Enterprise risk prediction method, device, equipment and storage medium

Abstract

The present invention discloses a method, device, equipment and storage medium for enterprise risk prediction, and relates to the field of artificial intelligence technology. The method includes obtaining characteristic data corresponding to each enterprise in the enterprise set; for each enterprise, based on the characteristic data corresponding to the current enterprise and the pre-trained risk prediction model, obtaining the risk prediction result corresponding to the current enterprise, the risk prediction result characterizes the probability that the current enterprise is a risk enterprise; obtaining the initial relationship matrix of each core enterprise and each chain enterprise, using an arbitrary order neighboring network to reconstruct the initial relationship matrix, and obtaining the reconstructed association relationship matrix; based on the association relationship matrix and the risk prediction results corresponding to each enterprise, obtaining the core enterprise risk prediction result set and the chain enterprise risk prediction result set. Using the above technical solution, the accuracy of enterprise risk assessment can be improved, helping enterprises avoid related enterprises with higher risks, thereby achieving the technical effect of obtaining high-quality cooperative enterprises.

Description

Technical Field

[0001] Embodiments of the present invention relate to the field of artificial intelligence technology, and in particular, to an enterprise risk prediction method, device, equipment and storage medium. Background Art

[0002] In the field of data mining, based on business needs, it is necessary to conduct risk assessments on multiple enterprises to determine whether each enterprise is a risky enterprise, so that high-quality customer groups can be expanded based on the assessment results and risky customer groups can be avoided. The multiple enterprises may include core enterprises and chain enterprises. Core enterprises include at least one designated enterprise of high importance, and chain enterprises may include upstream and downstream enterprises of each core enterprise.

[0003] Currently, a single machine learning model is mainly used to conduct risk assessment on each enterprise, and the accuracy of this risk assessment method is low. Summary of the invention

[0004] The embodiments of the present invention provide a method, device, equipment and storage medium for predicting enterprise risks, so as to optimize the existing solutions for predicting enterprise risks and improve the accuracy of enterprise risk prediction.

[0005] In a first aspect, an embodiment of the present invention provides a method for predicting enterprise risks, comprising:

[0006] Acquire characteristic data corresponding to each enterprise in the enterprise set; wherein the enterprise set includes at least one core enterprise and at least one chain enterprise;

[0007] For each of the enterprises, based on the characteristic data corresponding to the current enterprise and the pre-trained risk prediction model, a risk prediction result corresponding to the current enterprise is obtained, wherein the risk prediction result represents the probability that the current enterprise is a risky enterprise;

[0008] Obtaining an initial relationship matrix between each of the core enterprises and each of the chain enterprises, and reconstructing the initial relationship matrix using an arbitrary order proximity network to obtain a reconstructed association relationship matrix;

[0009] Based on the association relationship matrix and the risk prediction results corresponding to each of the enterprises, a core enterprise risk prediction result set and a chain enterprise risk prediction result set are obtained.

[0010] In a second aspect, an embodiment of the present invention further provides an enterprise risk prediction device, the device comprising:

[0011] A feature data acquisition module, used to acquire feature data corresponding to each enterprise in the enterprise set; wherein the enterprise set includes at least one core enterprise and at least one chain enterprise;

[0012] A first risk result obtaining module is used to obtain, for each of the enterprises, a risk prediction result corresponding to the current enterprise based on the characteristic data corresponding to the current enterprise and a pre-trained risk prediction model, wherein the risk prediction result represents the probability that the current enterprise is a risky enterprise;

[0013] A relationship matrix acquisition module is used to obtain the initial relationship matrix between each of the core enterprises and each of the chain enterprises, and reconstruct the initial relationship matrix using an arbitrary order neighboring network to obtain a reconstructed correlation relationship matrix;

[0014] The second risk result acquisition module is used to obtain a core enterprise risk prediction result set and a chain enterprise risk prediction result set based on the association relationship matrix and the risk prediction results corresponding to each of the enterprises.

[0015] In a third aspect, an embodiment of the present invention further provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, an enterprise risk prediction method as described in any one of the embodiments of the present invention is implemented.

[0016] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements an enterprise risk prediction method as described in any one of the embodiments of the present invention.

[0017] The enterprise risk prediction scheme provided by the embodiment of the present invention first inputs the acquired characteristic data corresponding to each enterprise into the pre-trained risk prediction model to obtain the risk prediction result corresponding to the current enterprise, and the risk prediction result characterizes the probability that the current enterprise is a risk enterprise, thereby clarifying the risk probability of each core enterprise and each chain enterprise; then obtains the initial relationship matrix of each core enterprise and each chain enterprise, and reconstructs the initial relationship matrix using an arbitrary order neighboring network to obtain the reconstructed association relationship matrix, which can clarify the closeness of the relationship between each core enterprise and each chain enterprise; based on the association relationship matrix and the risk prediction results corresponding to each enterprise, obtain the core enterprise risk prediction result set and the chain enterprise risk prediction result set, that is, based on the association relationship between the core enterprise and the chain enterprise, the risk prediction result set of each enterprise output by the risk prediction model is optimized. Using the above technical solution, the accuracy of enterprise risk assessment can be improved, helping enterprises avoid related enterprises with higher risks, thereby achieving the technical effect of obtaining high-quality cooperative enterprises. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0019] Figure 1a is a flow chart of a method for predicting enterprise risks provided by Embodiment 1 of the present invention;

[0020] Figure 1b It is a logic diagram of a method for implementing enterprise risk prediction provided by the first embodiment of the present invention;

[0021] Figure 2a is a flow chart of a method for predicting enterprise risks provided by Embodiment 2 of the present invention;

[0022] Figure 2b It is a logic diagram of a risk prediction model provided in the second embodiment of the present invention;

[0023] Figure 2c It is a logic diagram for obtaining a core enterprise and chain enterprise association relationship matrix provided by the second embodiment of the present invention;

[0024] Figure 3 It is a structural schematic diagram of an enterprise risk prediction device provided by Embodiment 3 of the present invention;

[0025] Figure 4 It is a structural diagram of a computer device provided in Embodiment 4 of the present invention. DETAILED DESCRIPTION

[0026] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention. It should also be noted that, for ease of description, only parts related to the present invention, rather than all structures, are shown in the accompanying drawings.

