=k) corresponding to that topic. The probability that document w is generated by the Dirichlet multinomial mixture model is:

[0109]

[0110] Where K is the number of all topics. The weights of each topic correspond to a multinomial distribution p(z=k)=Mult(z=k|θ)=θ k The prior distribution of this multinomial distribution is the Dirichlet distribution. in The distribution corresponding to each topic is a multinomial distribution p(x|z=k)=Mult(x|φ) k )=φ k,x The prior distribution for each topic is a Dirichlet distribution. in

[0111] like Figure 4 The diagram shows the probabilistic graphical model of the Dirichlet polynomial mixture model. The generation process of the Dirichlet polynomial mixture model is as follows:

[0112]

[0113] z u |θ~Mult(θ)

[0114]

[0115] Where p(u|φ) z,u )=Π x∈u Mult(x|φ z,u ).

[0116] Compared to traditional LDA topic models, Dirichlet Multinomial Mixture Models (DMMs) do not require generating high-dimensional vectors of dictionary length; they only need to generate words from the text, thus better describing the generation process of short texts. When used for clustering problems, DMMs are typically solved using the Gibbs sampling algorithm. Considering that the random variables θ and φ from the DMM are not needed in short text representation, they are marginalized during the solution process, and the conditional distribution is directly solved. The conditional probability that short text u is generated from topic z is:

[0117]

[0118] Where N is the total number of all texts, This refers to the number of texts in topic z without considering document u. When there is more text in topic z, text u will be more likely to choose a topic with more documents, thus achieving better integrity. That is, texts that should actually be generated by the same topic model can be assigned to the same topic.

[0119] This refers to the total number of words in topic z when text u is not considered. This represents the frequency of word x in topic z without considering text u. When more texts in topic z have similar words to text w, text w is more likely to be generated through topic z, thus having better similarity. That is, texts assigned to the same topic actually belong to the same topic.

[0120] The learning process of the topic model is a parameter estimation process. First, K topics are initialized, and texts are randomly assigned to a topic. During the iteration process, the probability of each short text belonging to each topic is calculated, and the text is reassigned according to the probability distribution. Each iteration makes the topic model develop in a more accurate direction. Finally, the probability distribution of each news document on the topic can be obtained, which can be used for subsequent risk detection models.

[0121] Next, we will explain the impact of parameters α and β on the results. Based on the conditional probability formula for short text u being generated from topic z, we can see that: when α = 0, the probability of a text selecting an empty topic is zero; as α increases, the probability of a text selecting an empty topic increases accordingly, and a larger α makes it more likely that a topic with only one text will appear. When β is small, the probability of a text selecting a certain topic will... The value of β is more sensitive and focuses more on the number of times word x appears in topic z, that is, the similarity of topics between texts. Therefore, it can form relatively more topics and higher similarity. When β is larger, it forms relatively fewer topics and higher similarity.

[0122] In addition, to explore the impact of the initial number of topics K on the results, the number of iterations was set to 10, with initial α = 0.2 and β = 0.2. It can be found that the larger the initial number of topics K is, the closer the number of topics is to the actual number of topics when the algorithm finally stabilizes.

[0123] Step 4: Risk detection model construction and training.

[0124] The processed structured financial data and unstructured text data are concatenated into a wide table (corresponding to the detection information set obtained by merging the customer's financial indicator set and text topic probability distribution data in the above embodiment). The representation of the unstructured text data includes word2vec vectorization representation and the probability distribution of the text on the topic. The risk detection model is constructed using the Adaboost ensemble learning algorithm.

[0125] During training, the data is divided into training set X. train and test set X test Let the training sample set be T={(x1,y1),(x2,y2),…,(x m ,y m The output is {-1, +1}. The AdaBoost model uses decision trees as the base classifier. During training, different weights are assigned to the classifiers to improve multiple weak classifiers into strong classifiers. Let the weight of the input sample of the k-th weak learner in the training set be D(k) = (w...). k1 ,w k2 ,…,w kmThe weights of the training samples for the next classifier will be adjusted accordingly based on the classification performance of the previous classifier.

