Enterprise digital risk control methods and platforms based on big data

Abstract

The enterprise digital risk control method and platform based on big data include: obtaining multiple types of source data, preprocessing the source data, and storing the preprocessed source data in a distributed database. The fluctuation characteristics of each type of source data are extracted by Fourier transform. The fluctuation characteristics are used as the input of the trained convolutional neural network, the fluctuation characteristics are convolved through the convolution layer, and the feature data is extracted according to the convolution result, so as to calculate the feature vector corresponding to each type of source data from the feature data through the pooling layer. The fully connected layer in the convolutional neural network is replaced with a graph neural network, and the type graph matrix corresponding to each type of source data is calculated based on the feature vector and the graph neural network, so as to output the risk prediction result according to the type graph matrix through the graph neural network. This method improves the accuracy of sensitive information identification and strengthens the security of enterprise digital risk control through a deep learning algorithm.

Description

Technical Field

[0001] The present invention relates to the field of big data technology, and in particular to a digital enterprise risk control method and platform based on big data. Background Art

[0002] In recent years, the digital economy has flourished. While promoting the development of productivity and the transformation of production relations, enterprises have also experienced rapid development and new opportunities. However, in the process of digital transformation and upgrading, enterprises have also encountered some information and digital security risks, such as corporate data leakage incidents. As people's awareness of their own privacy and data security protection gradually increases, data security has also received more and more attention from the country and all walks of life. Enterprises will face risks once data leakage occurs.

[0003] At present, the traditional enterprise digital risk control method mainly screens and preprocesses the original data, then identifies the processed data to identify sensitive information data, and adds corresponding sensitive labels, and then performs risk control on the data with sensitive labels. However, this method can only perform risk control on sensitive information data identified by the system, and the system has a limited range of sensitive information identification. It can only identify sensitive information according to the set system rules, and it is difficult to accurately identify data outside the system rules, which can easily lead to indirect leakage of sensitive information.

[0004] Therefore, due to the limitations of sensitive information identification, the intelligence level of data risk control in existing enterprise digital risk control methods needs to be further improved, which is more likely to lead to indirect leakage of sensitive information. Summary of the invention

[0005] In order to solve the deficiencies in the prior art, the purpose of the present invention is to solve the above-mentioned defects and further propose a digital risk control method and platform for enterprises based on big data.

[0006] The present invention adopts the following technical solution.

[0007] The first aspect of the present invention discloses a digital risk control method for an enterprise based on big data, the method comprising:

[0008] Acquire various types of source data, preprocess the source data, and store the preprocessed source data in a distributed database, wherein the preprocessing includes data cleaning, deduplication, and standardization;

[0009] Extracting fluctuation characteristics of each type of source data by Fourier transform, wherein the fluctuation characteristics are used to characterize the data trend of each type of source data changing over time;

[0010] The fluctuation feature is used as an input of a trained convolutional neural network, the fluctuation feature is convolved through a convolution layer, and feature data is extracted according to the convolution result, so as to calculate a feature vector corresponding to each type of source data from the feature data through a pooling layer;

[0011] The fully connected layer in the convolutional neural network is replaced with a graph neural network, and a type graph matrix corresponding to each type of source data is calculated based on the feature vector and the graph neural network, so as to output a risk prediction result according to the type graph matrix through the graph neural network;

[0012] Among them, the multiple types of source data include historical market transaction data, economic indicator data, industry report data, social media data and historical risk assessment data, and the trained convolutional neural network is trained by using the risk assessment value, expected loss, default probability, tail risk and volatility in the historical risk assessment data as the input of the convolutional neural network and using the historical risk assessment value as the output.

[0013] Furthermore, the historical market transaction data is represented as:

[0014] ;

[0015] in, Respectively indicate the number Assets at time The price, trading volume, number of open contracts, highest price, lowest price, closing price, bid-ask spread and yield rate at the moment. The historical market transaction data;

[0016] The economic indicator data are expressed as:

[0017] ;

[0018] in, For the economic indicator data, Respectively indicate time The GDP growth rate, inflation rate, unemployment rate, base interest rate and currency exchange rate at that moment.

[0019] Furthermore, the industry report data is expressed as:

[0020] ;

[0021] in, Report data for the industry in question, Respectively indicate the number Industry at time Market share, annual growth rate, innovation indicators, consumption trends and profit margins at all times;

[0022] The social media data is represented as:

[0023] ;

[0024] in, For the social media data, Respectively indicate the number The brand theme in time The emotional index and overall rating of the moment;

[0025] ;

[0026] in, is the historical risk assessment value, Respectively indicate time The forecasted value at risk, expected loss, probability of default, tail risk, and volatility at a given time, including a confidence level.