[0027] It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings. At the same time, in the description of the present invention, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0028] Embodiment 1

[0029] Figure 1aThis is a flow chart of a method for predicting enterprise risks provided in the first embodiment of the present invention. This embodiment can be applied to the situation of assessing enterprise risks. The method can be executed by the enterprise risk prediction device in the embodiment of the present invention. The device can be implemented in software and / or hardware. Figure 1a As shown, the method specifically comprises the following steps:

[0030] S110, obtaining characteristic data corresponding to each enterprise in the enterprise set.

[0031] Taking the financial supply chain scenario as an example, financial institutions need to provide comprehensive financial services to core enterprises and their upstream and downstream chain enterprises based on the transaction process, based on the characteristics of financial products and centered around the core enterprises in the supply chain. Generally speaking, the participants in supply chain finance include: core enterprises, financing enterprises (such as upstream suppliers, downstream distributors and retailers), financial institutions and third-party logistics companies. In supply chain finance, the collection of all participating enterprises can be understood as an enterprise collection.

[0032] In this embodiment, the enterprise set includes at least one core enterprise and at least one chain enterprise.

[0033] The core enterprise can be understood as an enterprise whose business is mainly different from other enterprises in the enterprise collection. In the above supply chain finance example, the remaining enterprises except the core enterprise can be called chain enterprises. In a supply chain, there is more than one core enterprise.

[0034] Generally, when conducting risk assessment on enterprises, if the core enterprise is an enterprise with a higher risk level, the risk level of the chain enterprises related to the core enterprise is also relatively high. Therefore, this embodiment obtains the characteristic data corresponding to each enterprise in the enterprise set by analysis to predict the risk of each enterprise.

[0035] Optionally, the characteristic data of the above-mentioned enterprise may include: capital trend data, scale data, financial indicator data, and compliance indicator data of the enterprise, etc. The characteristic data of the specific enterprise is not required here, and is subject to actual analysis needs.

[0036] S120. For each enterprise, based on the characteristic data corresponding to the current enterprise and the pre-trained risk prediction model, a risk prediction result corresponding to the current enterprise is obtained, and the risk prediction result represents the probability that the current enterprise is a risky enterprise.

[0037] The risk prediction model is a corresponding model that can output the current risk prediction result by inputting the characteristic data corresponding to the current enterprise. Accordingly, the risk prediction model can be a classification model and / or a regression model. The classification model can learn the characteristics corresponding to the enterprise characteristic data in different dimensions to classify the enterprise in different dimensions. The classification result includes enterprise classifications in multiple dimensions. The regression model can make judgments on the enterprise characteristics in different dimensions based on the characteristic data corresponding to the enterprise and calculate the risk value of the enterprise. Specifically, the risk prediction model can be one or more of a logistic regression model, a regression tree model, a support vector machine model, a linear classification model, a decision tree model or a random forest model. The type of specific risk prediction model and the number of models included in the risk prediction model are not limited here.

[0038] By inputting the characteristic data corresponding to each core enterprise and the chain enterprises related to the core enterprise into the pre-trained risk prediction model, the risk prediction results corresponding to each enterprise can be obtained, wherein the risk prediction results represent the probability that the current enterprise is a risky enterprise. If the risk prediction result corresponding to the current enterprise is greater than the preset threshold (for example, 80%), it indicates that the probability that the current enterprise is a risky enterprise is high. The subsequent steps will not calculate the enterprises with a high current risk probability, and the enterprises that do not meet the requirements can be filtered out to reduce the difficulty of calculation, so as to further judge the enterprises whose risk prediction results meet the preset threshold.

[0039] S130, obtaining an initial relationship matrix between each core enterprise and each chain enterprise, and reconstructing the initial relationship matrix using an arbitrary order neighboring network to obtain a reconstructed association relationship matrix.

[0040] The initial relationship matrix can represent the degree of correlation between the core enterprise and the chain enterprise business. The initial relationship matrix can be obtained by combining the original upstream and downstream associations between the core enterprise and the chain enterprises, as well as the source of the characteristic data corresponding to each enterprise, to obtain the initial relationship matrix of the initial core enterprise and the chain.

[0041] The initial relationship matrix can be composed of 0 and 1, the horizontal axis can represent the core enterprise, and the vertical axis can represent the chain enterprise. If the core enterprise has an association relationship with the chain enterprise, it is recorded as 1, otherwise it is recorded as 0. Correspondingly, the coordinate (i, j) in the initial relationship matrix can indicate whether the i-th core enterprise has an association relationship with the j-th chain enterprise.

[0042] Through the 0 or 1 in the initial relationship matrix, it can be clear whether each core enterprise is associated with each chain enterprise, but the degree of association cannot be determined. In this case, the initial relationship matrix is ​​further reconstructed using an arbitrary order neighboring network to obtain a reconstructed association relationship matrix. The reconstructed association relationship matrix can characterize the closeness of the relationship between each core enterprise and each chain enterprise, so that if a certain enterprise is a risky enterprise, its impact on the current enterprise can be judged based on the closeness.

[0043] For example, if it is determined that a certain enterprise is more likely to be a venture enterprise, but the correlation matrix shows that the degree of correlation between the enterprise and the current enterprise is relatively low, then the current enterprise is less likely to be a venture enterprise.

[0044] Among them, the arbitrary order neighbor network can be implemented by using the K-Nearest Neighbor classification algorithm (KNN) or the Deep Walk algorithm (DeepWalk), and the specific algorithm used by the arbitrary order neighbor network is not limited here.

