Method, device, electronic equipment and storage medium for predicting corporate illegal fund-raising risks

Abstract

The embodiment of the present invention relates to a method, device, electronic device and storage medium for predicting the risk of illegal fund-raising of an enterprise. The method comprises: obtaining enterprise data of a target enterprise to be predicted for the risk of illegal fund-raising; extracting features from the enterprise data to obtain enterprise features of the target enterprise; inputting the enterprise features into at least one trained risk prediction model for illegal fund-raising to obtain the risk probability of illegal fund-raising of at least one target enterprise; and determining whether the target enterprise has the risk of illegal fund-raising according to the risk probability of illegal fund-raising of at least one target enterprise. Thus, the accuracy of the prediction result of predicting the risk of illegal fund-raising of an enterprise can be improved.

Description

Technical Field

[0001] Embodiments of the present invention relate to the field of financial security technology, and in particular to a method, device, electronic device and storage medium for predicting the risk of illegal fund-raising by enterprises. Background Art

[0002] Illegal fund-raising refers to the act of an organization or individual raising funds from the public by issuing stocks, bonds, lottery tickets, investment fund securities or other debt certificates without approval from the relevant departments in accordance with legal procedures, and promising to repay the principal and interest or give returns to investors in cash, in kind or in other ways within a certain period of time. This behavior seriously interferes with the normal economic and financial order, causing participants to suffer economic losses and even put their lives in trouble. It is very likely to cause social instability and a large number of social security problems, and even cause social unrest in local areas. It is a criminal activity. Therefore, it is of great significance to predict whether an enterprise has the risk of illegal fund-raising and to intervene and deal with enterprises with higher risks of illegal fund-raising in a timely manner to safeguard the public interest, financial order and social stability.

[0003] At present, in the supervision of corporate financial security, the risk prediction of illegal fund-raising by enterprises is mainly based on the previous work experience, financial policies and market environment of professionals. The risk indicators of enterprises are calculated from different dimensions to discover the risk information of illegal fund-raising by enterprises.

[0004] However, this method mainly relies on the personal experience and ability of professionals, with large subjective factors and inconsistent standards. Therefore, this method has poor predictive ability for the risk of illegal fundraising by enterprises, and the accuracy of the prediction results is questionable. Summary of the invention

[0005] In view of this, in order to solve the technical problem that the existing methods have poor prediction ability for the risk of illegal fund-raising by enterprises and the accuracy of the prediction results is questionable, the embodiments of the present invention provide a method, device, electronic device and storage medium for predicting the risk of illegal fund-raising by enterprises.

[0006] In a first aspect, an embodiment of the present invention provides a method for predicting the risk of illegal fund-raising by an enterprise, the method comprising:

[0007] Obtaining enterprise data of target enterprises for illegal fund-raising risk prediction;

[0008] Extracting features from the enterprise data to obtain enterprise features of the target enterprise;

[0009] Inputting the enterprise characteristics into at least one trained illegal fund-raising risk prediction model to obtain the risk probability of illegal fund-raising of at least one target enterprise;

[0010] Whether the target enterprise has a risk of illegal fund-raising is determined based on the risk probability of illegal fund-raising of at least one of the target enterprises.

[0011] In a possible implementation, inputting the enterprise characteristics into at least one trained illegal fund-raising risk prediction model to obtain the risk probability of illegal fund-raising of at least one target enterprise includes:

[0012] Inputting the enterprise characteristics into two or more trained illegal fund-raising risk prediction models respectively to obtain the risk probabilities of illegal fund-raising of the two or more target enterprises;

[0013] The determining whether the target enterprise has the risk of illegal fund-raising according to the risk probability of illegal fund-raising of at least one target enterprise comprises:

[0014] Determine the average value of the risk probability of illegal fund-raising of two or more target enterprises;

[0015] If the average value is greater than or equal to a set threshold, it is determined that the target enterprise has a risk of illegal fund-raising; if the average value is less than the set threshold, it is determined that the target enterprise does not have a risk of illegal fund-raising.

[0016] In a possible implementation, inputting the enterprise characteristics into at least one trained illegal fund-raising risk prediction model to obtain the risk probability of illegal fund-raising of at least one target enterprise includes:

[0017] Inputting the enterprise characteristics into two or more trained illegal fund-raising risk prediction models respectively to obtain the risk probabilities of illegal fund-raising of the two or more target enterprises;

[0018] Constructing a feature vector using the risk probability of illegal fund-raising of two or more target enterprises;

[0019] Inputting the feature vector into a trained logistic regression model to obtain the final risk probability of illegal fund-raising of the target enterprise; the logistic regression model is trained based on two or more trained illegal fund-raising risk prediction models;

[0020] The determining whether the target enterprise has the risk of illegal fund-raising according to the risk probability of illegal fund-raising of at least one target enterprise comprises:

[0021] Comparing the risk probability of illegal fund-raising by the ultimate target enterprise with a set threshold;

[0022] If the risk probability of illegal fundraising of the final target enterprise is greater than or equal to the set threshold, it is determined that the target enterprise has the risk of illegal fundraising; if the risk probability of illegal fundraising of the final target enterprise is less than the set threshold, it is determined that the target enterprise does not have the risk of illegal fundraising.

[0023] In a possible implementation, the illegal fund-raising risk prediction model is trained through the following process:

[0024] A plurality of enterprise data are obtained as training data and classified into a training data set, and an illegal fund-raising mark is used as a data label of the training data in the training data set, wherein the illegal fund-raising mark is used to indicate whether the enterprise has been filed for investigation due to illegal fund-raising;

[0025] Performing feature extraction on multiple pieces of training data in the training data set to obtain a training feature matrix;

[0026] The training feature matrix and the data labels corresponding to each training feature in the training feature matrix are trained using a machine learning algorithm to obtain an illegal fund-raising risk prediction model.

