Credit risk prediction method, system and device

Abstract

The embodiment of the present application discloses a credit risk prediction method and system thereof, which is used to predict credit risk based on the text of the sender, and is simple to operate and has high accuracy. The method of the embodiment of the present application includes: obtaining a target prediction model, wherein the target prediction model is obtained by machine learning training of an initial prediction model by text feature vectors and text feature statistics of historical texts and credit risk probability values corresponding to the historical texts, and the target prediction model stores a target correspondence between the text feature vectors and text feature statistics and the credit risk probability values corresponding to the historical texts; obtaining the text feature vectors and text feature statistics of the target text; inputting the text feature vectors and text statistical feature values of the target text into the target prediction model, and obtaining the credit risk probability prediction value output by the target prediction model according to the target correspondence.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of credit risk prediction, and in particular to a credit risk prediction method, system, and apparatus thereof. Background Art

[0002] Credit risk management involves formulating information policies, guiding and coordinating the business activities of various institutions, and comprehensively overseeing and controlling all aspects of the process, from customer credit investigation and payment method selection to credit limit determination and payment collection, to ensure the safe and timely collection of receivables. Credit risk management presents challenges in quantitative analysis and measurement. Compared to market risk management, which utilizes ample data and extensive mathematical and statistical models, traditional credit risk management lacks scientific quantitative analysis and instead relies more heavily on qualitative analysis and managers' subjective experience and judgment.

[0003] Specifically, existing models for quantitative credit risk analysis mostly rely on manually pre-setting keyword dictionaries based on experience. These dictionaries then retrieve keywords from the sender's text, analyzing and assessing the sender's credit risk based on the keyword hit rate and the type of keywords found. Alternatively, the sender's text is directly converted into text vectors, which are then fed into pre-trained neural network models to analyze the text and output a credit risk assessment.

[0004] However, keyword dictionaries rely on manual experience to generate and require continuous maintenance, making them difficult to cope with massive amounts of continuously updated text. Furthermore, evaluating keyword hit rates and hit keyword types is cumbersome and labor-intensive. The use of text vectors for machine learning analysis lacks manual intervention and suffers from the low accuracy of neural network models, resulting in low accuracy in the final credit risk assessment results. Summary of the Invention

[0005] The embodiments of the present application provide a credit risk prediction method, system, and apparatus thereof, which are used to predict credit risk based on the text of the sender, with simple operation and high accuracy.

[0006] The credit risk prediction method provided in the embodiment of the present application includes:

[0007] Obtaining a target prediction model, wherein the target prediction model is obtained by performing machine learning training on an initial prediction model based on text feature vectors and text feature statistics of historical texts and credit risk probability values corresponding to the historical texts, the credit risk probability values corresponding to the historical texts being obtained by analyzing the text feature vectors and text feature statistics of the historical texts, and the target prediction model storing a target correspondence between the text feature vectors and text feature statistics and the credit risk probability values;

[0008] Obtain text feature vectors and text feature statistics of the target text;

[0009] The text feature vector and text statistical feature value of the target text are input into the target prediction model to obtain the credit risk probability prediction value output by the target prediction model according to the target correspondence relationship, and the credit risk probability prediction value is used to represent the credit risk of the target text.

[0010] Optionally, before obtaining the target prediction model, the method further includes:

[0011] Obtaining a text feature vector and text feature statistics of the historical text and a credit risk probability value corresponding to the historical text;

[0012] The text feature vector and text feature statistical value of the historical text and the credit risk probability value corresponding to the historical text are input into the initial prediction model as training samples, and the initial prediction model is subjected to machine learning training using the training samples to obtain the target prediction model, in which the target prediction model stores the target correspondence between the text feature vector and text feature statistical value and the credit risk probability value.

[0013] Optionally, obtaining the text feature vector and text feature statistics of the historical text includes:

[0014] Using a text vector algorithm to convert the text data of the historical text into a text feature vector of the historical text;

[0015] The text data of the historical text is processed to obtain text feature statistics of the historical text.