[0045] S140. Based on the correlation matrix and the risk prediction results corresponding to each enterprise, a core enterprise risk prediction result set and a chain enterprise risk prediction result set are obtained.

[0046] Please refer to Figure 1b , Figure b is a logic diagram of a method for implementing enterprise risk prediction provided by Example 1 of the present invention; the risk prediction results corresponding to each enterprise obtained in step S120, and the correlation relationship matrix obtained after reconstructing the initial relationship matrix of the core enterprise and the chain enterprise using an arbitrary order proximity network in step S130, after relevant mathematical calculations, can be obtained to obtain a set of risk prediction results for each core enterprise and a set of risk prediction results for the chain enterprises respectively.

[0047] Among them, the enterprise risk prediction result set can be represented in the form of a vector. According to the enterprise risk prediction result set, the enterprises related to the current enterprise can be sorted to clarify the enterprises most relevant to the current enterprise, and the impact of the risk levels of the remaining enterprises on the current enterprise can also be clarified to better predict the risks of the current enterprise.

[0048] The enterprise risk prediction method provided in this embodiment first inputs the acquired characteristic data corresponding to each enterprise into the pre-trained risk prediction model to obtain the risk prediction result corresponding to the current enterprise. The risk prediction result characterizes the probability that the current enterprise is a risk enterprise, thereby clarifying the risk probability of each core enterprise and each chain enterprise; then obtains the initial relationship matrix of each core enterprise and each chain enterprise, and reconstructs the initial relationship matrix using an arbitrary order neighboring network to obtain the reconstructed association relationship matrix, which can clarify the closeness of the relationship between each core enterprise and each chain enterprise; based on the association relationship matrix and the risk prediction results corresponding to each enterprise, obtain the core enterprise risk prediction result set and the chain enterprise risk prediction result set, that is, based on the association relationship between the core enterprise and the chain enterprise, the risk prediction result set of each enterprise output by the risk prediction model is optimized. Using the above technical solution, the accuracy of enterprise risk assessment can be improved, helping enterprises avoid related enterprises with higher risks, thereby achieving the technical effect of obtaining high-quality cooperative enterprises.

[0049] Embodiment 2

[0050] Figure 2a A flowchart of a method for enterprise risk prediction provided in the second embodiment of the present invention is provided. This embodiment is optimized based on the above embodiment, and optimizes the step of obtaining the risk prediction result corresponding to the current enterprise based on the feature data corresponding to the current enterprise and the pre-trained risk prediction model, including: inputting the feature data corresponding to the current enterprise into the 1st to nth sub-models contained in the pre-trained risk prediction model; wherein n is an integer greater than 1; inputting the result data output by the n sub-models and the feature data corresponding to the current enterprise into the n+1th sub-model contained in the risk prediction model; and obtaining the risk prediction result corresponding to the current enterprise output by the n+1th sub-model. The advantage of such a setting is that different sub-models can extract different feature results of the feature data corresponding to the current enterprise from different dimensions, and the feature results output by the risk prediction model can be made more comprehensive through the n sub-models.

[0051] The steps of reconstructing the initial relationship matrix using an arbitrary order neighboring network and obtaining the reconstructed association relationship matrix are also optimized, including: obtaining the core enterprise Gaussian matrix and the chain enterprise Gaussian matrix corresponding to the initial relationship matrix; determining the core enterprise similarity matrix according to the core enterprise Gaussian matrix, and determining the chain enterprise similarity matrix according to the chain enterprise Gaussian matrix; determining the core enterprise neighboring network embedding matrix based on the core enterprise similarity matrix and the arbitrary order neighboring network, and determining the chain enterprise neighboring network embedding matrix based on the chain enterprise similarity matrix and the arbitrary order neighboring network; obtaining the reconstructed association relationship matrix based on the core enterprise neighboring network embedding matrix and the chain enterprise neighboring network embedding matrix. The advantage of this setting is that the association relationship matrix represents the degree of association between the core enterprise and the chain enterprise, which is convenient for realizing the risk prediction of the enterprise.

[0052] The step of obtaining the core enterprise risk prediction result set and the chain enterprise risk prediction result set based on the association relationship matrix and the risk prediction results corresponding to each of the enterprises is also optimized, including: obtaining the initial core enterprise risk vector and the initial chain enterprise risk vector according to the risk prediction results corresponding to each of the enterprises; wherein the initial core enterprise risk vector contains the risk prediction results of each core enterprise, and the initial chain enterprise risk vector contains the risk prediction results of each chain enterprise; based on the association relationship matrix, the initial core enterprise risk vector and the initial chain enterprise risk vector, respectively determine the final core enterprise risk vector and the final chain enterprise risk vector. The advantage of this setting is that the final risk vector obtained by calculating the obtained association relationship matrix with the initial core enterprise risk vector and the initial chain enterprise risk vector can be more accurate.

[0053] like Figure 2a As shown, the method of this embodiment specifically includes the following steps:

[0054] S210: Obtain characteristic data corresponding to each enterprise in the enterprise set.

[0055] S220. Input the characteristic data corresponding to the current enterprise into the 1st to nth sub-models included in the pre-trained risk prediction model.

[0056] The enterprise risk prediction solution provided in this embodiment may include multiple sub-models in the risk prediction model. For details, please refer to Figure 2b , Figure 2b This is a logic diagram of a risk prediction model provided by the second embodiment of the present invention. Figure 2bIn the example, the characteristic data corresponding to the current enterprise are input into the 1st to nth sub-models respectively, and each sub-model can output n characteristics corresponding to the current characteristic data according to the characteristics of the current model. The advantage of this setting is that different types of sub-models can extract different characteristics of the characteristic data of the current enterprise from different dimensions, and the characteristics output by the risk prediction model can be made more accurate and comprehensive through n sub-models.