[0126] Through the above implementation methods, a credit risk detection model integrating multi-dimensional information is constructed. Based on traditional structured financial data, it focuses on the footnotes to financial statements and introduces unstructured news text data to analyze corporate credit risk from multiple information dimensions. Compared with existing technologies, it can more comprehensively and accurately detect corporate credit risk. Furthermore, in the representation of unstructured text data, it integrates the advantages of pre-trained models and topic models, while considering the vectorized representation of text and topic probability distribution to enrich the text information.

[0127] The following is a detailed description with reference to another embodiment.

[0128] Example 2

[0129] The credit risk detection device provided in this embodiment includes multiple implementation units, each of which corresponds to a specific implementation step in the above embodiment one. The specific implementation method and beneficial effects can be referred to the aforementioned method embodiment, and will not be repeated here.

[0130] Figure 5 This is a schematic diagram of an optional credit risk detection device according to an embodiment of the present invention, such as... Figure 3 As shown, the credit risk detection device may include: a data collection unit 51, a data screening unit 52, a data fusion unit 53, and a risk assessment unit 54.

[0131] The data collection unit 51 is used to collect structured financial data and unstructured news text data associated with the target customer in the credit business request after receiving the credit business request, and to preprocess the structured financial data and news text data.

[0132] The data filtering unit 52 is used to filter the preprocessed financial data, output a set of financial indicators for credit risk detection, and use a Dirichlet multinomial mixture model to perform topic representation on the preprocessed news text data, outputting the text topic probability distribution data of each news document.

[0133] The data fusion unit 53 is used to fuse the filtered set of financial indicators with the text topic probability distribution data after topic representation to generate a set of detection information containing structured financial information and unstructured text information.

[0134] Risk assessment unit 54 is used to input the detection information set into the credit risk detection model and output a risk assessment report on the current credit business initiated by the target customer and credit business approval suggestions.

[0135] The aforementioned credit risk detection device, upon receiving a credit business request, can collect structured financial data and unstructured news text data associated with the target customer in the credit business request through the data collection unit 51. It can preprocess the structured financial data and news text data, and then filter the preprocessed financial data through the data filtering unit 52, outputting a set of financial indicators for credit risk detection. It can also use a Dirichlet multinomial mixture model to perform thematic representation on the preprocessed news text data, outputting the text thematic probability distribution data for each news document. The data fusion unit 53 merges the filtered set of financial indicators with the thematically represented text thematic probability distribution data to generate a detection information set containing both structured financial information and unstructured text information. Finally, the risk assessment unit 54 inputs the detection information set into the credit risk detection model, outputting a risk assessment report and credit approval recommendations for the current credit business initiated by the target customer. In this embodiment, unstructured news text data can be introduced on the basis of traditional structured financial data to analyze customer credit risk from multiple information dimensions. This enables more comprehensive and accurate detection of corporate credit risk. At the same time, in the representation of unstructured text data, the advantages of pre-trained models and topic models are combined, while considering the vectorized representation of text and topic probability distribution, enriching text information, improving the accuracy of credit risk detection, enhancing the risk control capabilities and market competitiveness of financial institutions, reducing the property losses of financial institutions, and solving the technical problem in related technologies where financial institutions rely on a relatively single data source when conducting risk detection of customer credit business, resulting in incomplete information collection, affecting credit risk detection, and easily causing property losses for financial institutions.

[0136] Optionally, the data collection unit includes: a financial report collection module, used to collect financial reports related to the target customer, and filter out the balance sheet, income statement and cash flow statement from the financial reports to obtain a financial data table; a financial data extraction module, used to extract multiple financial indicators and the corresponding financial data from the financial data table to obtain structured financial data, wherein the multiple financial indicators include at least one of the following: current ratio, quick ratio, total asset growth rate, gross profit margin, net profit margin, period expense ratio, cash cost ratio and financial expense cash ratio; a news report collection module, used to collect news reports related to the target customer from financial news websites and financial engine databases, wherein the types of news reports include at least one of the following: corporate profit announcements, goodwill warnings, penalties for violations and industry news; and a news text extraction module, used to extract text data from news reports to obtain unstructured news text data.