[0027] Furthermore, the fluctuation characteristics of each type of source data are extracted by Fourier transform, and the fluctuation characteristics are used to characterize the data trend of each type of source data changing over time, including:

[0028] Obtain multiple data corresponding to various types of source data at different times, and establish a data matrix corresponding to each data in chronological order;

[0029] The data matrix corresponding to the social media data is expressed as:

[0030] ;

[0031] The fluctuation characteristics are expressed as:

[0032] ;

[0033] In the formula, is the data matrix, For time point, is the fluctuation characteristic, is an imaginary number, is a constant, and are the horizontal and vertical coordinates of the fluctuation feature, respectively, Represents the data matrix Line The element value of the column, and , .

[0034] Furthermore, the calculation formula of the type map matrix is:

[0035] ;

[0036] ;

[0037] In the formula, For the The transpose of the feature vector corresponding to the source data of each type, and , is the number of source data of all types, is the correlation matrix, is the degree matrix, is an activation function, which includes a sigmoid function and a ReLu function. For the The type map matrix of the layer, For the The type map matrix of the layer, , Indicates The weight matrix of the layer, the initial value of the weight matrix is obtained by setting according to the internal parameters of the graph neural network, is obtained through continuous iteration of the graph neural network. Represents the number of convolutional layers of the graph neural network.

[0038] Furthermore, the degree matrix is ​​a diagonal matrix, and the diagonal matrix is ​​composed of diagonal elements, and the expression of the diagonal elements is:

[0039] ;

[0040] In the formula, For the diagonal elements, the function Represents the incidence matrix Middle The number of elements in the row whose element values ​​are greater than the first threshold, used to describe the number of adjacent edges;

[0041] The expression of the association matrix is:

[0042] ;

[0043] in, The incidence matrix Middle Line The elements of the column, Respectively The feature vector corresponding to the source data of each type, and Respectively and The standard deviation of .

[0044] Furthermore, the method further comprises:

[0045] Setting a primary key for the multiple types of source data by an asymmetric encryption algorithm, and generating multiple secondary keys based on the primary key to encrypt each type of preprocessed source data;

[0046] A plurality of third-level secret keys are generated based on the second-level secret key, and the risk prediction results corresponding to each type of source data are encrypted respectively using the plurality of third-level secret keys.

[0047] The second aspect of the present invention discloses a digital risk control platform for enterprises based on big data, comprising:

[0048] A source data preprocessing module is used to obtain various types of source data, preprocess the source data, and store the preprocessed source data in a distributed database. The preprocessing includes data cleaning, deduplication, and standardization.

[0049] A feature extraction module, used for extracting the fluctuation feature of each type of source data through Fourier transform, wherein the fluctuation feature is used for characterizing the data trend of each type of source data changing over time;

[0050] A feature vector calculation module, used to use the fluctuation feature as an input of a trained convolutional neural network, convolve the fluctuation feature through a convolution layer, and extract feature data according to the convolution result, so as to calculate a feature vector corresponding to each type of source data from the feature data through a pooling layer;

[0051] A risk prediction module, used to replace the fully connected layer in the convolutional neural network with a graph neural network, and calculate a type graph matrix corresponding to each type of source data based on the feature vector and the graph neural network, so as to output a risk prediction result according to the type graph matrix through the graph neural network;

[0052] Among them, the multiple types of source data include historical market transaction data, economic indicator data, industry report data, social media data and historical risk assessment data, and the trained convolutional neural network is trained by using the risk assessment value, expected loss, default probability, tail risk and volatility in the historical risk assessment data as the input of the convolutional neural network and using the historical risk assessment value as the output.

[0053] The third aspect of the present invention discloses a terminal, including a processor and a storage medium; the characteristics are:

[0054] The storage medium is used to store instructions;

[0055] The processor is used to operate according to the instructions to execute the steps of the method described in the first aspect.

[0056] The fourth aspect of the present invention discloses a computer-readable storage medium having a computer program stored thereon, characterized in that when the program is executed by a processor, the steps of the method described in the first aspect are implemented.

[0057] The beneficial effects of the present invention are that, compared with the prior art, the present invention has the following advantages:

[0058] By obtaining multiple types of source data, preprocessing the source data, and storing the preprocessed source data in a distributed database, the fluctuation characteristics of each type of source data are extracted through Fourier transform, that is, the data trend of each type of source data changing over time. Subsequently, the fluctuation characteristics are used as the input of the trained convolutional neural network, the fluctuation characteristics are convolved through the convolution layer, and the feature data is extracted according to the convolution result, so as to calculate the feature vector corresponding to each type of source data from the feature data through the pooling layer. Finally, the fully connected layer in the convolutional neural network is replaced with a graph neural network, and the type graph matrix corresponding to each type of source data is calculated based on the feature vector and the graph neural network, so as to output the risk prediction result according to the type graph matrix through the graph neural network. This method establishes a digital risk control platform through a deep learning algorithm, which can perform intelligent analysis and information recognition on multiple types of source data. Compared with the traditional sensitive information recognition method, it is no longer limited to the set recognition rules, which improves the accuracy of sensitive information recognition, strengthens the security of enterprise digital risk control, and can effectively prevent the indirect leakage of sensitive information. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 It is a flowchart of a digital risk control method for enterprises based on big data;

[0060] Figure 2 It is a schematic diagram of the relationship between keys at each level and files at each level. DETAILED DESCRIPTION

[0061] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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.