[0027] In a possible implementation, after extracting features from the plurality of training data in the training data set to obtain a training feature matrix, the method further includes:

[0028] Feature encoding and missing value filling are performed on the training features in the training feature matrix.

[0029] In a possible implementation, filling missing values ​​of the training features in the training feature matrix includes:

[0030] Filling missing values ​​in the training feature matrix with set values; or,

[0031] Fill the missing values ​​in the training feature matrix with the average value of the column where the missing values ​​are located; or,

[0032] Fill the missing values ​​in the training feature matrix with the median of the column where the missing values ​​are located; or,

[0033] The missing values ​​in the training feature matrix are filled with the mode of the column where the missing values ​​are located.

[0034] In a second aspect, an embodiment of the present invention provides a device for predicting the risk of illegal fund-raising by an enterprise, the device comprising:

[0035] A data acquisition module, used to acquire enterprise data of target enterprises for illegal fund-raising risk prediction;

[0036] A feature extraction module, used to extract features from the enterprise data to obtain enterprise features of the target enterprise;

[0037] A model prediction module, used for inputting the enterprise characteristics into at least one trained illegal fund-raising risk prediction model to obtain the risk probability of illegal fund-raising of at least one of the target enterprises;

[0038] The determination module is used to determine whether the target enterprise has the risk of illegal fund-raising based on the risk probability of illegal fund-raising of at least one of the target enterprises.

[0039] In a possible implementation, the model prediction module is specifically used to:

[0040] Inputting the enterprise characteristics into two or more trained illegal fund-raising risk prediction models respectively to obtain the risk probabilities of illegal fund-raising of the two or more target enterprises;

[0041] The determination module is specifically used for:

[0042] Determine the average value of the risk probability of illegal fund-raising of two or more target enterprises;

[0043] If the average value is greater than or equal to a set threshold, it is determined that the target enterprise has a risk of illegal fund-raising; if the average value is less than the set threshold, it is determined that the target enterprise does not have a risk of illegal fund-raising.

[0044] In a possible implementation, the model prediction module includes:

[0045] The first prediction submodule is used to input the enterprise characteristics into two or more trained illegal fund-raising risk prediction models respectively to obtain the risk probability of illegal fund-raising of the two or more target enterprises;

[0046] A construction submodule, used to construct a feature vector using the risk probabilities of illegal fund-raising of two or more target enterprises;

[0047] A second prediction submodule is used to input the feature vector into a trained logistic regression model to obtain a final risk probability of illegal fund-raising of the target enterprise; the logistic regression model is trained based on two or more trained illegal fund-raising risk prediction models;

[0048] The determination module is specifically used for:

[0049] Comparing the risk probability of illegal fund-raising by the ultimate target enterprise with a set threshold;

[0050] If the risk probability of illegal fundraising of the final target enterprise is greater than or equal to the set threshold, it is determined that the target enterprise has the risk of illegal fundraising; if the risk probability of illegal fundraising of the final target enterprise is less than the set threshold, it is determined that the target enterprise does not have the risk of illegal fundraising.

[0051] In a possible implementation, the device further includes:

[0052] A model training module, used to obtain multiple pieces of enterprise data as training data and classify them into a training data set, and to use an illegal fund-raising mark as a data label for the training data in the training data set, wherein the illegal fund-raising mark is used to indicate whether the enterprise has been filed for investigation due to illegal fund-raising;

[0053] Performing feature extraction on multiple pieces of training data in the training data set to obtain a training feature matrix;

[0054] The training feature matrix and the data labels corresponding to each training feature in the training feature matrix are trained using a machine learning algorithm to obtain an illegal fund-raising risk prediction model.

[0055] In a possible implementation, the device further includes:

[0056] A feature encoding module, used for performing feature encoding on the training features in the training feature matrix;

[0057] A filling module is used to fill missing values ​​of the training features in the training feature matrix.

[0058] In a possible implementation manner, the filling module is specifically used for:

[0059] Filling missing values ​​in the training feature matrix with set values; or,

[0060] Fill the missing values ​​in the training feature matrix with the average value of the column where the missing values ​​are located; or,

[0061] Fill the missing values ​​in the training feature matrix with the median of the column where the missing values ​​are located; or,

[0062] The missing values ​​in the training feature matrix are filled with the mode of the column where the missing values ​​are located.

[0063] In a third aspect, an embodiment of the present invention provides an electronic device comprising: a processor and a memory, wherein the processor is used to execute a corporate illegal fund-raising risk prediction program stored in the memory to implement the corporate illegal fund-raising risk prediction method described in any one of the first aspects.

[0064] In a fourth aspect, an embodiment of the present invention provides a storage medium storing one or more programs, and the one or more programs can be executed by one or more processors to implement the enterprise illegal fund-raising risk prediction method described in any one of the first aspects.

[0065] The technical solution provided by the embodiment of the present invention obtains the enterprise data of the target enterprise to be predicted for the risk of illegal fund-raising, performs feature extraction on the enterprise data, obtains the enterprise characteristics of the target enterprise, inputs the enterprise characteristics into at least one trained illegal fund-raising risk prediction model, obtains the risk probability of illegal fund-raising of at least one target enterprise, and determines whether the target enterprise has the risk of illegal fund-raising based on the risk probability of illegal fund-raising of at least one target enterprise. This can achieve direct prediction of the risk of illegal fund-raising of the enterprise through the illegal fund-raising risk prediction model. Compared with the prior art of performing risk assessment based on risk assessment indicators of different dimensions, this can avoid the problems of subjectivity and inconsistent standards, thereby improving the accuracy of the prediction results for predicting the risk of illegal fund-raising of the enterprise.