[0137] Optionally, the data collection unit includes: a data preprocessing module for cleaning and handling missing values ​​of structured financial data, wherein data cleaning includes deleting duplicate values ​​and outliers, and missing value handling includes filling in missing data; cleaning news text data, wherein cleaning includes removing stop words, removing punctuation marks, and removing numbers and special characters; and a word segmentation module for segmenting the cleaned news text to obtain a set of text words and a set of text sentences.

[0138] Optionally, the data filtering unit includes: an IV value calculation module, used to calculate the IV value of each financial indicator in the structured financial data, and filter out financial indicators with IV values ​​greater than or equal to a preset threshold to obtain financial indicators associated with credit risk detection; a discretization module, used to discretize the continuous financial data in the structured financial data, and use a decision tree binning strategy to convert continuous variables into discrete variables; and a feature filtering module, used to use a chi-square test strategy to filter features of the discretized financial data, remove features with a correlation degree lower than a preset correlation degree threshold for credit risk detection, and output a set of financial indicators for credit risk detection.

[0139] Optionally, the data filtering unit further includes: a word vectorization processing module, used to perform word vectorization processing on the pre-processed news text using a pre-trained Word2Vec model to construct a word vector matrix; and a topic extraction module, used to extract topics from the word vector matrix using a Dirichlet multinomial mixture model to obtain the probability distribution of each news document on each topic.

[0140] Optionally, the Dirichlet multinomial mixture model is trained as follows: Initialize model parameters, including the number of topics and the hyperparameters of the Dirichlet distribution; acquire preprocessed historical news text data and randomly assign each historical news document to a topic as the initial state for model training; perform a topic probability calculation step, iteratively calculating the probability of each historical news document belonging to each topic, wherein, when calculating the probability of each historical news document belonging to each topic, a Gibbs sampling algorithm is used for probability estimation based on the topic features of each document and the model parameters; perform a probability redistribution step, redistributing each historical news document to the target topic according to the calculated probabilities; repeat the iterative topic probability calculation step and the probability redistribution step until the rate of change of document topic allocation is lower than a preset threshold or the maximum number of iterations is reached, completing the topic model convergence and obtaining the trained Dirichlet multinomial mixture model.

[0141] Optionally, the data fusion unit includes: a data merging module, which, based on a data table connection strategy, merges the target customer's financial indicator set and text topic probability distribution data into a detection information set, wherein, in the detection information set, independent columns are assigned to each financial indicator in the financial indicator set and each text subject in the text topic probability distribution data.

[0142] Optionally, the credit risk detection model is trained as follows: A historical detection information set containing historical structured financial information and historical unstructured text information is obtained, and this set is divided into a training set and a test set; an initial risk detection model is constructed using the Adaboost ensemble learning algorithm; a decision tree is selected as a weak classifier, and each training sample in the training set is assigned an equal weight; the classifier weight distribution adjustment step: following the iterative process of the Adaboost algorithm, a weak classifier is selected in each round, and risk detection training is performed based on the weight distribution of the training samples, wherein the weights of the weak classifiers are adjusted according to the classification accuracy; the sample weight update step: after each round of weak classifier training, the weights of the training samples are updated, wherein, when updating the weights of the training samples, the weights of misclassified training samples are increased, and the weights of correctly classified training samples are decreased; the classifier weight distribution adjustment step and the sample weight update step are repeated until a preset number of weak classifiers is reached, resulting in a strong classifier composed of multiple weak classifiers; the performance of the credit risk detection model is evaluated using a test set, and if the performance evaluation results meet the evaluation accuracy, the training is confirmed to be complete, and the credit risk detection model is obtained.