[0062] like Figure 1As shown, in one embodiment, the enterprise digital risk control method based on big data provided by the present invention includes the following steps:

[0063] Step S110, obtaining multiple types of source data, preprocessing the source data, and storing the preprocessed source data in a distributed database, wherein the preprocessing includes data cleaning, deduplication, and standardization.

[0064] Among them, the various types of source data include historical market transaction data, economic indicator data, industry report data, social media data, and historical risk assessment data.

[0065] In some embodiments, in the enterprise digital risk control method based on big data provided by the present invention, historical market transaction data is represented as:

[0066] .

[0067] in, Respectively indicate the number Assets at time The price, trading volume, number of open contracts, highest price, lowest price, closing price, bid-ask spread and yield rate at the moment. For historical market transaction data.

[0068] Economic indicator data are expressed as:

[0069] .

[0070] in, For economic indicator data, Respectively indicate time The GDP growth rate, inflation rate, unemployment rate, base interest rate and currency exchange rate at that moment.

[0071] Industry report data is expressed as:

[0072] .

[0073] in, Report data for the industry, Respectively indicate the number Industry at time Market share, annual growth rate, innovation indicators, consumption trends and profit margins at all times;

[0074] Social media data is represented as:

[0075] .

[0076] in, For social media data, Respectively indicate the number The brand theme in time The emotional index and overall rating of the moment.

[0077] .

[0078] in, is the historical risk assessment value, Respectively indicate time The risk value, expected loss, default probability, tail risk and volatility predicted at any given moment, including the confidence level.

[0079] Step S120, extracting the fluctuation characteristics of each type of source data through Fourier transform, where the fluctuation characteristics are used to characterize the data trend of each type of source data changing over time.

[0080] In some embodiments, the enterprise digital risk control method based on big data provided by the present invention extracts the fluctuation characteristics of each type of source data through Fourier transform, and the fluctuation characteristics are used to characterize the data trend of each type of source data changing over time, and specifically includes the following steps:

[0081] Step S121, obtaining multiple data corresponding to multiple types of source data at different times, and establishing a data matrix corresponding to each data in chronological order;

[0082] Among them, the data matrix corresponding to the social media data is expressed as:

[0083] .

[0084] The fluctuation characteristics are expressed as:

[0085] .

[0086] In the formula, is the data matrix, For time point, is the fluctuation characteristic, is an imaginary number, is a constant, and are the horizontal and vertical coordinates of the fluctuation characteristics, Represents the data matrix Line The element value of the column, and , .

[0087] In some embodiments, the enterprise digital risk control method based on big data provided by the present invention, the calculation formula of the type graph matrix is:

[0088] .

[0089] .

[0090] In the formula, For the The transpose of the feature vector corresponding to the source data of each type, and , is the number of source data of all types, is the correlation matrix, is the degree matrix, is the activation function, which includes sigmoid function and ReLu function. For the The type map matrix of the layer, For the The type map matrix of the layer, , Indicates The weight matrix of the layer, the initial value of the weight matrix It is obtained by setting the internal parameters of the graph neural network. It is obtained through continuous iteration of graph neural network. Represents the number of convolutional layers of the graph neural network.

[0091] In some embodiments, in the enterprise digital risk control method based on big data provided by the present invention, the degree matrix is ​​a diagonal matrix, and the diagonal matrix is ​​composed of diagonal elements. The expression of the diagonal elements is:

[0092] .

[0093] In the formula, are diagonal elements, the function Represents the incidence matrix Middle The number of elements in the row whose values ​​are greater than the first threshold is used to describe the number of adjacent edges.

[0094] The expression of the incidence matrix is:

[0095] .

[0096] in, is the incidence matrix Middle Line The elements of the column, Respectively The feature vector corresponding to the source data of each type, and Respectively and The standard deviation of .

[0097] Step S130, using the fluctuation feature as the input of the trained convolutional neural network, convolving the fluctuation feature through the convolution layer, and extracting feature data based on the convolution result, so as to calculate the feature vector corresponding to each type of source data from the feature data through the pooling layer.