[0066] At the same time, in the application, for different business scenarios such as different types and methods of illegal fundraising, different illegal fundraising risk prediction models can be trained using the embodiments of the present invention to adapt to different businesses. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 A flow chart of an embodiment of the training process of an illegal fund-raising risk prediction model provided by an embodiment of the present invention;

[0068] Figure 2 A flow chart of an embodiment of a method for predicting the risk of illegal fund-raising by an enterprise provided by an embodiment of the present invention;

[0069] Figure 3 A block diagram of an embodiment of a device for predicting the risk of illegal fund-raising in an enterprise provided by an embodiment of the present invention;

[0070] Figure 4 A schematic diagram of the structure of an electronic device for predicting the risk of illegal fund-raising in an enterprise provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0071] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0072] To facilitate understanding of the enterprise illegal fund-raising risk prediction method provided in the embodiment of the present invention, the following first describes the training process of the illegal fund-raising risk prediction model involved in the embodiment of the present invention with reference to the accompanying drawings and specific embodiments:

[0073] See also Figure 1 , is a flow chart of an embodiment of the training process of an illegal fund-raising risk prediction model provided by an embodiment of the present invention, comprising the following steps:

[0074] Step 101: obtain multiple pieces of enterprise data as training data and classify them into a training data set, and use illegal fund-raising marks as data labels for the training data in the training data set.

[0075] In the embodiment of the present invention, the above-mentioned enterprise data includes but is not limited to: basic enterprise information, public annual report information, tax information, change information, public opinion information, other information, etc. The above-mentioned illegal fund-raising mark is used to indicate whether the enterprise has been investigated for illegal fund-raising. Optionally, when the illegal fund-raising mark is 1, it indicates that the enterprise has been investigated for illegal fund-raising, and when the illegal fund-raising mark is 0, it indicates that the enterprise has not been investigated for illegal fund-raising.

[0076] Among them, the basic information of the enterprise includes but is not limited to: enterprise registration number, administrative division code of the registration authority, industry category code, industry subcategory code, enterprise type, enterprise subcategory, establishment date, approval date, business start time, business end time, business scope, business address, business status, registered capital, paid-in capital, organization code, number of employees, number of partners, number of executives, etc.

[0077] Public annual report information includes, but is not limited to: company registration number, annual report year, total capital, business operating status, whether there has been any shareholder equity transfer in the limited liability company this year, whether there is any external guarantee, whether there is a website or online store, whether the company has investment information or purchased equity in other companies, whether the number of employees is disclosed, the number of employees, the number of operators who are college graduates, the number of operators who are retired soldiers, the number of operators who are disabled, the number of laid-off workers, the number of employees who are college graduates, the number of employees who are retired soldiers, the number of employees who are disabled, the number of laid-off workers, etc.

[0078] Tax information includes but is not limited to: company registration number, start time, end time, tax type, tax item, tax calculation basis, tax rate, deduction amount, tax amount, etc.

[0079] Change information includes but is not limited to: company registration number, change type, information before change, information after change, change date, etc.

[0080] Public opinion information includes but is not limited to: company registration number, public opinion content, release date, etc.

[0081] Other information includes but is not limited to: company registration number, number of judicial documents, number of registered trademarks, number of patents, etc.

[0082] As an embodiment, the above-mentioned basic enterprise information, public annual report information, tax information, change information, other information, illegal fundraising marks, etc. can be obtained from a third-party platform, and the above-mentioned public opinion information can be crawled from the Internet through crawler technology.

[0083] Step 102: extract features from multiple training data in the training data set to obtain a training feature matrix.

[0084] In the embodiment of the present invention, taking a piece of training data as an example, feature extraction of the training data can be divided into the following parts, and the feature values ​​extracted from each part are combined into a training feature and included in the training feature matrix. The feature extraction of each part is described below:

[0085] (1) Feature extraction of basic enterprise information

[0086] As an embodiment, features extracted based on basic enterprise information include but are not limited to: administrative division code of registration authority, industry category code, industry subcategory code, enterprise type, enterprise subcategory, number of days of establishment, number of days of approval, number of days in operation, number of remaining days in operation, total number of days in operation, business scope characteristics, business address, business status, registered capital, paid-in capital, organizational code, number of employees, number of partners, number of executives, the number of enterprises with the same industry category code in the training set, the number of enterprises with the same industry subcategory code in the training set, the number of enterprises with the same enterprise type in the training set, the number of enterprises with the same enterprise subcategory in the training set, the number of enterprises with the same business address in the training set, intersection characteristics of industry category code and business status, intersection characteristics of enterprise type code and business status, intersection characteristics of industry category code and enterprise type code, etc.

[0087] Among them, the characteristics of the enterprise's registration authority administrative division code, industry category code, industry subcategory code, enterprise type, enterprise subcategory, business address, business status, registered capital, paid-in capital, organizational code, number of employees, number of partners, number of executives, etc. can directly use the original data values ​​as feature values.

[0088] The number of days since the establishment of an enterprise is the difference between the current date and the establishment date, the number of days since the approval is the difference between the current date and the approval date, the number of days since the operation is the difference between the current date and the start time of the operation, the number of days remaining since the operation is the difference between the current date and the end time of the operation, and the total number of days since the operation is the difference between the end time of the operation and the start time of the operation. For example, assuming that the establishment date of an enterprise is September 29, 2007, the approval date is November 16, 2020, the start time of the operation is September 29, 2007, the end time of the operation is September 28, 2037, and the current time is December 1, 2020, then the characteristic values ​​of the establishment days, approval days, operation days, remaining days of the operation, and total days of the operation of the enterprise are 4812, 15, 4812, 6145, and 10957, respectively.