[0016] Optionally, the processing of the text data of the historical text to obtain text feature statistics of the historical text includes:

[0017] Establishing an industry dictionary based on expert prior knowledge, wherein the industry dictionary includes the industry type of the sender of the historical text and industry characteristic keywords corresponding to the industry type;

[0018] Establishing a risk type vocabulary based on expert prior knowledge, wherein the risk type vocabulary is formed by expanding the risk type seed vocabulary based on expert prior knowledge using a new word discovery algorithm;

[0019] Classifying the first text data of the historical text according to the industry dictionary to obtain the industry type corresponding to the first text data of the historical text;

[0020] classifying the second text data of the historical text according to the risk type vocabulary to obtain a risk type corresponding to the second text data of the historical text;

[0021] Inputting the industry type and the risk type into a credit risk scoring model to obtain an industry type score and a risk type score corresponding to the historical text, wherein the credit risk scoring model stores a credit risk scoring system based on expert prior knowledge;

[0022] sorting the third text data of the historical text to obtain other characteristic statistical values of the historical text;

[0023] The industry type score, the risk type score, and the other feature statistical values are determined as text feature statistical values of the historical text.

[0024] Optionally, inputting the text feature vector and text feature statistics of the historical text and the credit risk probability value corresponding to the historical text as training samples into the initial prediction model includes:

[0025] The text feature vectors and text feature statistics of the historical text are classified and fused according to the sender of the historical text to form standardized format data as training samples, which are input into the initial prediction model.

[0026] A credit risk prediction system provided in an embodiment of the present application includes:

[0027] an acquisition unit, configured to acquire a target prediction model, wherein the target prediction model is obtained by performing machine learning training on an initial prediction model using text feature vectors and text feature statistics of historical texts and credit risk probability values, the credit risk probability values being obtained by analyzing the text feature vectors and text feature statistics of the historical texts, and the target prediction model storing a target correspondence between the text feature vectors and text feature statistics and the credit risk probability values;

[0028] The acquisition unit is further configured to acquire a text feature vector and text feature statistics of the target text;

[0029] An input unit is used to input the text feature vector and text statistical feature value of the target text into a target prediction model to obtain a credit risk probability prediction value output by the target prediction model according to the target correspondence relationship, and the credit risk probability prediction value is used to represent the credit risk of the target text.

[0030] Optionally, the acquisition unit is specifically configured to acquire a text feature vector and text feature statistics of the historical text and a credit risk probability value corresponding to the historical text;

[0031] The text feature vector and text feature statistical value of the historical text and the credit risk probability value corresponding to the historical text are input into the initial prediction model as training samples, and the initial prediction model is subjected to machine learning training using the training samples to obtain the target prediction model, in which the target prediction model stores the target correspondence between the text feature vector and text feature statistical value and the credit risk probability value.

[0032] Optionally, the acquisition unit is further configured to convert the text data of the historical text into a text feature vector of the historical text using a text vector algorithm;

[0033] The text data of the historical text is processed to obtain text feature statistics of the historical text.

[0034] Optionally, the acquisition unit is further configured to establish an industry dictionary based on expert prior knowledge, wherein the industry dictionary includes the industry type of the sender of the historical text and industry characteristic keywords corresponding to the industry type;

[0035] Establishing a risk type vocabulary based on expert prior knowledge, wherein the risk type vocabulary is formed by expanding the risk type seed vocabulary based on expert prior knowledge using a new word discovery algorithm;

[0036] Classifying the first text data of the historical text according to the industry dictionary to obtain the industry type corresponding to the first text data of the historical text;

[0037] classifying the second text data of the historical text according to the risk type vocabulary to obtain a risk type corresponding to the second text data of the historical text;

[0038] Inputting the industry type and the risk type into a credit risk scoring model to obtain an industry type score and a risk type score corresponding to the historical text, wherein the credit risk scoring model stores a credit risk scoring system based on expert prior knowledge;

[0039] sorting the third text data of the historical text to obtain other characteristic statistical values of the historical text;

[0040] The industry type score, the risk type score, and the other feature statistical values are determined as text feature statistical values of the historical text.

[0041] Optionally, the input unit is specifically configured to classify and fuse the text feature vectors and text feature statistics of the historical text according to the sender of the historical text to form standardized format data as training samples, and input the data into the initial prediction model.

[0042] The present application also provides a credit risk prediction device, which includes:

[0043] CPU, memory and input / output interfaces;

[0044] The memory is a short-term storage memory or a persistent storage memory;

[0045] The central processing unit is configured to communicate with the memory and execute instructions in the memory to perform the aforementioned credit risk prediction method.

[0046] An embodiment of the present application further provides a computer-readable storage medium comprising instructions, which, when executed on a computer, enable the computer to execute the aforementioned credit risk prediction method.