[0098] Among them, the trained convolutional neural network is obtained by training with the risk assessment value, expected loss, default probability, tail risk and volatility in the historical risk assessment data as the input of the convolutional neural network, and with the historical risk assessment value as the output.

[0099] Step S140, replace the fully connected layer in the convolutional neural network with a graph neural network, and calculate the type graph matrix corresponding to each type of source data based on the feature vector and the graph neural network, so as to output the risk prediction result according to the type graph matrix through the graph neural network.

[0100] In some embodiments, the enterprise digital risk control method based on big data provided by the present invention further includes the following steps:

[0101] Step S210, setting a primary key for multiple types of source data through an asymmetric encryption algorithm, and generating multiple secondary keys based on the primary key to encrypt each type of pre-processed source data.

[0102] Step S220, generate multiple third-level keys based on the second-level key, and encrypt the risk prediction results corresponding to each type of source data using the multiple third-level keys.

[0103] The above-mentioned enterprise digital risk control method based on big data obtains multiple types of source data, preprocesses the source data, and stores the preprocessed source data in a distributed database, and then extracts the fluctuation characteristics of each type of source data through Fourier transform, that is, the data trend of each type of source data changing over time. Subsequently, the fluctuation characteristics are used as the input of the trained convolutional neural network, the fluctuation characteristics are convolved through the convolution layer, and the feature data is extracted according to the convolution result, so as to calculate the feature vector corresponding to each type of source data from the feature data through the pooling layer. Finally, the fully connected layer in the convolutional neural network is replaced with a graph neural network, and the type graph matrix corresponding to each type of source data is calculated based on the feature vector and the graph neural network, so as to output the risk prediction result according to the type graph matrix through the graph neural network. This method establishes a digital risk control platform through a deep learning algorithm, which can perform intelligent analysis and information recognition on multiple types of source data. Compared with the traditional sensitive information recognition method, it is no longer limited to the set recognition rules, improves the accuracy of sensitive information recognition, strengthens the security of enterprise digital risk control, and can effectively prevent the indirect leakage of sensitive information.

[0104] In a specific embodiment, the enterprise digital risk control method based on big data provided by the present invention includes steps 1 to 3.

[0105] Step 1: Source data collection and preprocessing.

[0106] Source data include: historical market transaction data, economic indicator data, industry report data, social media data, historical risk assessment values ​​and other types of source data.

[0107] Preprocessing includes: data cleaning and processing, that is, using big data technology to clean, deduplicate and standardize the original data, and store the data in a distributed database for real-time access.

[0108] Specific, historical market transaction data At least it can be expressed as:

[0109] .

[0110] in, Can be represented by numbers Financial assets at time The price, trading volume, number of open contracts, highest price, lowest price, closing price, bid-ask spread and yield rate at the moment.

[0111] Economic indicators data At least it can be expressed as:

[0112] .

[0113] in, Can indicate time separately GDP growth rate, inflation rate, unemployment rate, base interest rate and currency exchange rate under the current situation.

[0114] Industry report data At least it can be expressed as:

[0115] .

[0116] in, Can be represented by numbers Industry at time Market share, annual growth rate, innovation indicators, consumption trends and profit margins under

[0117] Social media data At least it can be expressed as:

[0118] .

[0119] in, Can be represented by numbers The brand theme in time The sentiment index and overall score below.

[0120] Historical risk assessment At least it can be expressed as:

[0121] .

[0122] in, Can express time The predicted risk value (for example, with a confidence level of 95%), expected loss, default probability, tail risk and volatility forecasts are shown in Figure 2.

[0123] It should be noted that As input features in step 3, As the output feature in step 3, therefore, It can also represent the risk prediction result in step 3.

[0124] Step 2: Use Fourier transform to extract the fluctuation characteristics of each type of source data.

[0125] For each type of data, there is a trend over time. For example, The data at different times represent 7 types of data.

[0126] Each type of data, such as , the corresponding data matrix can be established according to time :

[0127] .

[0128] Fluctuation characteristics :

[0129] .

[0130] in, For time point, is an imaginary number, is a constant, and are the horizontal and vertical coordinates of the fluctuation characteristics, Represents the data matrix No. Line The element value of the column, where , Therefore, for In terms of .

[0131] Step 3: Based on the trained convolutional neural algorithm, input the fluctuation characteristics to obtain the risk prediction results.

[0132] Step 3.1, input the fluctuation characteristics into the trained convolutional neural network.

[0133] The convolutional neural network includes: convolution layer, feature extraction layer, pooling layer and fully connected layer. The convolution layer is used to convolve the pre-extracted fluctuation features, and the feature extraction layer is used to extract feature data from the convolution result. The pooling layer is used to calculate the feature vector corresponding to each type of data from the feature data. The fully connected layer does not perform any operation, and the fully connected layer is replaced by the graph neural network.