[0089] The number of enterprises with the same industry category code in the training set of an enterprise is the total number of enterprises with the same industry category code as the enterprise in the training data set, the number of enterprises with the same industry subcategory code in the training set is the total number of enterprises with the same industry subcategory code as the enterprise in the training data set, the number of enterprises with the same enterprise type in the training set is the total number of enterprises with the same enterprise type as the enterprise in the training data set, the number of enterprises with the same enterprise subcategory in the training set is the total number of enterprises with the same enterprise subcategory as the enterprise in the training data set, and the number of enterprises with the same business address in the training set is the total number of enterprises with the same business address as the enterprise in the training data set. For example, assuming that the industry category code of an enterprise is L, and there are 2301 enterprises with the industry category code L in the training data set, the characteristic value of the number of enterprises with the same industry category code in the training set of the enterprise is 2301.

[0090] The cross-feature of the enterprise's industry category code and operating status is the concatenation of the industry category code and the operating status. For example, if the industry category code of an enterprise is L and the operating status is opened, the characteristic value of the cross-feature of the enterprise's industry category code and operating status is L opened.

[0091] The business scope feature of an enterprise is a text feature. In the application, the TF-IDF method, text embedded representation method, etc. can be used to extract the business scope feature. Specifically, the TF-IDF method is used to extract the business scope feature, including: calculating the word frequency-inverse text frequency of the word as the weight of the word, and constructing the word vector of the text as the business scope feature of the enterprise. The text embedded representation method is used to extract the business scope feature, including: using the word2vec model to calculate the word vector, and then calculating the average value of the word vector of the text to obtain the vector representation of the text as the business scope feature of the enterprise.

[0092] (2) Feature extraction of the company’s public annual report information

[0093] As an embodiment, the features extracted based on the public annual report information include but are not limited to: the latest year of the company's annual report, the company's operating status in the latest year, whether there is any shareholder equity transfer in the limited liability company in the public annual report this year, whether the company has any external guarantees in the public annual report, whether the company has a website or online store in the public annual report, whether the company has investment information or purchases other company shares in the public annual report, whether the number of employees in the latest year is public, and the latest year number, total value, mean, maximum value, minimum value and standard deviation of the total amount of funds, number of employees, number of college graduates, number of operators, number of retired soldiers, number of disabled operators, number of laid-off workers, number of college graduates, number of retired soldiers, number of disabled employees, and number of laid-off workers.

[0094] The latest year of the enterprise annual report is the year value of the enterprise's latest annual report. For example, if the enterprise's latest annual report is 2018, then the characteristic value of the latest annual report of the enterprise annual report is 2018.

[0095] The business operating status of the latest year is the business operating status value in the latest annual report.

[0096] Whether a limited liability company has a shareholder equity transfer in the current year in the public annual report, whether the company has external guarantees in the public annual report, whether the company has a website or online store in the public annual report, and whether the company has investment information or purchases other company equity in the public annual report are obtained based on all the public annual reports of the company. Specifically, taking whether a limited liability company has a shareholder equity transfer in the current year in the public annual report as an example, if the value of whether a limited liability company has a shareholder equity transfer in the current year in the annual report of any year is 1, then the characteristic value of whether a limited liability company has a shareholder equity transfer in the current year in the public annual report is 1, indicating that the limited liability company has a shareholder equity transfer in the current year; if the value of whether a limited liability company has a shareholder equity transfer in the current year in the annual report of all years is 0, then the characteristic value of whether a limited liability company has a shareholder equity transfer in the current year in the public annual report is 0, indicating that the limited liability company has no shareholder equity transfer in the current year.

[0097] Whether the number of employees in the latest year is disclosed is determined based on the mark value of the number of employees in the latest annual report of the enterprise. Specifically, if the number of employees in the latest annual report of the enterprise has a mark value, the feature value of whether the number of employees in the latest year is disclosed is 1, indicating that the number of employees in the latest year has been disclosed; if the number of employees in the latest annual report of the enterprise does not have a mark value, the feature value of whether the number of employees in the latest year is disclosed is 0, indicating that the number of employees in the latest year has not been disclosed.

[0098] The latest year numbers of total funds, number of employees, number of university graduates, number of managers, number of retired soldiers, number of disabled managers, number of laid-off workers, number of university graduates, number of retired soldiers, number of disabled employees and number of laid-off workers are obtained based on the latest annual report of the enterprise, and the total value, mean, maximum, minimum value and standard deviation of each year are obtained based on all publicly available annual reports of the enterprise.

[0099] (3) Feature extraction of corporate tax information

[0100] As an embodiment, features extracted based on tax information include but are not limited to: the number of tax types, the number of tax items, the total number, maximum, minimum and average of tax bases, and the total number, maximum, minimum and average of tax amounts.

[0101] (4) Feature extraction of enterprise change information

[0102] As an embodiment, the features extracted based on the change information include but are not limited to: the number of enterprise change types and the number of changes of each type.

[0103] Among them, the above-mentioned enterprise change types include: name change, address change, legal representative change, registered capital change, business scope change, type change, shareholder and equity change, and other changes, a total of 8 categories.

[0104] For example, an enterprise has three changes in total, two of which are changes in shareholders and equity, and one is a change in registered capital. The characteristic value of the number of enterprise change types of this enterprise is 2, the characteristic value of the number of shareholder and equity changes is 2, and the characteristic value of the number of registered capital changes is 1.

[0105] (5) Feature extraction of corporate public opinion information

[0106] As an embodiment, features extracted based on public opinion information include but are not limited to: the number of days from the current time when corporate public opinion is released, the maximum value, the minimum value and the mean value, the total amount of corporate public opinion, the number of positive public opinions, the number of negative public opinions, the number of neutral public opinions, the proportion of positive public opinions, the proportion of negative public opinions, the proportion of neutral public opinions, etc.

[0107] For example, a company has 5 public opinions, and the release time of these 5 public opinions is 1, 3, 2, 2, and 3 days from the current time. Then the number of days from the release time of the company's public opinions to the current time is 3, the maximum value is 3, the minimum value is 1, and the average is 2.2.

[0108] The total amount of corporate public opinion is the amount of public opinion information of the company. For example, in the above example, the characteristic value of the total amount of corporate public opinion is 5.