[0057] Before inputting the characteristic data corresponding to the current enterprise into the pre-risk prediction model respectively, it is necessary to train the 1st to nth sub-models included therein.

[0058] Optionally, the training method of the risk prediction model includes: obtaining a training sample set; using part of the training samples in the training sample set to train the 1st to nth sub-models; using another part of the training samples in the training sample set to train the risk prediction model including the trained 1st to nth sub-models and the n+1th sub-model to be trained.

[0059] The acquired training sample set is the feature data corresponding to the sample enterprise (for example, F), and the feature data corresponding to the sample enterprise contained in the current training sample set is divided into two parts, and one part of the feature data (for example, F1) is used as a training sample to train the 1st to nth sub-models. Accordingly, the training method for the 1st to nth sub-models can be:

[0060] The sample labels of each training sample in F1 are set in advance, and then the training samples are input into the 1st to nth sub-models respectively to output the features corresponding to each model. Then, after each feature is output, the sample labels corresponding to the training samples are compared to determine the accuracy of the current model training, and the relevant parameters contained in the corresponding sub-model are adjusted according to the difference between the output results and the sample labels. After the relevant parameters contained in the sub-model are adjusted, the training samples contained in F1 are input into the 1st to nth sub-models, and the above process is performed again to further optimize each sub-model in the risk prediction model, until the output results are within the error range after the relevant test data is input into each sub-model, indicating that the current 1st to nth sub-model training is completed.

[0061] Furthermore, after the feature data (F1) corresponding to the current sample enterprise is input into the 1st to nth sub-models, the result data set corresponding to the 1st to nth sub-models (for example, F2) can be obtained; and the result data set (F2) corresponding to the current 1st to nth sub-models and another feature data (for example, F-F1) can be integrated to obtain a new feature data set (for example, F3, where F3 = F-F1+F2), and the new feature data set (F3) is further used to train the risk prediction models of the trained 1st to nth sub-models and the n+1th sub-model to be trained. The specific process of training the n+1th sub-model is the same as the process of training the 1st to nth sub-models, which will not be repeated here.

[0062] Optionally, the 1st to nth sub-models can be classification models and regression models. The classification model can learn the features corresponding to the enterprise feature data in different dimensions to classify the enterprise in different dimensions. The classification results include enterprise classifications in multiple dimensions. The regression model can make judgments on the enterprise features in different dimensions based on the feature data corresponding to the enterprise and calculate the risk value of the enterprise. Each sub-model in the 1st to nth sub-models provided in this embodiment can be a logistic regression model, a regression tree model, a support vector machine model, a linear classification model, or a decision tree model, etc.; the n+1th sub-model can be a random forest model, and the type of the specific sub-model is not limited here.

[0063] The characteristic data corresponding to the current enterprise are further input into the pre-trained 1st to nth risk prediction sub-models. Each sub-model can analyze the characteristic data of the current enterprise from different dimensions, thereby outputting the result data corresponding to the 1st to nth sub-models.

[0064] S221. Input the result data output by the n sub-models and the characteristic data corresponding to the current enterprise into the n+1th sub-model included in the risk prediction model to obtain the risk prediction result corresponding to the current enterprise output by the n+1th sub-model.

[0065] After the current result data and the characteristic data corresponding to the current enterprise are input into the n+1th sub-model contained in the pre-trained risk prediction model, the n+1th sub-model can output the risk prediction result corresponding to the current enterprise.

[0066] Accordingly, each core enterprise and each chain enterprise is input into the trained n+1th sub-model respectively, and the risk prediction results of each enterprise can be obtained respectively.

[0067] Among them, when the n+1th sub-model outputs the risk prediction result corresponding to the current enterprise, it can be output in the form of probability, for example, predicting that the probability of the current enterprise having risk is 20%; correspondingly, it can also be output in the form of risk level, for example, the risk level of the current enterprise is 0, 1 or 2, and the larger the number, the higher the risk level. The specific form of the risk level output by the n+1th sub-model is not limited here.

[0068] It should be noted that at least one core enterprise and at least one chain enterprise included in the enterprise set must execute the above steps S220-S221 so that the n+1th sub-model outputs the risk prediction results corresponding to each enterprise.

[0069] S230, obtaining an initial relationship matrix between each core enterprise and each chain enterprise.

[0070] S240, obtaining the Gaussian matrix of the core enterprise and the Gaussian matrix of the chain enterprises corresponding to the initial relationship matrix.

[0071] Please refer to Figure 2c , Figure 2c This is a logic diagram for obtaining a core enterprise and chain enterprise association matrix provided by the second embodiment of the present invention. Figure 2c Here, H represents the core enterprise, K represents the chain enterprise, A represents the initial relationship matrix between the core enterprise and the chain enterprise, KC represents the Gaussian matrix of the core enterprise, KU represents the Gaussian matrix of the chain enterprise, CS represents the semantic matrix of the core enterprise, US represents the semantic matrix of the chain enterprise, SC represents the similarity matrix of the core enterprise, SU represents the similarity matrix of the chain enterprise, SCD represents the neighboring network embedding matrix of the core enterprise, SUD represents the neighboring network embedding matrix of the chain enterprise, and CU represents the association relationship matrix.

[0072] First, the core enterprises and chain enterprises related to the initial matrix are expressed by the following expressions:

[0073]

[0074] In the above formula, i and j can be understood as the rows and columns of the initial matrix respectively, where H ij＝1 Indicates that in the initial relationship matrix, the i-th core enterprise and the j-th chain enterprise have an association relationship; H ij＝0 It means that there is no association between the i-th core enterprise and the j-th chain enterprise in the initial relationship matrix.