[0134] Step 3.2, pass through the convolutional layer, feature extraction layer, and pooling layer in the convolutional neural network in turn, and output the corresponding feature vector.

[0135] It should be noted that each type of data corresponds to one of the above-mentioned feature vectors.

[0136] Step 3.3, based on each feature vector and graph neural network, calculate the type graph matrix.

[0137] The calculation formula of the type matrix is:

[0138] ,

[0139] .

[0140] in, For the The transpose of the eigenvectors corresponding to the types, where , for all types of quantities; is the correlation matrix, is the degree matrix, is the activation function, which can be a sigmoid function or a ReLu function; For the The type map matrix of the layer, For the The type map matrix of the layer, , Indicates The weight matrix of the layer, the initial value of the weight matrix It is obtained by setting the internal parameters of the graph neural network. It is obtained through continuous iteration of graph neural network. Represents the number of convolutional layers of the graph neural network.

[0141] The degree matrix is ​​a diagonal matrix with diagonal elements The expression is:

[0142] .

[0143] Among them, the function Represents the incidence matrix Middle The number of elements in the row whose value is greater than 0.5, used to describe the number of adjacent edges.

[0144] Incidence Matrix The expression is:

[0145] .

[0146] in, is the incidence matrix Middle Line The elements of the column, Respectively The characteristic vector corresponding to each type, and Respectively and The standard deviation of .

[0147] Step 3.4: Based on the graph neural network, output the risk prediction results according to the type graph matrix.

[0148] In this embodiment, for the enterprise digital risk control platform based on big data, from obtaining source data (documents), to data preprocessing (equivalent to generating corresponding secondary data (documents)), to convolutional neural network learning (equivalent to generating corresponding tertiary data (documents)). In order to prevent information leakage, each piece of data has the obligation to be encrypted. Among them, the source data may include multiple copies of primary data (documents).

[0149] In this embodiment, the data encryption process includes steps S1-S2.

[0150] Step S1, setting a primary key for source data.

[0151] It should be noted that an encryption algorithm is a set of rules for performing encryption and decryption operations, but it requires a key to perform the actual encryption and decryption operations. Therefore, without a key, the encryption algorithm cannot generate meaningful ciphertext or restore ciphertext to plaintext. The role of the key is to determine how to apply the rules of the algorithm, ensuring that the ciphertext generated by the same algorithm under different keys is different, and each key is a unique input to the algorithm. In this way, even if the same encryption algorithm is used, different keys will generate different encryption results. This uniqueness of encryption ensures the security of the data, because even if the algorithm is made public, as long as the key is not leaked, the encrypted data cannot be easily decrypted.

[0152] Step S2: Generate multiple secondary keys based on the primary key to match the secondary data corresponding to the source data.

[0153] from Figure 2 It is not difficult to see that the data files at all levels of the risk control platform can be similar to the project branches in tortoiseGit or svn. However, unlike tortoiseGit, the risk control platform does not focus on data merging, but is more concerned with the security of upper and lower level data.

[0154] like Figure 2 As shown in the figure, for the risk control platform, on the one hand, setting a separate key for each file is a feasible solution, but the disadvantage is that it lacks manageability. The manageability here should be understood from two aspects: (1) Once the key of a file is forgotten, it is troublesome. (2) For project managers (not actual workers, such as source data acquisition workers, data cleaning workers, algorithm processing workers, etc., project managers refer to people who assign work), when assigning work setting permissions, different keys need to be entered, which greatly reduces work efficiency and is not conducive to data security.

[0155] On the other hand, as in conventional technology, setting associated keys will lead to the following problem: if an employee has access to multiple keys, such as k11, k21, k411, k12 and k121, then the associated matrix or function can be analyzed through their association, and then other keys, such as k412, k1, etc., can be brute-force cracked.

[0156] It is understandable that kx or kxx mentioned below refers to the key corresponding to the file, such as Figure 2 As shown, x or xx can be, for example: 1, 2, 3 or 11, 12, 31, 32, etc.

[0157] In some embodiments, the principle of RSA (asymmetric encryption algorithm) may be used to generate a key, and the generated key k1 may be shown as follows:

[0158] ,

[0159] ,

[0160] .

[0161] in, Can be an RSA function, is the secondary encryption key, and It is a preset arbitrary function, but it needs to meet the following conditions:

[0162] (1) As long as the input is different, the output will definitely be different.

[0163] (2) are all prime numbers that meet the RSA conditions. All of them are RSA-compliant bases. Indicates the node number (equivalent to the j-th child node).

[0164] It should be noted that Can be fixed throughout the project, and (a separate source data), that is, the project manager is only responsible for holding these four keys. Once the secondary key is forgotten, as long as the primary key is not forgotten, the secondary key can be restored through a limited number of attempts based on the primary key.