[0109] For the number of positive public opinions, the number of negative public opinions, the number of neutral public opinions, the proportion of positive public opinions, the proportion of negative public opinions, the proportion of neutral public opinions, etc., in the application, the text sentiment classification algorithm can be used to classify the sentiment tendency of each public opinion information, and the sentiment tendency of each public opinion information can be marked according to the classification results, including "positive", "negative", and "neutral". In this way, the number of positive public opinions, the number of negative public opinions, the number of neutral public opinions, the proportion of positive public opinions, the proportion of negative public opinions, the proportion of neutral public opinions, etc. can be counted according to the sentiment tendency marking of each public opinion information.

[0110] (6) Feature extraction of other enterprise information

[0111] As an embodiment, features extracted based on other information include but are not limited to: the number of corporate judicial documents, the number of registered trademarks, the number of patents, etc.

[0112] Step 103: Use a machine learning algorithm to train the training feature matrix and the data labels corresponding to each training feature in the training feature matrix to obtain an illegal fund-raising risk prediction model.

[0113] The data label corresponding to the above training features refers to the data label of the training data corresponding to the training features, that is, the above illegal fund-raising mark.

[0114] The above machine learning algorithms include but are not limited to: logistic regression, decision tree, random forest, xgboost, lightGBM, catboost, etc.

[0115] In the embodiment of the present invention, the above-mentioned illegal fund-raising risk prediction model uses enterprise characteristics as input values ​​and uses the risk probability of illegal fund-raising by the enterprise as output values.

[0116] In addition, further, in the application, for the lightGBM and catboost models, the training feature matrix without feature encoding and missing value filling can be input for model training, while for other models, the training feature matrix with feature encoding and missing value filling needs to be input for model training. Therefore, after executing step 103, the training features in the training feature matrix can be feature encoded and the missing values ​​can be filled.

[0117] Among them, feature encoding of training features includes: numerical encoding of feature values ​​of string type in training features, and one-hot encoding of feature values ​​of category type in training features. Furthermore, the numerical encoding method includes: sorting all possible values ​​of the feature value in character order, and then, according to the sorting result, encoding and mapping numbers starting from 0 and in ascending order, and finally replacing the original feature value with the number obtained by the encoding mapping. For example, the industry category code is a feature value of string type, with a total of 20 possible values. The encoding mapping relationship between characters and numbers obtained by the above encoding method is shown in Table 1 below:

[0118] Table 1

[0119]

[0120]

[0121] Filling missing values ​​in training features includes: filling missing values ​​in training features with set values, such as 0; or filling missing values ​​in training features with the average value of the column where the missing values ​​are located; or filling missing values ​​in training features with the median of the column where the missing values ​​are located; or filling missing values ​​in training features with the mode of the column where the missing values ​​are located.

[0122] pass Figure 1 The process shown realizes the training of an illegal fund-raising risk prediction model that can be used to predict the risk probability of illegal fund-raising by enterprises.

[0123] In addition, in the application, the above Figure 1 In the process shown, during each execution, the enterprise data, that is, the training data set, may be different, and the machine learning algorithm used may also be different. In this way, multiple different illegal fund-raising risk prediction models can be trained that can be used to predict the risk probability of illegal fund-raising by enterprises.

[0124] The method for predicting the risk of illegal fund-raising by enterprises provided by the present invention is further explained below with reference to specific embodiments in conjunction with the accompanying drawings. The embodiments do not constitute a limitation on the embodiments of the present invention.

[0125] See also Figure 2 , is a flow chart of an embodiment of a method for predicting the risk of illegal fund-raising in an enterprise provided by an embodiment of the present invention. Figure 2 As shown, the process may include the following steps:

[0126] Step 201: Obtain enterprise data of a target enterprise for which illegal fund-raising risk prediction is to be performed.

[0127] For the convenience of description, in the embodiment of the present invention, the enterprise to be subjected to illegal fund-raising risk prediction is referred to as a target enterprise.

[0128] For details of the target company’s corporate data, please refer to the above Figure 1 The relevant descriptions in the process shown will not be repeated here.

[0129] Step 202: extract features from the enterprise data to obtain enterprise features of the target enterprise.

[0130] As for how to extract the characteristics of the enterprise data in step 202 to obtain the enterprise characteristics of the target enterprise, see Figure 1 The relevant descriptions in the process shown will not be repeated here.

[0131] In addition, by Figure 1 As can be seen from the description of the process shown, in some cases, the enterprise characteristics can also be coded and missing values ​​can be filled. Figure 2 In the process shown, when filling missing values ​​for enterprise characteristics, the missing values ​​in the enterprise characteristics can be filled with set values. As for how to encode enterprise characteristics, please refer to Figure 1 The relevant descriptions in the process shown will not be repeated here.

[0132] Step 203: Input the enterprise characteristics into at least one trained illegal fund-raising risk prediction model to obtain the risk probability of illegal fund-raising of at least one target enterprise.

[0133] Step 204: Determine whether a target enterprise has a risk of illegal fund-raising based on the risk probability of illegal fund-raising of at least one target enterprise.

[0134] The following is a unified description of step 203 and step 204:

[0135] As an embodiment, in step 203, the enterprise characteristics are respectively input into two or more trained illegal fund-raising risk prediction models to obtain the risk probabilities of illegal fund-raising of two or more target enterprises, wherein the greater the risk probability, the higher the possibility of illegal fund-raising of the target enterprise. Based on this, in step 204, the average value of the risk probabilities of illegal fund-raising of two or more target enterprises can be determined. If the average value is greater than or equal to the set threshold, it is determined that the target enterprise has the risk of illegal fund-raising. If the average value is less than the set threshold, it is determined that the target enterprise does not have the risk of illegal fund-raising. Through this processing, the final prediction accuracy can be improved.