[0143] The aforementioned credit risk detection device may also include a processor and a memory. The aforementioned data collection unit 51, data filtering unit 52, data fusion unit 53, risk assessment unit 54, etc., are all stored in the memory as program units, and the processor executes the aforementioned program units stored in the memory to realize the corresponding functions.

[0144] The aforementioned processor contains a kernel, which retrieves the corresponding program units from memory. One or more kernels can be configured, and credit risk detection based on multi-dimensional information can be achieved by adjusting kernel parameters.

[0145] The aforementioned memory may include non-permanent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash memory, and the memory includes at least one memory chip.

[0146] Example 3

[0147] Embodiments of this application may provide an electronic device. Figure 6This is a structural block diagram of an electronic device according to an embodiment of this application. Figure 6 As shown, the electronic device may include: one or more ( Figure 6 (Only one is shown) processor 602, memory 604, memory controller, and peripheral interface, wherein the peripheral interface is connected to the radio frequency module, audio module and display.

[0148] The memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the credit risk detection method and apparatus in this application embodiment. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, thereby realizing the aforementioned credit risk detection method. The memory may include high-speed random access memory, and may also include 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 memory remotely located relative to the processor, and these remote memories can be connected to the terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0149] The processor can access information and applications stored in memory via a transmission device to perform the following steps: Upon receiving a credit business request, it collects structured financial data and unstructured news text data associated with the target customer in the credit business request, and preprocesses the structured financial data and news text data; it filters the preprocessed financial data, outputs a set of financial indicators for credit risk detection, and uses a Dirichlet multinomial mixture model to perform topic representation on the preprocessed news text data, outputting the text topic probability distribution data for each news document; it merges the filtered set of financial indicators with the topic-represented text topic probability distribution data to generate a detection information set containing structured financial information and unstructured text information; it inputs the detection information set into the credit risk detection model, and outputs a risk assessment report and credit business approval recommendations for the current credit business initiated by the target customer.

[0150] The processor can also access information and applications stored in memory via a transmission device to perform the following steps: collect financial reports related to the target customer, and filter out the balance sheet, income statement, and cash flow statement from the financial reports to obtain a financial data table; extract multiple financial indicators and the corresponding financial data from the financial data table to obtain structured financial data, wherein the multiple financial indicators include at least one of the following: current ratio, quick ratio, total asset growth rate, gross profit margin, net profit margin, period expense ratio, cash cost ratio, and financial expense cash ratio; collect news reports related to the target customer from financial news websites and financial engine databases, wherein the types of news reports include at least one of the following: corporate profit announcements, goodwill warnings, penalties for violations, and industry news; extract text data from the news reports to obtain unstructured news text data.

[0151] The processor can also access information and applications stored in the memory via a transmission device to perform the following steps: data cleaning and missing value processing of structured financial data, wherein data cleaning includes: deleting duplicate values ​​and deleting outliers, and missing value processing includes: filling in missing data; cleaning news text data, wherein cleaning includes removing stop words, removing punctuation marks, and removing numbers and special characters; and performing word segmentation on the cleaned news text to obtain a set of text words and a set of text sentences.

[0152] The processor can also call the information and application programs stored in the memory through the transmission device to perform the following steps: calculate the IV value of each financial indicator in the structured financial data, filter out financial indicators with IV values ​​greater than or equal to a preset threshold, and obtain financial indicators associated with credit risk detection; discretize the continuous financial data in the structured financial data, and use a decision tree binning strategy to convert continuous variables into discrete variables; use a chi-square test strategy to perform feature filtering on the discretized financial data, remove features with a correlation degree lower than a preset correlation degree threshold for credit risk detection, and output a set of financial indicators for credit risk detection.

[0153] The processor can also access information and applications stored in the memory via a transmission device to perform the following steps: using a pre-trained Word2Vec model to vectorize the pre-processed news text and construct a word vector matrix; using a Dirichlet multinomial mixture model to extract topics from the word vector matrix and obtain the probability distribution of each news document on each topic.