[0165] Specifically, one encryption method is used to encrypt data to generate a corresponding key, and another encryption method can be used to encrypt the data to generate another key. Here, k1 and k2 represent the keys of the two different encryption methods for the data. Then, based on the keys k1 and k2, the data can be encrypted twice by the same encryption method as the key k1 or k2 or other encryption methods.

[0166] So, assuming:

[0167] .

[0168] .

[0169] in, , They are two secondary keys obtained by encrypting data on the basis of k1 and k2 respectively in the same encryption method as key k1.

[0170] Similarly, for data using the same secondary encryption key, different encryption methods can be used to perform tertiary encryption, thereby generating different tertiary keys. Indicates the same secondary encryption key, then for the secondary key Using different encryption methods to encrypt data at three levels, different three-level keys can at least be expressed as , as well as .

[0171] Therefore, combining the above encryption formula, we can get:

[0172] ,

[0173] ,

[0174] .

[0175] It is not difficult to find from the above rules: function The second parameter in indicates that the child node is the parent node and The number of child nodes, and the function The parameters in represent the prime numbers corresponding to all parent nodes of the child node.

[0176] Therefore, let's assume that:

[0177] .

[0178] According to the formula of the generated key k1, we can get:

[0179] ,

[0180] .

[0181] Similarly, , The secondary encryption keys are The keys corresponding to two different levels of encryption.

[0182] Assume again:

[0183] .

[0184] It is not difficult to deduce that:

[0185] ,

[0186] .

[0187] The enterprise digital risk control platform based on big data provided by the present invention is described below. The enterprise digital risk control platform based on big data described below and the enterprise digital risk control method based on big data described above can be referenced to each other.

[0188] In one embodiment, a big data-based enterprise digital risk control platform includes a source data preprocessing module, a feature extraction module, a feature vector calculation module, and a risk prediction module.

[0189] The source data preprocessing module is used to obtain various types of source data, preprocess the source data, and store the preprocessed source data in a distributed database. The preprocessing includes data cleaning, deduplication, and standardization.

[0190] The feature extraction module is used to extract the fluctuation characteristics of each type of source data through Fourier transform, and the fluctuation characteristics are used to characterize the data trend of each type of source data changing over time.

[0191] The feature vector calculation module is used to take the fluctuation feature as the input of the trained convolutional neural network, convolve the fluctuation feature through the convolution layer, and extract feature data based on the convolution result to calculate the feature vector corresponding to each type of source data from the feature data through the pooling layer.

[0192] The risk prediction module is used to replace the fully connected layer in the convolutional neural network with a graph neural network, and calculate the type graph matrix corresponding to each type of source data based on the feature vector and the graph neural network, so as to output the risk prediction result according to the type graph matrix through the graph neural network.

[0193] Among them, various types of source data include historical market transaction data, economic indicator data, industry report data, social media data and historical risk assessment data. The trained convolutional neural network is trained with the risk assessment value, expected loss, default probability, tail risk and volatility in the historical risk assessment data as the input of the convolutional neural network and with the historical risk assessment value as the output.

[0194] In this embodiment, the enterprise digital risk control platform based on big data provided by the present invention, the historical market transaction data is represented as:

[0195] .

[0196] in, Respectively indicate the number Assets at time The price, trading volume, number of open contracts, highest price, lowest price, closing price, bid-ask spread and yield rate at the moment. For historical market transaction data.

[0197] Economic indicator data are expressed as:

[0198] .

[0199] in, For economic indicator data, Respectively indicate time The GDP growth rate, inflation rate, unemployment rate, base interest rate and currency exchange rate at that moment.

[0200] In this embodiment, the enterprise digital risk control platform based on big data provided by the present invention, the industry report data is represented as follows:

[0201] .

[0202] in, Report data for the industry, Respectively indicate the number Industry at time Market share, annual growth rate, innovation indicators, consumption trends and profit margins at all times;

[0203] Social media data is represented as:

[0204] .

[0205] in, For social media data, Respectively indicate the number The brand theme in time The emotional index and overall rating of the moment.

[0206] .

[0207] in, is the historical risk assessment value, Respectively indicate time The risk value, expected loss, default probability, tail risk and volatility predicted at any given moment, including the confidence level.

[0208] In this embodiment, the enterprise digital risk control platform based on big data provided by the present invention, the feature extraction module is specifically used for:

[0209] Acquire multiple data corresponding to various types of source data at different times, and establish a data matrix corresponding to each data in chronological order.

[0210] Among them, the data matrix corresponding to the social media data is expressed as:

[0211] .

[0212] The fluctuation characteristics are expressed as:

[0213] .