[0136] As another example, the Figure 1The two or more illegal fund-raising risk prediction models trained by the process shown are regarded as the first-level model, and the logistic regression model trained based on the first-level model is regarded as the second-level model. Based on this, in step 203, the enterprise characteristics are respectively input into each first-level model to obtain the risk probability of illegal fund-raising of two or more target enterprises. Further, a feature vector is constructed using the risk probability of illegal fund-raising of two or more target enterprises, and the feature vector is input into the trained second-level model, that is, the logistic regression model, to obtain the final risk probability of illegal fund-raising of the target enterprise. In step 204, the final risk probability of illegal fund-raising of the target enterprise is compared with the set threshold. If the final risk probability of illegal fund-raising of the target enterprise is greater than or equal to the set threshold, it is determined that the target enterprise has the risk of illegal fund-raising; if the final risk probability of illegal fund-raising of the target enterprise is less than the set threshold, it is determined that the target enterprise does not have the risk of illegal fund-raising. Through this processing, the final prediction accuracy can be improved.

[0137] In the application, the process of obtaining the second-level model based on the first-level model training includes: dividing the training data into N parts, taking N-1 parts as the training data set, and Figure 1 The process shown trains the first-level model, and then uses the first-level model to predict the probability of illegal fund-raising of the remaining 1 training data, repeating the above process N times until the probability of illegal fund-raising of each training data is obtained. After that, the probability of illegal fund-raising of each training data is formed into a feature matrix and input into the initial logistic regression algorithm model for training to obtain the second-level model.

[0138] The technical solution provided by the embodiment of the present invention obtains the enterprise data of the target enterprise to be predicted for the risk of illegal fund-raising, performs feature extraction on the enterprise data, obtains the enterprise characteristics of the target enterprise, inputs the enterprise characteristics into at least one trained illegal fund-raising risk prediction model, obtains the risk probability of illegal fund-raising of at least one target enterprise, and determines whether the target enterprise has the risk of illegal fund-raising based on the risk probability of illegal fund-raising of at least one target enterprise. This can achieve direct prediction of the risk of illegal fund-raising of the enterprise through the illegal fund-raising risk prediction model. Compared with the prior art of performing risk assessment based on risk assessment indicators of different dimensions, this can avoid the problems of subjectivity and inconsistent standards, thereby improving the accuracy of the prediction results for predicting the risk of illegal fund-raising of the enterprise.

[0139] At the same time, in the application, for different business scenarios such as different types and methods of illegal fundraising, different illegal fundraising risk prediction models can be trained respectively using the embodiments of the present invention, so as to adapt to the needs of different business scenarios and improve the reliability and scalability of risk prediction for illegal fundraising of enterprises.

[0140] See also Figure 3, is a block diagram of an embodiment of a device for predicting the risk of illegal fund-raising in an enterprise provided by an embodiment of the present invention. Figure 3 As shown, the device includes: a data acquisition module 31, a feature extraction module 32, a model prediction module 33 and a determination module 34.

[0141] The data acquisition module 31 is used to acquire the enterprise data of the target enterprise for which illegal fund-raising risk prediction is to be performed;

[0142] A feature extraction module 32 is used to extract features from the enterprise data to obtain enterprise features of the target enterprise;

[0143] The model prediction module 33 is used to input the enterprise characteristics into at least one trained illegal fund-raising risk prediction model to obtain the risk probability of illegal fund-raising of at least one target enterprise;

[0144] The determination module 34 is used to determine whether the target enterprise has the risk of illegal fund-raising according to the risk probability of illegal fund-raising of at least one of the target enterprises.

[0145] In a possible implementation, the model prediction module 33 is specifically used to:

[0146] Inputting the enterprise characteristics into two or more trained illegal fund-raising risk prediction models respectively to obtain the risk probabilities of illegal fund-raising of the two or more target enterprises;

[0147] The determination module 34 is specifically used for:

[0148] Determine the average value of the risk probability of illegal fund-raising of two or more target enterprises;

[0149] If the average value is greater than or equal to a set threshold, it is determined that the target enterprise has a risk of illegal fund-raising; if the average value is less than the set threshold, it is determined that the target enterprise does not have a risk of illegal fund-raising.

[0150] In a possible implementation, the model prediction module 33 includes ( Figure 3 Not shown):

[0151] The first prediction submodule is used to input the enterprise characteristics into two or more trained illegal fund-raising risk prediction models respectively to obtain the risk probability of illegal fund-raising of the two or more target enterprises;

[0152] A construction submodule, used to construct a feature vector using the risk probabilities of illegal fund-raising of two or more target enterprises;

[0153] A second prediction submodule is used to input the feature vector into a trained logistic regression model to obtain a final risk probability of illegal fund-raising of the target enterprise; the logistic regression model is trained based on two or more trained illegal fund-raising risk prediction models;

[0154] The determination module is specifically used for:

[0155] Comparing the risk probability of illegal fund-raising by the ultimate target enterprise with a set threshold;

[0156] If the risk probability of illegal fundraising of the final target enterprise is greater than or equal to the set threshold, it is determined that the target enterprise has the risk of illegal fundraising; if the risk probability of illegal fundraising of the final target enterprise is less than the set threshold, it is determined that the target enterprise does not have the risk of illegal fundraising.

[0157] In a possible implementation, the device further includes ( Figure 3 Not shown):

[0158] A model training module, used to obtain multiple pieces of enterprise data as training data and classify them into a training data set, and to use an illegal fund-raising mark as a data label for the training data in the training data set, wherein the illegal fund-raising mark is used to indicate whether the enterprise has been filed for investigation due to illegal fund-raising;

[0159] Performing feature extraction on multiple pieces of training data in the training data set to obtain a training feature matrix;

[0160] The training feature matrix and the data labels corresponding to each training feature in the training feature matrix are trained using a machine learning algorithm to obtain an illegal fund-raising risk prediction model.