[0047] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:

[0048] Conduct machine learning training on historical texts in advance to establish a credit risk prediction model, and obtain the text feature vector and text feature statistics of the target text. The text feature statistics of the target text are obtained based on expert prior knowledge, and the text feature vector of the target text is obtained using a text vector algorithm. These two types of data are input into the pre-established credit risk prediction model to obtain the credit risk probability prediction value corresponding to the target text. The operation is simple and the accuracy is high. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A schematic diagram of a credit risk prediction method provided in an embodiment of the present application;

[0050] Figure 2 Another schematic diagram of the credit risk prediction method provided in an embodiment of the present application;

[0051] Figure 3 A schematic diagram of a credit risk prediction system provided in an embodiment of the present application;

[0052] Figure 4 A schematic diagram of a credit risk prediction device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0053] The embodiments of the present application provide a credit risk prediction method and system thereof, which are used to predict credit risk based on the text of the sender, with simple operation and high accuracy.

[0054] Credit risk management involves formulating information policies, guiding and coordinating the business activities of various institutions, and comprehensively overseeing and controlling all aspects of the process, from customer credit investigation and payment method selection to credit limit determination and payment collection, to ensure the safe and timely collection of receivables. Credit risk management presents challenges in quantitative analysis and measurement. Compared to market risk management, which utilizes ample data and extensive mathematical and statistical models, traditional credit risk management lacks scientific quantitative analysis and instead relies more heavily on qualitative analysis and managers' subjective experience and judgment.

[0055] Specifically, existing models for quantitative credit risk analysis mostly rely on manually pre-setting keyword dictionaries based on experience. These dictionaries then retrieve keywords from the sender's text, analyzing and assessing the sender's credit risk based on the keyword hit rate and the type of keywords found. Alternatively, the sender's text is directly converted into text vectors, which are then fed into pre-trained neural network models to analyze the text and output a credit risk assessment.

[0056] However, keyword dictionaries rely on manual experience to generate and require ongoing maintenance, making them difficult to handle with massive amounts of constantly updated text. Furthermore, evaluating keyword hit rates and types is cumbersome and labor-intensive. Machine learning analysis using text vectors suffers from a lack of human intervention and the low accuracy of neural network models, resulting in low accuracy in the final credit risk assessment results. Therefore, a credit risk assessment method is needed that is both less reliant on manual experience and labor-intensive to maintain, while also providing a high degree of accuracy for quantitative analysis of credit risk.

[0057] See also Figure 1 An implementation of a credit risk prediction method provided in an embodiment of the present application includes steps 101 to 103.

[0058] 101. Obtain a target prediction model.

[0059] The target prediction model is established in advance and is used to predict the credit risk probability value. It is obtained by performing machine learning training on the initial prediction model based on the text feature vectors and text feature statistics of the historical text and the credit risk probability values corresponding to the historical text. Specifically, the XGBoost algorithm can be used to perform machine learning training on the initial prediction model. There are many other algorithms that can achieve the same or similar technical effects, and there is no specific restriction on which algorithm to use here.

[0060] The credit risk probability value corresponding to the historical text is obtained by analyzing the text feature vector and text feature statistical value of the historical text in combination with manual experience. The target prediction model stores the target correspondence between the text feature vector and text feature statistical value and the credit risk probability value. This allows the target prediction model to output the credit risk probability value corresponding to the target text based on the target correspondence after the text feature vector and text feature statistical value of the target text are input into the target prediction model.

[0061] 102. Obtain text feature vectors and text feature statistics of the target text.

[0062] A text vector algorithm is used to convert the target text's text data into a text feature vector of the target text, and the target text data is processed to obtain text feature statistics of the target text. The text vector algorithm may be an embedding algorithm, specifically a word2vec model or a BERT model, to convert the target text data into a text feature vector of the target text. Other algorithms that achieve the same or similar effects may also be used to convert text into vectors, and the specifics are not limited here.

[0063] Specifically, the target text's corresponding industry type is determined based on a pre-established industry dictionary based on expert prior knowledge. This dictionary includes the sender's industry type and corresponding keywords. For example, if the target text contains "fenqile," it's considered to belong to the lending industry; if it contains "XX Bank," it's considered to belong to the banking industry, and so on.