[0214] In the formula, is the data matrix, For time point, is the fluctuation characteristic, is an imaginary number, is a constant, and are the horizontal and vertical coordinates of the fluctuation characteristics, Represents the data matrix Line The element value of the column, and , .

[0215] In this embodiment, the enterprise digital risk control platform based on big data provided by the present invention has a calculation formula of the type graph matrix as follows:

[0216] .

[0217] .

[0218] In the formula, For the The transpose of the feature vector corresponding to the source data of each type, and , is the number of source data of all types, is the correlation matrix, is the degree matrix, is the activation function, which includes sigmoid function and ReLu function. For the The type map matrix of the layer, For the The type map matrix of the layer, , Indicates The weight matrix of the layer, the initial value of the weight matrix It is obtained by setting the internal parameters of the graph neural network. It is obtained through continuous iteration of graph neural network. Represents the number of convolutional layers of the graph neural network.

[0219] In this embodiment, the enterprise digital risk control platform based on big data provided by the present invention has a degree matrix which is a diagonal matrix. The diagonal matrix is ​​composed of diagonal elements. The expression of the diagonal elements is:

[0220] .

[0221] In the formula, are diagonal elements, the function Represents the incidence matrix Middle The number of elements in the row whose values ​​are greater than the first threshold is used to describe the number of adjacent edges.

[0222] The expression of the incidence matrix is:

[0223] .

[0224] in, is the incidence matrix Middle Line The elements of the column, Respectively The feature vector corresponding to the source data of each type, and Respectively and The standard deviation of .

[0225] In this embodiment, the enterprise digital risk control platform based on big data provided by the present invention further includes a data hierarchical encryption module for:

[0226] A primary key is set for various types of source data through an asymmetric encryption algorithm, and multiple secondary keys are generated based on the primary key to encrypt each type of preprocessed source data.

[0227] Multiple third-level keys are generated based on the second-level key, and the risk prediction results corresponding to each type of source data are encrypted using the multiple third-level keys.

[0228] The present disclosure may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0229] A computer-readable storage medium may be a tangible device that can hold and store instructions used by an instruction execution device. A computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the above. More specific examples (a non-exhaustive list) of computer-readable storage media include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a raised structure in a groove on which instructions are stored, and any suitable combination of the above. The computer-readable storage medium used herein is not to be interpreted as a transient signal itself, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse through a fiber optic cable), or an electrical signal transmitted through a wire.

[0230] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0231] The computer program instructions for performing the operations of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages, such as Smalltalk, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer, partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect through the Internet). In some embodiments, by using the state information of the computer-readable program instructions to personalize an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), the electronic circuit may execute the computer-readable program instructions, thereby implementing various aspects of the present disclosure.

[0232] Various aspects of the present disclosure are described herein with reference to flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present disclosure. It should be understood that each box in the flowchart and / or block diagram and the combination of boxes in the flowchart and / or block diagram can be implemented by computer-readable program instructions.

[0233] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine, so that when these instructions are executed by the processor of the computer or other programmable data processing device, a device that implements the functions / actions specified in one or more boxes in the flowchart and / or block diagram is generated. These computer-readable program instructions can also be stored in a computer-readable storage medium, and these instructions cause the computer, programmable data processing device, and / or other equipment to work in a specific manner, so that the computer-readable medium storing the instructions includes a manufactured product, which includes instructions for implementing various aspects of the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0234] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device so that a series of operating steps are performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to implement the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0235] The flow chart and block diagram in the accompanying drawings show the possible architecture, function and operation of the system, method and computer program product according to multiple embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a part of a module, program segment or instruction, and a part of the module, program segment or instruction includes one or more executable instructions for realizing the specified logical function. In some alternative implementations, the function marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two continuous square boxes can actually be executed substantially in parallel, and they can sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs the specified function or action, or can be implemented with a combination of special hardware and computer instructions.