[0161] In a possible implementation, the device further includes ( Figure 3 Not shown):

[0162] A feature encoding module, used for performing feature encoding on the training features in the training feature matrix;

[0163] A filling module is used to fill missing values ​​of the training features in the training feature matrix.

[0164] In a possible implementation manner, the filling module is specifically used for:

[0165] Filling missing values ​​in the training feature matrix with set values; or,

[0166] Fill the missing values ​​in the training feature matrix with the average value of the column where the missing values ​​are located; or,

[0167] Fill the missing values ​​in the training feature matrix with the median of the column where the missing values ​​are located; or,

[0168] The missing values ​​in the training feature matrix are filled with the mode of the column where the missing values ​​are located.

[0169] Figure 4 A schematic diagram of the structure of an electronic device for predicting the risk of illegal fund-raising in an enterprise provided by an embodiment of the present invention. Figure 4 The electronic device 400 shown includes: at least one processor 401, a memory 402, at least one network interface 404 and other user interfaces 403. The various components in the electronic device 400 are coupled together via a bus system 405. It is understood that the bus system 405 is used to achieve connection and communication between these components. In addition to the data bus, the bus system 405 also includes a power bus, a control bus and a status signal bus. However, for the sake of clarity, the bus system 405 is not shown in FIG. Figure 4 Various buses are labeled as bus system 405 .

[0170] The user interface 403 may include a display, a keyboard, or a pointing device (eg, a mouse, a trackball, a touch pad, or a touch screen).

[0171] It can be understood that the memory 402 in the embodiment of the present invention can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memories. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory can be a random access memory (RAM), which is used as an external cache. By way of example but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM) and direct memory bus random access memory (DRRAM). The memory 402 described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0172] In some implementations, the memory 402 stores the following elements, executable units or data structures, or a subset thereof, or an extended set thereof: an operating system 4021 and application programs 4022 .

[0173] The operating system 4021 includes various system programs, such as a framework layer, a core library layer, a driver layer, etc., which are used to implement various basic services and process hardware-based tasks. The application 4022 includes various application programs, such as a media player (MediaPlayer), a browser (Browser), etc., which are used to implement various application services. The program for implementing the method of the embodiment of the present invention can be included in the application 4022.

[0174] In the embodiment of the present invention, by calling the program or instruction stored in the memory 402, specifically, the program or instruction stored in the application 4022, the processor 401 is used to execute the method steps provided by each method embodiment, for example, including:

[0175] Obtaining enterprise data of target enterprises for illegal fund-raising risk prediction;

[0176] Extracting features from the enterprise data to obtain enterprise features of the target enterprise;

[0177] Inputting the enterprise characteristics into at least one trained illegal fund-raising risk prediction model to obtain the risk probability of illegal fund-raising of at least one target enterprise;

[0178] Whether the target enterprise has a risk of illegal fund-raising is determined based on the risk probability of illegal fund-raising of at least one of the target enterprises.

[0179] The method disclosed in the above embodiment of the present invention can be applied to the processor 401, or implemented by the processor 401. The processor 401 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit in the processor 401 or the instruction in the form of software. The above processor 401 can be a general processor, a digital signal processor (Digital Signal Processor, DSP), an application specific integrated circuit (Application Specific Integrated Circuit, ASIC), a field programmable gate array (Field Programmable Gate Array, FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present invention can be implemented or executed. The general processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the embodiment of the present invention can be directly embodied as a hardware decoding processor to execute, or the hardware and software units in the decoding processor can be executed. The software unit can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory 402, and the processor 401 reads the information in the memory 402 and completes the steps of the above method in combination with its hardware.

[0180] It is understood that the embodiments described herein can be implemented by hardware, software, firmware, middleware, microcode or a combination thereof. For hardware implementation, the processing unit can be implemented in one or more application specific integrated circuits (ASIC), digital signal processors (DSP), digital signal processing devices (DSPDevice, DSPD), programmable logic devices (PLD), field programmable gate arrays (FPGA), general-purpose processors, controllers, microcontrollers, microprocessors, other electronic units for performing the functions described in the present application or a combination thereof.

[0181] For software implementation, the technology described herein can be implemented by a unit that performs the functions described herein. The software code can be stored in a memory and executed by a processor. The memory can be implemented in the processor or outside the processor.

[0182] The electronic device provided in this embodiment may be Figure 4 The electronic device shown in FIG. 1 may perform the following steps: Figure 1-2 All steps of the risk prediction method for illegal fund-raising in enterprises, and then achieve Figure 1-2 For details, please refer to the technical effect of the enterprise illegal fund-raising risk prediction method shown in Figure 1-2 For the sake of brevity, the relevant description is not repeated here.

[0183] The embodiment of the present invention also provides a storage medium (computer-readable storage medium). The storage medium here stores one or more programs. The storage medium may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a read-only memory, a flash memory, a hard disk or a solid-state drive; the memory may also include a combination of the above-mentioned types of memory.

[0184] When one or more programs in the storage medium can be executed by one or more processors, the above-mentioned method for predicting the risk of illegal fund-raising by enterprises executed on the electronic device side can be implemented.

[0185] The processor is used to execute the enterprise illegal fund-raising risk prediction program stored in the memory to implement the following steps of the enterprise illegal fund-raising risk prediction method executed on the electronic device side:

[0186] Obtaining enterprise data of target enterprises for illegal fund-raising risk prediction;

[0187] Extracting features from the enterprise data to obtain enterprise features of the target enterprise;

[0188] Inputting the enterprise characteristics into at least one trained illegal fund-raising risk prediction model to obtain the risk probability of illegal fund-raising of at least one target enterprise;

[0189] Whether the target enterprise has a risk of illegal fund-raising is determined based on the risk probability of illegal fund-raising of at least one of the target enterprises.