[0075] Furthermore, the Gaussian matrix KC of the core enterprise and the Gaussian matrix KU of the chain enterprise can be calculated by the following expressions:

[0076] KC=exp(-γ c ||IP(C i)-IP(C j )|| 2 ) (2)

[0077] KU=exp(γ u ||IP(U i )-IP(U j )|| 2 ) (3)

[0078] in,

[0079]

[0080]

[0081] In the above formula, IP(C i ) and IP(U j ) represents the distance between the i-th row vector of the core enterprise and the j-th column vector of the chain enterprise in the initial relationship matrix, γ c and γ u Represent the adjustment coefficients of core enterprises and chain enterprises, γ c ' and γ u 'represent the original coefficients of core enterprises and chain enterprises respectively.

[0082] S241. Determine the core enterprise similarity matrix based on the core enterprise Gaussian matrix, and determine the chain enterprise similarity matrix based on the chain enterprise Gaussian matrix.

[0083] Optionally, the process of determining the core enterprise similarity matrix SC according to the core enterprise Gaussian matrix KC may be:

[0084] Firstly, the core enterprise semantic matrix CS containing the similarity values ​​between core enterprises is determined, and then the core enterprise similarity matrix SC is determined according to the core enterprise semantic matrix CS and the core enterprise Gaussian matrix KC.

[0085] The present application can construct all enterprises into a directed acyclic graph (DAG). The specific construction method of the directed acyclic graph is not limited here and can be obtained according to the construction method of the prior art.

[0086] The parameter value of the semantic value of the core enterprise of DAG (C) can be expressed by the following expression:

[0087]

[0088] In the above formula, ε is the influencing factor. If the child nodes of C are far away, the influencing factor is smaller than the nearest one.

[0089] Furthermore, the semantic value of enterprise D is added to all enterprises in the DAG, expressed as CV. iand C j The semantic similarity score of can be defined as:

[0090] CV1(C)=∑ c∈T(C) C1 C (c) (7)

[0091]

[0092] Among them, the semantic value of the core enterprise can be measured by the following expression:

[0093]

[0094]

[0095] Furthermore, according to formula (9) and formula (10), the core enterprise semantic matrix CS can be obtained:

[0096]

[0097] In the above formula, SS1 and SS2 refer to the representation of the core enterprise semantic matrix at different levels.

[0098] Accordingly, the core enterprise similarity matrix SC can be expressed by the following expression:

[0099] SC=CS⊙α+KC⊙(1-α) (12)

[0100] In the above formula, α∈(0, 1) represents the comprehensive weight.

[0101] Furthermore, the process of determining the chain enterprise similarity matrix SU according to the chain enterprise Gaussian matrix KU can be: first determine the chain enterprise functional similarity matrix SU containing similarity values ​​between each chain enterprise, and then determine the chain enterprise similarity matrix SU according to the chain enterprise functional similarity matrix SU and the chain enterprise Gaussian matrix KU.

[0102] Accordingly, it can be expressed as follows:

[0103] S(ct,DT)=max 1≤i≤nc (S(dt,dt i )) (13)

[0104]

[0105] SU=US⊙β+KU⊙(1-β) (15)

[0106] In the above formula, β∈(0, 1) represents the comprehensive weight.

[0107] S242. Determine the embedding matrix of the core enterprise neighboring network based on the core enterprise similarity matrix and the arbitrary order neighboring network, and determine the embedding matrix of the chain enterprise neighboring network based on the chain enterprise similarity matrix and the arbitrary order neighboring network.

[0108] An optional embodiment determines the core enterprise neighbor network embedding matrix SCD based on the core enterprise similarity matrix SC and the arbitrary order neighbor network, including: determining a first arbitrary order neighbor matrix based on the core enterprise similarity matrix SC; performing singular value decomposition in descending order on the first arbitrary order neighbor matrix, and selecting the first r eigenvalues ​​in the characteristic matrix after the current singular value decomposition in descending order; wherein r is an integer greater than 1; reconstructing the first principal characteristic matrix of the singular value decomposition through the currently selected first r eigenvalues, and determining the first principal characteristic matrix as the core enterprise neighbor network embedding matrix SCD.

[0109] Specifically, the first arbitrary order proximity matrix is ​​determined using the following formula:

[0110] S1=w1A1+w2A1 2 +…+w n A1 n (16)

[0111] Among them, S1 is the first arbitrary order proximity matrix, A1 is the core enterprise similarity matrix, A1 i represents the i-th order of A1, w i =β i , β i ∈(0,1), i takes values ​​in [1, n], and n is the preset order.

[0112] In formula (16), w represents the weight. The weight keeps decreasing, indicating that the distance from the node means the weight is small, which ensures the convergence of the function. It is worth noting that we call the n-order proximity the weighted combination of all orders from 1st to nth order, rather than just nth order. The formula of the proximity matrix function of any order is as follows:

[0113]

[0114] Correspondingly, based on the chain enterprise similarity matrix SU and the arbitrary order neighboring network, the chain enterprise neighboring network embedding matrix SUD is determined, including: determining a second arbitrary order neighboring matrix based on the chain enterprise similarity matrix; performing singular value decomposition in descending order on the second arbitrary order neighboring matrix, and selecting the first r eigenvalues ​​in the characteristic matrix after the current singular value decomposition in descending order; reconstructing the second main characteristic matrix of the singular decomposition through the currently selected first r eigenvalues, and determining the second main characteristic matrix as the chain enterprise neighboring network embedding matrix.

[0115] The second arbitrary order proximity matrix is ​​determined using the following formula:

[0116] S2=w1A2+w2A2 2 +…+w n A2 n (18)

[0117] Among them, S2 is the second arbitrary order proximity matrix, A2 is the chain enterprise similarity matrix, A2 i represents the i-th order of A2, w i =β i , β i ∈(0,1), i takes values ​​in [1, n], and n is the preset order.

[0118] It should be noted that the method of determining the second arbitrary order proximity matrix based on the chain enterprise similarity matrix is ​​the same as the method of determining the first arbitrary order proximity matrix based on the core enterprise similarity matrix, which will not be repeated here.