[0154] The processor can also access information and applications stored in the memory via a transmission device to execute the following steps: The Dirichlet multinomial mixture model is trained as follows: Initialize model parameters, including the number of topics and the hyperparameters of the Dirichlet distribution; acquire preprocessed historical news text data and randomly assign each historical news document to a topic as the initial state for model training; topic probability calculation step, iteratively calculating the probability of each historical news document belonging to each topic, wherein, when calculating the probability of each historical news document belonging to each topic, a Gibbs sampling algorithm is used for probability estimation based on the topic features of each document and the model parameters; probability redistribution step, redistributing each historical news document to the target topic according to the calculated probability; repeat the iterative calculation of topic probability step and probability redistribution step until the rate of change of document topic allocation is lower than a preset threshold or the maximum number of iterations is reached, completing the topic model convergence and obtaining the trained Dirichlet multinomial mixture model.

[0155] The processor can also call the information and application stored in the memory through the transmission device to perform the following steps: based on the data table connection strategy, merge the target customer's financial indicator set and text topic probability distribution data into the detection information set, wherein, in the detection information set, each financial indicator in the financial indicator set and each text subject in the text topic probability distribution data are assigned an independent column.

[0156] The processor can also access information and applications stored in memory via a transmission device to execute the following steps: The credit risk detection model is trained as follows: A historical detection information set containing historical structured financial information and historical unstructured text information is obtained, and this set is divided into a training set and a test set; an initial risk detection model is constructed using the Adaboost ensemble learning algorithm; a decision tree is selected as a weak classifier, and equal weights are assigned to each training sample in the training set; the classifier weight distribution adjustment steps are as follows: following the iterative process of the Adaboost algorithm, a weak classifier is selected in each round, based on the weight distribution of the training samples. The risk detection training process involves adjusting the weights of weak classifiers based on classification accuracy. The sample weight update step involves updating the weights of training samples after each round of weak classifier training, increasing the weights of misclassified samples and decreasing the weights of correctly classified samples. This process is repeated until a preset number of weak classifiers is reached, resulting in a strong classifier composed of multiple weak classifiers. The credit risk detection model is then evaluated using a test set. If the performance evaluation results meet the accuracy target, the training is considered complete, and the credit risk detection model is obtained.

[0157] Those skilled in the art will understand that Figure 6 The structure shown is for illustrative purposes only. Electronic devices can also be smartphones, tablets, handheld computers, mobile internet devices (MIDs), PADs, and other terminal devices. Figure 6 This does not limit the structure of the aforementioned electronic device. For example, electronic devices may also include components that are more... Figure 6 The more or fewer components shown (such as network interfaces, display devices, etc.), or having the same Figure 6 The different configurations shown.

[0158] Those skilled in the art will understand that all or part of the steps in the various credit risk detection methods of the above embodiments can be implemented by a program instructing the hardware of the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0159] Example 4

[0160] Embodiments of this application also provide a storage medium. Optionally, in this embodiment, the storage medium can be used to store the program code executed by the credit risk detection method provided in Embodiment 1.

[0161] According to another aspect of the present invention, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored computer program, wherein, when the computer program is running, it controls the device where the computer-readable storage medium is located to execute any of the credit risk detection methods in the first embodiment described above.

[0162] Optionally, in this embodiment, the storage medium may be located in any computer terminal in a group of computer terminals in a computer network, or in any mobile terminal in a group of mobile terminals.

[0163] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the credit risk detection method described in various embodiments of this application.

[0164] This application also provides a computer program product, including a non-volatile computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the credit risk detection method described in various embodiments of this application.