[0190] The professionals should further realize that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to the function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0191] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0192] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for predicting the risk of illegal fund-raising in an enterprise, characterized in that: The method comprises: Obtaining enterprise data of target enterprises for illegal fund-raising risk prediction; Extracting features from the enterprise data to obtain enterprise features of the target enterprise; Inputting the enterprise characteristics into at least one trained illegal fund-raising risk prediction model to obtain the risk probability of illegal fund-raising of at least one target enterprise; Determining whether the target enterprise has a risk of illegal fund-raising based on at least one risk probability of illegal fund-raising of the target enterprise; The step of inputting the enterprise characteristics into at least one trained illegal fund-raising risk prediction model to obtain the risk probability of illegal fund-raising of at least one target enterprise includes: Inputting the enterprise characteristics into two or more trained illegal fund-raising risk prediction models respectively to obtain the risk probabilities of illegal fund-raising of the two or more target enterprises; Constructing a feature vector using the risk probability of illegal fund-raising of two or more target enterprises; The feature vector is input into a trained logistic regression model to obtain the final risk probability of illegal fund-raising of the target enterprise; the logistic regression model is trained based on two or more trained illegal fund-raising risk prediction models.

2. The method according to claim 1, characterized in that The step of inputting the enterprise characteristics into at least one trained illegal fund-raising risk prediction model to obtain the risk probability of illegal fund-raising of at least one target enterprise includes: Inputting the enterprise characteristics into two or more trained illegal fund-raising risk prediction models respectively to obtain the risk probabilities of illegal fund-raising of the two or more target enterprises; The determining whether the target enterprise has the risk of illegal fund-raising according to the risk probability of illegal fund-raising of at least one target enterprise comprises: Determine the average value of the risk probability of illegal fund-raising of two or more target enterprises; If the average value is greater than or equal to a set threshold, it is determined that the target enterprise has a risk of illegal fund-raising; if the average value is less than the set threshold, it is determined that the target enterprise does not have a risk of illegal fund-raising.

3. The method according to claim 1, characterized in that The determining whether the target enterprise has the risk of illegal fund-raising according to the risk probability of illegal fund-raising of at least one target enterprise comprises: Comparing the risk probability of illegal fund-raising by the ultimate target enterprise with a set threshold; If the risk probability of illegal fundraising of the final target enterprise is greater than or equal to the set threshold, it is determined that the target enterprise has the risk of illegal fundraising; if the risk probability of illegal fundraising of the final target enterprise is less than the set threshold, it is determined that the target enterprise does not have the risk of illegal fundraising.

4. The method according to claim 1, characterized in that The illegal fund-raising risk prediction model is trained through the following process: A plurality of enterprise data are obtained as training data and classified into a training data set, and an illegal fund-raising mark is used as a data label of the training data in the training data set, wherein the illegal fund-raising mark is used to indicate whether the enterprise has been filed for investigation due to illegal fund-raising; Performing feature extraction on multiple pieces of training data in the training data set to obtain a training feature matrix; The training feature matrix and the data labels corresponding to each training feature in the training feature matrix are trained using a machine learning algorithm to obtain an illegal fund-raising risk prediction model.

5. The method according to claim 4, characterized in that After extracting features from the plurality of training data in the training data set to obtain a training feature matrix, the method further includes: Feature encoding and missing value filling are performed on the training features in the training feature matrix.

6. The method according to claim 5, characterized in that The filling missing values ​​of the training features in the training feature matrix includes: Filling missing values ​​in the training feature matrix with set values; or, Fill the missing values ​​in the training feature matrix with the average value of the column where the missing values ​​are located; or, Fill the missing values ​​in the training feature matrix with the median of the column where the missing values ​​are located; or, The missing values ​​in the training feature matrix are filled with the mode of the column where the missing values ​​are located.

7. A risk prediction device for illegal fund-raising in enterprises, characterized in that: The device comprises: A data acquisition module, used to acquire enterprise data of target enterprises for illegal fund-raising risk prediction; A feature extraction module, used to extract features from the enterprise data to obtain enterprise features of the target enterprise; A model prediction module, used for inputting the enterprise characteristics into at least one trained illegal fund-raising risk prediction model to obtain the risk probability of illegal fund-raising of at least one of the target enterprises; A determination module, configured to determine whether the target enterprise has a risk of illegal fund-raising according to a risk probability of illegal fund-raising of at least one of the target enterprises; The step of inputting the enterprise characteristics into at least one trained illegal fund-raising risk prediction model to obtain the risk probability of illegal fund-raising of at least one target enterprise includes: Inputting the enterprise characteristics into two or more trained illegal fund-raising risk prediction models respectively to obtain the risk probabilities of illegal fund-raising of the two or more target enterprises; Constructing a feature vector using the risk probability of illegal fund-raising of two or more target enterprises; The feature vector is input into a trained logistic regression model to obtain the final risk probability of illegal fund-raising of the target enterprise; the logistic regression model is trained based on two or more trained illegal fund-raising risk prediction models.

8. The device according to claim 7, characterized in that The device also includes: The model training module is used to obtain multiple enterprise data as training data and classify them into a training data set, and to use an illegal fundraising mark as a data label for the training data in the training data set, wherein the illegal fundraising mark is used to indicate whether the enterprise has been investigated for illegal fundraising; to perform feature extraction on the multiple training data in the training data set to obtain a training feature matrix; and to use a machine learning algorithm to train the training feature matrix and the data label corresponding to each training feature in the training feature matrix to obtain an illegal fundraising risk prediction model.

9. An electronic device, characterized in that: include: A processor and a memory, wherein the processor is used to execute the enterprise illegal fund-raising risk prediction program stored in the memory to implement the enterprise illegal fund-raising risk prediction method according to any one of claims 1 to 6.

10. A storage medium, characterized in that: The storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the enterprise illegal fund-raising risk prediction method according to any one of claims 1 to 6.

Images