[0119] S243. Obtain a reconstructed correlation relationship matrix based on the core enterprise neighboring network embedding matrix and the chain enterprise neighboring network embedding matrix.

[0120] Among them, the core enterprise neighboring network embedding matrix SCD can be obtained according to the core enterprise similarity matrix SC, and the chain enterprise neighboring network embedding matrix SUD can be obtained according to the chain enterprise similarity matrix, and further according to the core enterprise neighboring network embedding matrix SCD and the chain enterprise neighboring network embedding matrix SUD, the reconstructed association relationship matrix CU is obtained.

[0121] Optionally, the association relationship matrix CU may be obtained by calculating the product of the core enterprise neighboring network embedding matrix SCD and the chain enterprise neighboring network embedding matrix SUD, and determining the product as the reconstructed association relationship matrix.

[0122] By vector-multiplying the core enterprise neighboring network embedding matrix SCD and the chain enterprise neighboring network embedding matrix SUD, we can finally obtain the correlation matrix CU of the core enterprise and the chain enterprise. According to the ranking in the obtained correlation matrix, we can know the chain enterprises most related to the core enterprise.

[0123] S250. According to the risk prediction results corresponding to each enterprise, an initial core enterprise risk vector and an initial chain enterprise risk vector are obtained.

[0124] In step S221, the risk prediction results corresponding to each core enterprise and each chain enterprise can be obtained. Further, the initial core enterprise risk vector CU is obtained according to the risk prediction results of the core enterprise, and the initial chain enterprise risk vector UD is obtained according to the output chain enterprise risk prediction results.

[0125] The method of obtaining the corresponding enterprise risk vector from the corresponding risk prediction result may be to perform a correlation matrix transformation or a correlation function calculation, and the specific method is not limited here.

[0126] S251. Based on the correlation matrix, the initial core enterprise risk vector and the initial chain enterprise risk vector, respectively determine the final core enterprise risk vector and the final chain enterprise risk vector.

[0127] Based on the association matrix CU obtained in S243 and the initial core enterprise risk vector UD and the initial chain enterprise risk vector CD obtained in S250, relevant vector operations are performed to obtain the final core enterprise risk vector SCD and the chain enterprise risk vector SUD, thereby obtaining the chain enterprise that can represent the degree of association with the core enterprise.

[0128] Optionally, the following formula (19) and formula (20) are used to determine the final core enterprise risk vector and the final chain enterprise risk vector respectively:

[0129] SCD z =UD⊙CU+CD (19)

[0130] SUD z =CD⊙CU -1 +UD (20)

[0131] Among them, SCD z For the final core enterprise risk vector, SUD z is the final chain enterprise risk vector, UD is the initial chain enterprise risk vector, CD is the initial core enterprise risk vector, and CU is the association relationship matrix.

[0132] The enterprise risk prediction scheme provided by this embodiment takes into account the characteristic data corresponding to each enterprise, and uses multiple risk prediction sub-models to extract characteristic data from different dimensions, so that multiple enterprises with risks in different dimensions can be mined from point to surface. The scheme provided by this embodiment can be used to mine the relationship between core enterprises and chain enterprises, and the risk customer groups of core enterprises and chain enterprises with multi-attribute characteristics can be accurately and quickly obtained. This scheme takes into account the multi-faceted comparison of the characteristic data corresponding to each enterprise, so as to determine the risky enterprises, and determine the risky enterprise group by obtaining the correlation between the core enterprise and the chain enterprise. Compared with the traditional risk assessment of enterprises through a single model, our considerations are more comprehensive, so the accuracy of predicting the risks of each enterprise is better.

[0133] Embodiment 3

[0134] Figure 3Schematic diagram of the structure of a device for predicting enterprise risk provided by the third embodiment of the present invention. This embodiment is applicable to the case of predicting enterprise risk. The device can be implemented in software and / or hardware, such as Figure 3 As shown, the device of the enterprise risk prediction device specifically includes: a feature data acquisition module 30, a first risk result acquisition module 31, a relationship matrix acquisition module 32 and a second risk result acquisition module 33, wherein:

[0135] The characteristic data acquisition module 30 is used to acquire characteristic data corresponding to each enterprise in the enterprise set; wherein the enterprise set includes at least one core enterprise and at least one chain enterprise;

[0136] The first risk result obtaining module 31 is used to obtain, for each of the enterprises, a risk prediction result corresponding to the current enterprise based on the characteristic data corresponding to the current enterprise and a pre-trained risk prediction model, wherein the risk prediction result represents the probability that the current enterprise is a risky enterprise;

[0137] A relationship matrix acquisition module 32 is used to acquire an initial relationship matrix between each of the core enterprises and each of the chain enterprises, and reconstruct the initial relationship matrix using an arbitrary order neighboring network to obtain a reconstructed correlation relationship matrix;

[0138] The second risk result obtaining module 33 is used to obtain a core enterprise risk prediction result set and a chain enterprise risk prediction result set based on the association relationship matrix and the risk prediction results corresponding to each of the enterprises.

[0139] The enterprise risk prediction device provided in this embodiment first inputs the acquired characteristic data corresponding to each enterprise into the pre-trained risk prediction model to obtain the risk prediction result corresponding to the current enterprise. The risk prediction result characterizes the probability that the current enterprise is a risk enterprise, thereby clarifying the risk probability of each core enterprise and each chain enterprise; then obtains the initial relationship matrix of each core enterprise and each chain enterprise, and reconstructs the initial relationship matrix using an arbitrary order neighboring network to obtain the reconstructed association relationship matrix, which can clarify the closeness of the relationship between each core enterprise and each chain enterprise; based on the association relationship matrix and the risk prediction results corresponding to each enterprise, obtain the core enterprise risk prediction result set and the chain enterprise risk prediction result set, that is, based on the association relationship between the core enterprise and the chain enterprise, the risk prediction result set of each enterprise output by the risk prediction model is optimized. Using the above technical solution, the accuracy of enterprise risk assessment can be improved, helping enterprises avoid related enterprises with higher risks, thereby achieving the technical effect of obtaining high-quality cooperative enterprises.