[0064] The credit risk type corresponding to the target text is obtained from a pre-established risk type vocabulary based on expert prior knowledge. The risk type vocabulary is formed by applying a new word discovery algorithm to a risk type seed vocabulary based on expert prior knowledge. Specifically, this involves first establishing a risk type seed vocabulary based on expert prior knowledge, then expanding this seed vocabulary using a machine learning algorithm to form a more comprehensive risk type vocabulary. This expansion process using machine learning algorithms requires manual setting of corpus labels and corpus similarity thresholds to ensure the completeness and accuracy of the resulting risk type vocabulary.

[0065] The industry type and risk type corresponding to the historical text are input into the credit risk scoring model to obtain the industry type score and risk type score corresponding to the historical text. The credit risk scoring model here is different from the aforementioned target prediction model. The credit risk scoring model stores a credit risk scoring system based on expert prior knowledge, which is used to score the industry type and risk type corresponding to the historical text obtained in the aforementioned steps.

[0066] In addition, some text data in the historical text does not need to be classified and scored. By summarizing and organizing this part of the text data, other feature statistical values that can be used for the target prediction model can be obtained, and the industry type score, risk type score and other feature statistical values can be determined as the text feature statistical values of the historical text.

[0067] 103. Input the text feature vector and text statistical feature value of the target text into the target prediction model to obtain the credit risk probability prediction value output by the target prediction model according to the target corresponding relationship.

[0068] The text feature vector and text statistical feature value of the target text are input into the target prediction model to obtain the credit risk probability prediction value output by the target prediction model according to the target correspondence relationship. The credit risk probability prediction value is used to represent the credit risk of the target text.

[0069] In this embodiment, machine learning training is performed on historical texts in advance to establish a credit risk prediction model, and the text feature vector and text feature statistics of the target text are obtained. The text feature statistics of the target text are obtained based on expert prior knowledge, and the text feature vector of the target text is obtained using a text vector algorithm. These two types of data are input into the pre-established credit risk prediction model to obtain the credit risk probability prediction value corresponding to the target text. The operation is simple and the accuracy is high.

[0070] See also Figure 2 Another implementation of a credit risk prediction method provided in an embodiment of the present application includes steps 201 to 204.

[0071] 201. Obtain a text feature vector and text feature statistics of a historical text and a credit risk probability value corresponding to the historical text.

[0072] The text vector algorithm is used to convert the text data of the historical text into the text feature vector of the historical text, and the text data of the historical text is classified and scored to obtain the text feature statistical value of the historical text.

[0073] Specifically, the industry type corresponding to the historical text is obtained based on a pre-established industry dictionary based on expert prior knowledge. Based on expert prior knowledge, that is, the industry dictionary references the experience of industry experts. The industry dictionary includes the industry type of the sender of the historical text and the corresponding industry-specific keywords. For example, if the historical text contains "Fenqi Le", it is considered to belong to the lending industry, and if it contains "XX Bank", it is considered to belong to the banking industry, and so on.

[0074] Based on a pre-established risk type vocabulary based on expert prior knowledge, we retrieve the credit risk types corresponding to historical texts. This risk type vocabulary is formed by applying a new word discovery algorithm to the risk type seed vocabulary established based on expert prior knowledge. Specifically, this involves first establishing a risk type seed vocabulary based on expert prior knowledge, then expanding this seed vocabulary using a machine learning algorithm to form a more comprehensive risk type vocabulary. This expansion process using machine learning algorithms requires manually setting corpus labels and corpus similarity thresholds to ensure the completeness and accuracy of the resulting risk type vocabulary. For example, based on expert prior knowledge, we first determine the risk seed word "severely overdue | overdue by more than 90 days." We classify this word as belonging to the "severely overdue" risk type. Then, based on this seed word, we use a new word discovery algorithm to discover more new risk words belonging to the "severely overdue" risk type, such as "malicious overdraft | loan fraud | will file a lawsuit in court | file a case," and so on. These words are then added to the "severely overdue" risk type vocabulary. This process continues by expanding the seed vocabulary for various risk types to form a more comprehensive risk type vocabulary.

[0075] 202. Input the text feature vector and text feature statistics of the historical text and the credit risk probability value corresponding to the historical text as training samples into the initial prediction model, and use the training samples to perform machine learning training on the initial prediction model to obtain a target prediction model.