[0236] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A digital risk control method for enterprises based on big data, characterized in that: The method comprises: Acquire various types of source data, preprocess the source data, and store the preprocessed source data in a distributed database, wherein the preprocessing includes data cleaning, deduplication, and standardization; Extracting fluctuation characteristics of each type of source data by Fourier transform, wherein the fluctuation characteristics are used to characterize the data trend of each type of source data changing over time; The fluctuation feature is used as an input of a trained convolutional neural network, the fluctuation feature is convolved through a convolution layer, and feature data is extracted according to the convolution result, so as to calculate a feature vector corresponding to each type of source data from the feature data through a pooling layer; The fully connected layer in the convolutional neural network is replaced with a graph neural network, and a type graph matrix corresponding to each type of source data is calculated based on the feature vector and the graph neural network, so as to output a risk prediction result according to the type graph matrix through the graph neural network; The multiple types of source data include historical market transaction data, economic indicator data, industry report data, social media data, and historical risk assessment data; the trained convolutional neural network is obtained by training with the risk assessment value, expected loss, default probability, tail risk, and volatility in the historical risk assessment data as inputs of the convolutional neural network and with the historical risk assessment value as output; The calculation formula of the type map matrix is: ； ； In the formula, For the The transpose of the feature vector corresponding to the source data of each type, and , is the number of source data of all types, is the correlation matrix, is the degree matrix, is an activation function, which includes a sigmoid function and a ReLu function. For the The type map matrix of the layer, For the The type map matrix of the layer, , Indicates The weight matrix of the layer, the initial value of the weight matrix is obtained by setting according to the internal parameters of the graph neural network, is obtained through continuous iteration of the graph neural network. Represents the number of convolutional layers of the graph neural network; The degree matrix is ​​a diagonal matrix, which is composed of diagonal elements. The expression of the diagonal elements is: ； In the formula, For the diagonal elements, the function Represents the incidence matrix Middle The number of elements in the row whose element values ​​are greater than the first threshold, used to describe the number of adjacent edges; The expression of the correlation matrix is: ； in, The incidence matrix Middle Line The elements of the column, Respectively The feature vector corresponding to the source data of each type, and Respectively and The standard deviation of .

2. The enterprise digital risk control method based on big data according to claim 1 is characterized in that: The historical market transaction data is represented as: ； in, Respectively indicate the number Assets at time The price, trading volume, number of open contracts, highest price, lowest price, closing price, bid-ask spread and yield rate at the moment. The historical market transaction data; The economic indicator data are expressed as: ； in, For the economic indicator data, Respectively indicate time The GDP growth rate, inflation rate, unemployment rate, base interest rate and currency exchange rate at that moment.

3. The enterprise digital risk control method based on big data according to claim 2 is characterized in that: The industry report data is expressed as: ； in, Report data for the industry in question, Respectively indicate the number Industry at time Market share, annual growth rate, innovation indicators, consumption trends and profit margins at all times; The social media data is represented as: ； in, For the social media data, Respectively indicate the number The brand theme in time The emotional index and overall rating of the moment; ； in, is the historical risk assessment value, Respectively indicate time The forecasted value at risk, expected loss, probability of default, tail risk, and volatility at a given time, including a confidence level.

4. The enterprise digital risk control method based on big data according to claim 3 is characterized in that: The fluctuation characteristics of each type of source data are extracted by Fourier transform, and the fluctuation characteristics are used to characterize the data trend of each type of source data changing over time, including: Obtain multiple data corresponding to various types of source data at different times, and establish a data matrix corresponding to each data in chronological order; The data matrix corresponding to the social media data is expressed as: ； The fluctuation characteristics are expressed as: ； In the formula, is the data matrix, For time point, is the fluctuation characteristic, is an imaginary number, is a constant, and are the horizontal and vertical coordinates of the fluctuation feature, respectively, Represents the data matrix Line The element value of the column, and , .

5. The enterprise digital risk control method based on big data according to any one of claims 1 to 4, characterized in that: The method further comprises: Setting a primary key for the multiple types of source data by an asymmetric encryption algorithm, and generating multiple secondary keys based on the primary key to encrypt each type of preprocessed source data; A plurality of third-level secret keys are generated based on the second-level secret key, and the risk prediction results corresponding to each type of source data are encrypted respectively using the plurality of third-level secret keys.

6. A digital enterprise risk control platform based on big data, used to execute any of the methods described in claims 1-5, characterized in that: include: A source data preprocessing module is used to obtain various types of source data, preprocess the source data, and store the preprocessed source data in a distributed database. The preprocessing includes data cleaning, deduplication, and standardization. A feature extraction module, used to extract the fluctuation feature of each type of source data through Fourier transform, wherein the fluctuation feature is used to characterize the data trend of each type of source data changing over time; A feature vector calculation module, used to use the fluctuation feature as an input of a trained convolutional neural network, convolve the fluctuation feature through a convolution layer, and extract feature data according to the convolution result, so as to calculate a feature vector corresponding to each type of source data from the feature data through a pooling layer; A risk prediction module, used to replace the fully connected layer in the convolutional neural network with a graph neural network, and calculate a type graph matrix corresponding to each type of source data based on the feature vector and the graph neural network, so as to output a risk prediction result according to the type graph matrix through the graph neural network; Among them, the multiple types of source data include historical market transaction data, economic indicator data, industry report data, social media data and historical risk assessment data, and the trained convolutional neural network is trained by using the risk assessment value, expected loss, default probability, tail risk and volatility in the historical risk assessment data as the input of the convolutional neural network and using the historical risk assessment value as the output.

7. A terminal comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to execute the steps of the method according to any one of claims 1-5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images