[0140] Optionally, the first risk result obtaining module 31 includes: a training unit and a prediction result output unit, wherein:

[0141] A training unit, used to input the characteristic data corresponding to the current enterprise into the 1st to nth sub-models included in the pre-trained risk prediction model; wherein n is an integer greater than 1; and also used to input the result data output by the n sub-models and the characteristic data corresponding to the current enterprise into the n+1th sub-model included in the risk prediction model;

[0142] The prediction result output unit is used to obtain the risk prediction result corresponding to the current enterprise output by the n+1th sub-model.

[0143] The above-mentioned product can execute the method provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0144] Optionally, each sub-model in the 1st to nth sub-models is a logistic regression model, a regression tree model, a support vector machine model, a linear classification model, or a decision tree model; and the n+1th model is a random forest model.

[0145] Optionally, the training unit includes: an acquisition subunit and a training subunit, wherein:

[0146] An acquisition subunit is used to acquire a training sample set;

[0147] A training sub-unit is used to use part of the training samples in the training sample set to train the 1st to nth sub-models; and is also used to use another part of the training samples in the training sample set to train the risk prediction model including the trained 1st to nth sub-models and the n+1th sub-model to be trained.

[0148] Optionally, the relationship matrix acquisition module 32 includes: an acquisition unit, a determination unit and an acquisition unit; wherein:

[0149] An acquisition unit, used to acquire a Gaussian matrix of core enterprises and a Gaussian matrix of chain enterprises corresponding to the initial relationship matrix;

[0150] A determination unit, used to determine a core enterprise similarity matrix according to the core enterprise Gaussian matrix, and to determine a chain enterprise similarity matrix according to the chain enterprise Gaussian matrix; and also used to determine a core enterprise neighbor network embedding matrix based on the core enterprise similarity matrix and an arbitrary order neighbor network, and to determine a chain enterprise neighbor network embedding matrix based on the chain enterprise similarity matrix and an arbitrary order neighbor network;

[0151] The obtaining unit is used to obtain a reconstructed association relationship matrix according to the core enterprise neighboring network embedding matrix and the chain enterprise neighboring network embedding matrix.

[0152] Optionally, the determination unit is also used to determine a core enterprise semantic matrix containing similarity values ​​between core enterprises, and determine the core enterprise similarity matrix based on the core enterprise semantic matrix and the core enterprise Gaussian matrix; determine a chain enterprise functional similarity matrix containing similarity values ​​between chain enterprises, and determine the chain enterprise similarity matrix based on the chain enterprise functional similarity matrix and the chain enterprise Gaussian matrix.

[0153] Optionally, the determining unit includes: a determining subunit and a selecting subunit, wherein:

[0154] A determination subunit, used to determine a first arbitrary order proximity matrix based on the core enterprise similarity matrix;

[0155] A selection subunit is used to perform a singular value decomposition in descending order on the first arbitrary order neighboring matrix, and select the first r eigenvalues ​​of the characteristic matrix after the current singular value decomposition in descending order; wherein r is an integer greater than 1;

[0156] The determination subunit is further used to reconstruct the first main characteristic matrix of singular value decomposition through the first r eigenvalues ​​currently selected, and determine the first main characteristic matrix as the core enterprise neighboring network embedding matrix;

[0157] Optionally, the determination subunit is further used to determine a second arbitrary order proximity matrix based on the chain enterprise similarity matrix;

[0158] The selection subunit is further used to perform a descending singular value decomposition on the second arbitrary order neighboring matrix, and select the first r eigenvalues ​​in the characteristic matrix after the current singular value decomposition is descended;

[0159] The subunit is determined, and the second main characteristic matrix of the singular decomposition is reconstructed through the first r eigenvalues ​​currently selected, and the second main characteristic matrix is ​​determined as the chain enterprise neighboring network embedding matrix.

[0160] Optionally, the first arbitrary order neighbor matrix is ​​determined by the following formula 1, and the second arbitrary order neighbor matrix is ​​determined by the following formula 2:

[0161] Formula 1: S1=w1A1+w2A1 2 +…+w n A1 n ;

[0162] Formula 2: S2 = w1A2 + w2A2 2 +…+w n A2 n ;

[0163] Among them, S1 is the first arbitrary order proximity matrix, S2 is the second arbitrary order proximity matrix, A1 is the core enterprise similarity matrix, A2 is the chain enterprise similarity matrix, A1 irepresents the i-th order of A1, A2 i represents the i-th order of A2, w i =β i , β i ∈(0,1), i takes values ​​in [1, n], and n is the preset order.

[0164] Optionally, the determination unit is further used to calculate the product result of the core enterprise neighboring network embedding matrix and the chain enterprise neighboring network embedding matrix, and determine the product result as the reconstructed association relationship matrix.

[0165] Optionally, the second risk result obtaining module includes: an initial vector obtaining unit and a final vector determining unit, wherein:

[0166] An initial vector obtaining unit, used to obtain an initial core enterprise risk vector and an initial chain enterprise risk vector according to the risk prediction results corresponding to each of the enterprises; wherein the initial core enterprise risk vector includes the risk prediction results of each core enterprise, and the initial chain enterprise risk vector includes the risk prediction results of each chain enterprise;

[0167] The final vector determination unit is used to determine the final core enterprise risk vector and the final chain enterprise risk vector respectively based on the association relationship matrix, the initial core enterprise risk vector and the initial chain enterprise risk vector.