[0165] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0166] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0167] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0168] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0169] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0170] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0171] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A credit risk detection method, characterized in that, include: Upon receiving a credit business request, the system collects structured financial data and unstructured news text data associated with the target customer in the credit business request, and preprocesses the structured financial data and the news text data. The preprocessed financial data is filtered to output a set of financial indicators for credit risk detection. The preprocessed news text data is then characterized using a Dirichlet multinomial mixture model to output the text topic probability distribution data for each news document. The step of filtering the preprocessed financial data and outputting a set of financial indicators for credit risk detection includes: calculating the IV value of each financial indicator in the structured financial data, filtering out financial indicators with IV values ​​greater than or equal to a preset threshold to obtain financial indicators associated with credit risk detection; discretizing the continuous financial data in the structured financial data by using a decision tree binning strategy to convert continuous variables into discrete variables; and using a chi-square test strategy to perform feature filtering on the discretized financial data, removing features with a correlation degree lower than a preset correlation degree threshold for credit risk detection, and outputting the set of financial indicators for credit risk detection. The step of using a Dirichlet multinomial mixture model to perform topic representation on the preprocessed news text data includes: using a pre-trained Word2Vec model to perform word vectorization on the preprocessed news text to construct a word vector matrix; and using the Dirichlet multinomial mixture model to extract topics from the word vector matrix to obtain the probability distribution of each news document on each topic. The Dirichlet multinomial mixture model is trained as follows: Initialize model parameters, including the number of topics and the hyperparameters of the Dirichlet distribution; acquire preprocessed historical news text data and randomly assign each historical news document to a topic as the initial state for model training; perform a topic probability calculation step, iteratively calculating the probability of each historical news document belonging to each topic, wherein, when calculating the probability of each historical news document belonging to each topic, a Gibbs sampling algorithm is used for probability estimation based on the topic features of each document and the model parameters; perform a probability redistribution step, redistributing each historical news document to the target topic according to the calculated probabilities; repeat the iterative topic probability calculation step and the probability redistribution step until the rate of change in document topic allocation is lower than a preset threshold or the maximum number of iterations is reached, completing the topic model convergence and obtaining the trained Dirichlet multinomial mixture model. The selected set of financial indicators is fused with the text topic probability distribution data after topic characterization to generate a set of detection information containing structured financial information and unstructured text information. The detection information set is input into the credit risk detection model, which outputs a risk assessment report and credit approval recommendations for the current credit business initiated by the target customer.

2. The method according to claim 1, characterized in that, The step of fusing the filtered set of financial indicators with the text topic probability distribution data after topic representation to generate a detection information set containing structured financial information and unstructured text information includes: Based on the data table join strategy, the set of financial indicators of the target customer and the probability distribution data of the text topics are merged into the detection information set, wherein each financial indicator in the set of financial indicators and each text subject in the probability distribution data of the text topics are assigned an independent column in the detection information set.

3. The method according to claim 1, characterized in that, The credit risk detection model was trained in the following manner: Obtain a set of historical detection information containing historical structured financial information and historical unstructured text information, and divide the set of historical detection information into a training set and a test set; An initial risk detection model was constructed using the Adaboost ensemble learning algorithm; A decision tree is selected as a weak classifier, and each training sample in the training set is assigned an equal weight. Classifier weight distribution adjustment steps: Following the iterative process of the Adaboost algorithm, a weak classifier is selected in each round, and risk detection training is performed based on the weight distribution of the training samples. The weights of the weak classifier are adjusted according to the classification accuracy. Sample weight update steps: After each round of weak classifier training is completed, the weights of the training samples are updated. When updating the weights of the training samples, the weights of misclassified training samples are increased, and the weights of correctly classified training samples are decreased. Repeat the classifier weight distribution adjustment step and the sample weight update step until the preset number of weak classifiers is reached, and a strong classifier composed of multiple weak classifiers is obtained. The credit risk detection model is evaluated using the test set. If the performance evaluation results meet the evaluation accuracy, the training is confirmed to be complete, and the credit risk detection model is obtained.