[0076] Since each sender may correspond to multiple historical texts, there may be information overlap or data overlap in the text feature vectors and text feature statistics of one or more historical texts. However, when making predictions using the target prediction model, only one data value or vector of one data dimension needs to be input into the target prediction model. Therefore, it is necessary to first classify and fuse the text feature vectors and text feature statistics of the historical texts according to the senders of the historical texts to form standardized format data as training samples, which are then input into the initial prediction model. Specifically, the classification and fusion includes classifying similar text feature vectors and text feature statistics in historical texts of the same sender and fusing similar text feature vectors and text feature statistics into standardized format data as training samples, which are then input into the initial prediction model. The training samples are used to perform machine learning training on the initial prediction model to obtain the target prediction model. The target prediction model stores the target correspondence between text feature vectors, text feature statistics, and credit risk probability values.

[0077] 203. Obtain text feature vectors and text feature statistics of the target text.

[0078] A text vector algorithm is used to convert the target text's text data into a text feature vector of the target text, and the target text data is processed to obtain text feature statistics of the target text. The text vector algorithm may be an embedding algorithm, specifically a word2vec model or a BERT model, to convert the target text data into a text feature vector of the target text. Other algorithms that achieve the same or similar effects may also be used to convert text into vectors, and the specifics are not limited here.

[0079] Specifically, the target text's corresponding industry type is determined based on a pre-established industry dictionary based on expert prior knowledge. This dictionary includes the sender's industry type and corresponding keywords. For example, if the target text contains "fenqile," it's considered to belong to the lending industry; if it contains "XX Bank," it's considered to belong to the banking industry, and so on.

[0080] The credit risk type corresponding to the target text is obtained from a pre-established risk type vocabulary based on expert prior knowledge. The risk type vocabulary is formed by applying a new word discovery algorithm to a risk type seed vocabulary based on expert prior knowledge. Specifically, this involves first establishing a risk type seed vocabulary based on expert prior knowledge, then expanding this seed vocabulary using a machine learning algorithm to form a more comprehensive risk type vocabulary. This expansion process using machine learning algorithms requires manual setting of corpus labels and corpus similarity thresholds to ensure the completeness and accuracy of the resulting risk type vocabulary.

[0081] The industry type and risk type corresponding to the historical text are input into the credit risk scoring model to obtain the industry type score and risk type score corresponding to the historical text. The credit risk scoring model here is different from the above-mentioned target prediction model. The credit risk scoring model stores a credit risk scoring system based on expert prior knowledge, which is used to score the industry type and risk type corresponding to the historical text obtained in the above steps.

[0082] In addition, some text data in the historical text does not need to be classified and scored. By summarizing and organizing this part of the text data, other feature statistical values that can be used for the target prediction model can be obtained, and the industry type score, risk type score and other feature statistical values can be determined as the text feature statistical values of the historical text.

[0083] 204. Input the text feature vector and text statistical feature value of the target text into the target prediction model to obtain the credit risk probability prediction value output by the target prediction model according to the target corresponding relationship.

[0084] Since each sender may correspond to multiple target texts, there may be information overlap or data overlap in the text feature vectors and text feature statistics of one or more target texts. However, when making predictions through the target prediction model, only one data value or vector of one data dimension needs to be input into the target prediction model. Therefore, it is necessary to first classify and fuse the text feature vectors and text feature statistics of the target text according to the sender of the target text to form standardized format data as training samples, which are then input into the initial prediction model. Among them, the classification and fusion specifically includes classifying the text feature vectors and text feature statistics of the same type in the target texts of the same sender and fusing the text feature vectors and text feature statistics of the same type into standardized format data as the input of the target prediction model, thereby obtaining the credit risk probability prediction value output by the target prediction model based on the target correspondence relationship. The credit risk probability prediction value is used to represent the credit risk of the target text.

[0085] In this embodiment, machine learning training is performed on historical texts in advance to establish a credit risk prediction model, and the text feature vector and text feature statistics of the target text are obtained. The text feature statistics of the target text are obtained based on expert prior knowledge, and the text feature vector of the target text is obtained using a text vector algorithm. These two types of data are input into the pre-established credit risk prediction model to obtain the credit risk probability prediction value corresponding to the target text. The operation is simple and the accuracy is high.

[0086] The following describes a credit risk prediction system provided by this application. Figure 3 , an implementation of a credit risk prediction system provided by this application includes:

[0087] An acquisition unit 301 is configured to acquire a target prediction model, wherein the target prediction model is obtained by performing machine learning training on an initial prediction model using text feature vectors and text feature statistics of historical texts and credit risk probability values. The credit risk probability values are obtained by analyzing the text feature vectors and text feature statistics of the historical texts. The target prediction model stores a target correspondence between the text feature vectors and text feature statistics and the credit risk probability values.

[0088] The acquisition unit 301 is further configured to acquire a text feature vector and text feature statistics of the target text;

[0089] The input unit 302 is used to input the text feature vector and text statistical feature value of the target text into the target prediction model to obtain the credit risk probability prediction value output by the target prediction model according to the target correspondence relationship. The credit risk probability prediction value is used to represent the credit risk of the target text.

[0090] The acquisition unit 301 is specifically configured to acquire a text feature vector and text feature statistics of a historical text and a credit risk probability value corresponding to the historical text;

[0091] The text feature vectors and text feature statistics of the historical text and the credit risk probability values corresponding to the historical text are input into the initial prediction model as training samples. The initial prediction model is trained by machine learning using the training samples to obtain a target prediction model. The target prediction model stores the target correspondence between the text feature vectors and text feature statistics and the credit risk probability values.

[0092] The acquisition unit 301 is further configured to convert the text data of the historical text into a text feature vector of the historical text using a text vector algorithm;

[0093] The text data of the historical text is processed to obtain text feature statistics of the historical text.

[0094] The acquisition unit 301 is further configured to establish an industry dictionary based on expert prior knowledge, wherein the industry dictionary includes the industry type of the sender of the historical text and industry characteristic keywords corresponding to the industry type;

[0095] Establish a risk type vocabulary based on expert prior knowledge. The risk type vocabulary is formed by expanding the risk type seed vocabulary based on expert prior knowledge using a new word discovery algorithm.

[0096] Classifying the first text data of the historical text according to the industry dictionary to obtain the industry type corresponding to the first text data of the historical text;

[0097] Classifying the second text data of the historical text according to the risk type vocabulary to obtain the risk type corresponding to the second text data of the historical text;

[0098] Input the industry type and risk type into the credit risk scoring model to obtain the industry type score and risk type score corresponding to the historical text. The credit risk scoring model stores a credit risk scoring system based on expert prior knowledge.

[0099] Arranging the third text data of the historical text to obtain other characteristic statistical values of the historical text;

[0100] The industry type score, risk type score and other feature statistics are determined as text feature statistics of the historical text.

[0101] The input unit 302 is specifically configured to classify and fuse text feature vectors and text feature statistics of historical texts according to senders of the historical texts to form standardized format data as training samples, and input the data into the initial prediction model.

[0102] The functions and processes performed by each unit in the credit risk prediction system of this embodiment are similar to those of the aforementioned Figures 1 to 2 The functions and processes performed by the credit risk prediction system are similar and will not be repeated here.

[0103] Figure 4 4 is a schematic diagram of the structure of the credit risk prediction device provided in this application. The credit risk prediction device 400 may include one or more central processing units (CPU) 401 and a memory 405. The memory 405 stores one or more applications or data.

[0104] Memory 405 can be volatile or persistent storage. The program stored in memory 405 can include one or more modules, each of which can include a series of instruction operations for the credit risk prediction system. Furthermore, central processing unit 401 can be configured to communicate with memory 405 and execute the series of instruction operations in memory 405 on credit risk prediction system 400.

[0105] The credit risk prediction device 400 may also include one or more power supplies 402, one or more wired or wireless network interfaces 403, one or more input and output interfaces 404, and / or one or more operating systems, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, etc.

[0106] The CPU 401 can execute the aforementioned Figures 1 to 2 The operations performed by the credit risk prediction system in the illustrated embodiment will not be described in detail here.

[0107] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0108] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0109] The units described as separate components may or may not be physically separate, and 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 network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0110] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0111] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially 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, and the computer software product is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, read-only memory), random access memory (RAM, random access memory), disk or optical disk, and other media that can store program code.