4. The method according to claim 1, characterized in that, The steps of collecting structured financial data and unstructured news text data associated with the target customer in the credit transaction request include: Collect financial reports related to the target customers, and filter out the balance sheet, income statement and cash flow statement from the financial reports to obtain a financial data table; The structured financial data is obtained by extracting multiple financial indicators and the corresponding financial data for each financial indicator from the financial data table. The multiple financial indicators include at least one of the following: current ratio, quick ratio, total asset growth rate, gross profit margin, net profit margin, period expense ratio, cash cost ratio, and financial expense cash ratio. Collect news reports related to the target customer from financial news websites and financial search engine databases, wherein the types of news reports include at least one of the following: corporate profit announcements, goodwill warnings, penalties for violations, and industry news. Extract the text data from the news report to obtain unstructured news text data.

5. The method according to claim 1, characterized in that, The steps for preprocessing the structured financial data and the news text data include: The structured financial data is cleaned and missing value is handled, wherein the data cleaning includes: deleting duplicate values ​​and deleting outliers, and the missing value handling includes: filling in missing data; The news text data is cleaned, including removing stop words, punctuation marks, numbers, and special characters. The news text obtained after cleaning is segmented into words to obtain a set of text words and a set of text sentences.

6. A credit risk detection device, characterized in that, include: The data collection unit is used to collect structured financial data and unstructured news text data associated with the target customer in the credit business request after receiving the credit business request, and to preprocess the structured financial data and the news text data. The data filtering unit is used to filter the preprocessed financial data, output a set of financial indicators for credit risk detection, and use a Dirichlet multinomial mixture model to perform topic representation on the preprocessed news text data, outputting the text topic probability distribution data of each news document. The data filtering unit includes: an IV value calculation module, used to calculate the IV value of each financial indicator in the structured financial data, and filter out financial indicators with IV values ​​greater than or equal to a preset threshold to obtain financial indicators associated with credit risk detection; a discretization module, used to discretize the continuous financial data in the structured financial data, and use a decision tree binning strategy to convert continuous variables into discrete variables; and a feature filtering module, used to use a chi-square test strategy to filter features of the discretized financial data, remove features with a correlation degree lower than a preset correlation degree threshold for credit risk detection, and output the set of financial indicators used for credit risk detection. The data filtering unit further includes: a word vectorization processing module, used to perform word vectorization processing on the pre-processed news text using a pre-trained Word2Vec model to construct a word vector matrix; and a topic extraction module, used to extract topics from the word vector matrix using the Dirichlet multinomial mixture model to obtain the probability distribution of each news document on each topic. The Dirichlet multinomial mixture model is trained as follows: Initialize model parameters, including the number of topics and the hyperparameters of the Dirichlet distribution; acquire preprocessed historical news text data and randomly assign each historical news document to a topic as the initial state for model training; perform a topic probability calculation step, iteratively calculating the probability of each historical news document belonging to each topic, wherein, when calculating the probability of each historical news document belonging to each topic, a Gibbs sampling algorithm is used for probability estimation based on the topic features of each document and the model parameters; perform a probability redistribution step, redistributing each historical news document to the target topic according to the calculated probabilities; repeat the iterative topic probability calculation step and the probability redistribution step until the rate of change in document topic allocation is lower than a preset threshold or the maximum number of iterations is reached, completing the topic model convergence and obtaining the trained Dirichlet multinomial mixture model. The data fusion unit is used to fuse the filtered set of financial indicators with the text topic probability distribution data after topic characterization to generate a set of detection information containing structured financial information and unstructured text information. The risk assessment unit is used to input the detection information set into the credit risk detection model and output a risk assessment report and credit approval suggestions for the current credit business initiated by the target customer.

7. 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, it controls the device containing the computer-readable storage medium to perform the credit risk detection method according to any one of claims 1 to 5.

8. An electronic device, characterized in that, It includes one or more processors and a memory, the memory being 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 cause the one or more processors to implement the credit risk detection method according to any one of claims 1 to 5.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the credit risk detection method according to any one of claims 1 to 5.

Images