Claims

1. A credit risk prediction method, characterized in that: include: Obtain a target prediction model, wherein the target prediction model is obtained by performing machine learning training on the initial prediction model using the text feature vector and text feature statistics of the historical text and the credit risk probability value corresponding to the historical text as inputs of the initial prediction model, the credit risk probability value corresponding to the historical text is obtained by analyzing the text feature vector and text feature statistics of the historical text, and the target prediction model stores a target correspondence between the text feature vector and text feature statistics and the credit risk probability value; the text feature statistics of the historical text are obtained based on expert prior knowledge; the text feature statistics of the historical text include an industry type score and a risk type score corresponding to the historical text, and the industry type score and the risk type score corresponding to the historical text are obtained by using the industry type and risk type corresponding to the historical text as inputs of the credit risk scoring model; Obtaining a text feature vector and text feature statistics of a target text; the text feature statistics of the target text are obtained based on expert prior knowledge; The text feature vector and text statistical feature value of the target text are input into the target prediction model to obtain the credit risk probability prediction value output by the target prediction model according to the target correspondence relationship, and the credit risk probability prediction value is used to represent the credit risk of the target text.

2. The credit risk prediction method according to claim 1, characterized in that: The step of obtaining the text feature vector and text feature statistics of the historical text includes: Using a text vector algorithm to convert the text data of the historical text into a text feature vector of the historical text; The text data of the historical text is processed to obtain text feature statistics of the historical text.

3. The credit risk prediction method according to claim 2, characterized in that: The processing of the text data of the historical text to obtain text feature statistics of the historical text includes: Establishing an industry dictionary based on expert prior knowledge, wherein the industry dictionary includes the industry type of the sender of the historical text and industry characteristic keywords corresponding to the industry type; Establishing a risk type vocabulary based on expert prior knowledge, wherein the risk type vocabulary is formed by expanding the risk type seed vocabulary based on expert prior knowledge using a new word discovery algorithm; Classifying the first text data of the historical text according to the industry dictionary to obtain the industry type corresponding to the first text data of the historical text; classifying the second text data of the historical text according to the risk type vocabulary to obtain a risk type corresponding to the second text data of the historical text; Inputting the industry type and the risk type into a credit risk scoring model to obtain an industry type score and a risk type score corresponding to the historical text, wherein the credit risk scoring model stores a credit risk scoring system based on expert prior knowledge; sorting the third text data of the historical text to obtain other characteristic statistical values of the historical text; The industry type score, the risk type score, and the other feature statistical values are determined as text feature statistical values of the historical text.

4. The credit risk prediction method according to claim 1, characterized in that: The method of using the text feature vector and text feature statistics of the historical text and the credit risk probability value corresponding to the historical text as inputs of the initial prediction model includes: The text feature vectors and text feature statistics of the historical text are classified and fused according to the sender of the historical text to form standardized format data as training samples, which are input into the initial prediction model.

5. A credit risk prediction system, characterized in that: include: An acquisition unit is used to acquire a target prediction model, wherein the target prediction model is obtained by using the text feature vector, text feature statistics, and credit risk probability value of the historical text as inputs of the initial prediction model and performing machine learning training on the initial prediction model, the credit risk probability value is obtained by analyzing the text feature vector and text feature statistics of the historical text, and the target prediction model stores a target correspondence between the text feature vector, text feature statistics, and credit risk probability value; the text feature statistics of the historical text are obtained based on expert prior knowledge; the text feature statistics of the historical text include an industry type score and a risk type score corresponding to the historical text, and the industry type score and the risk type score corresponding to the historical text are obtained by using the industry type and risk type corresponding to the historical text as inputs of the credit risk scoring model; The acquisition unit is further configured to acquire a text feature vector and text feature statistics of the target text; the text feature statistics of the target text are acquired based on expert prior knowledge; An input unit is used to input the text feature vector and text statistical feature value of the target text into a target prediction model to obtain a credit risk probability prediction value output by the target prediction model according to the target correspondence relationship, and the credit risk probability prediction value is used to represent the credit risk of the target text.

6. The credit risk prediction system according to claim 5, characterized in that: The acquisition unit is further configured to convert the text data of the historical text into a text feature vector of the historical text using a text vector algorithm; The text data of the historical text is processed to obtain text feature statistics of the historical text.

7. A credit risk prediction device, characterized in that: include: CPU, memory and input / output interfaces; The memory is a transient storage memory or a persistent storage memory; The central processing unit is configured to communicate with the memory and execute instructions in the memory to perform the method according to any one of claims 1 to 4.

8. A computer-readable storage medium comprising instructions, which, when executed on a computer, enable the computer to perform the method according to any one of claims 1 to 4.

Images