[0168] Optionally, the final core enterprise risk vector is determined by the following formula 3, and the final chain enterprise risk vector is determined by the following formula 4:

[0169] Formula 3: SCD z =UD⊙CU+CD;

[0170] Formula 4: SUD z =CD⊙CU -1 +UD;

[0171] Among them, SCD z For the final core enterprise risk vector, SUD z is the final chain enterprise risk vector, UD is the initial chain enterprise risk vector, CD is the initial core enterprise risk vector, and CU is the association relationship matrix.

[0172] Optionally, the characteristic data of the enterprise includes: at least one of the enterprise's capital trend data, scale data, financial indicator data, and compliance indicator data; and / or,

[0173] The element value in the i-th row and j-th column of the initial relationship matrix indicates whether there is an association relationship between the i-th core enterprise and the j-th chain enterprise.

[0174] The enterprise risk prediction device provided by the embodiment of the present invention can execute the enterprise risk prediction method provided by any of the above embodiments, and has the corresponding functional modules and beneficial effects of the execution method.

[0175] Embodiment 4

[0176] Figure 4 This is a schematic diagram of the structure of a computer device in Embodiment 4 of the present invention. Figure 4 A block diagram of an exemplary computer device 12 suitable for use in implementing embodiments of the present invention is shown. Figure 4 The computer device 12 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0177] like Figure 4 As shown, the computer device 12 is in the form of a general-purpose computing device. The components of the computer device 12 may include, but are not limited to: one or more processors or processing units 16, a system memory 28, and a bus 18 that connects various system components (including the system memory 28 and the processing unit 16).

[0178] Bus 18 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processor or a local bus using any of a variety of bus architectures. By way of example, these architectures include, but are not limited to, an Industry Standard Architecture (ISA) bus, a Micro Channel Architecture (MAC) bus, an Enhanced ISA bus, a Video Electronics Standards Association (VESA) local bus, and a Peripheral Component Interconnect (PCI) bus.

[0179] The computer device 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the computer device 12, including volatile and non-volatile media, removable and non-removable media.

[0180] The system memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache memory 32. The computer device 12 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, the storage system 34 may be used to read and write non-removable, non-volatile magnetic media ( Figure 4 not shown, usually called a "hard drive"). Although Figure 4Not shown in the figure, a disk drive for reading and writing to a removable non-volatile disk (e.g., a "floppy disk"), and an optical disk drive for reading and writing to a removable non-volatile optical disk (e.g., a CD-ROM, a DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to the bus 18 via one or more data medium interfaces. The memory 28 may include at least one program product having a set (e.g., at least one) of program modules that are configured to perform the functions of the various embodiments of the present invention.

[0181] A program / utility 40 having a set (at least one) of program modules 42 may be stored, for example, in the memory 28. Such program modules 42 include, but are not limited to, an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment. The program modules 42 generally perform the functions and / or methods of the embodiments described herein.

[0182] The computer device 12 may also communicate with one or more external devices 14 (e.g., keyboard, pointing device, display 24, etc.), may also communicate with one or more devices that enable a user to interact with the computer device 12, and / or communicate with any device that enables the computer device 12 to communicate with one or more other computing devices (e.g., network card, modem, etc.). Such communication may be performed through an input / output (I / O) interface 22. In addition, the computer device 12 in this embodiment, the display 24 does not exist as an independent individual, but is embedded in the mirror surface, and when the display surface of the display 24 is not displayed, the display surface of the display 24 and the mirror surface are visually integrated. In addition, the computer device 12 may also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN) and / or public network, such as the Internet) through a network adapter 20. As shown in the figure, the network adapter 20 communicates with other modules of the computer device 12 through a bus 18. It should be understood that although not shown in the figures, other hardware and / or software modules may be used in conjunction with the computer device 12, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0183] The processing unit 16 executes various functional applications and data processing by running programs stored in the system memory 28, such as implementing the enterprise risk prediction method provided in the embodiment of the present invention.

[0184] The computer device provided in the embodiment of the present invention can execute the enterprise risk prediction method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects for executing the method.

[0185] Embodiment 5

[0186] Embodiment 5 of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the enterprise risk prediction method provided by all the embodiments of the present invention: acquiring characteristic data corresponding to each enterprise in an enterprise set; wherein the enterprise set includes at least one core enterprise and at least one chain enterprise;

[0187] For each of the enterprises, based on the characteristic data corresponding to the current enterprise and the pre-trained risk prediction model, a risk prediction result corresponding to the current enterprise is obtained, wherein the risk prediction result represents the probability that the current enterprise is a risky enterprise;

[0188] Obtaining an initial relationship matrix between each of the core enterprises and each of the chain enterprises, and reconstructing the initial relationship matrix using an arbitrary order proximity network to obtain a reconstructed association relationship matrix;

[0189] Based on the association relationship matrix and the risk prediction results corresponding to each of the enterprises, a core enterprise risk prediction result set and a chain enterprise risk prediction result set are obtained.

[0190] Any combination of one or more computer-readable media can be used. Computer-readable media can be computer-readable signal media or computer-readable storage media. Computer-readable storage media can be, for example, but not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or devices, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: electrical connections with one or more wires, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above. In this document, computer-readable storage media can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.

[0191] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0192] The program code embodied on the computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0193] Computer program code for performing the operation of the present invention may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code 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 and 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., via the Internet using an Internet service provider).

[0194] Note that the above are only preferred embodiments of the present invention and the technical principles used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. A method for predicting enterprise risks, characterized in that: The method comprises: Acquire characteristic data corresponding to each enterprise in the enterprise set; wherein the enterprise set includes at least one core enterprise and at least one chain enterprise; For each of the enterprises, based on the characteristic data corresponding to the current enterprise and the pre-trained risk prediction model, a risk prediction result corresponding to the current enterprise is obtained, wherein the risk prediction result represents the probability that the current enterprise is a risky enterprise; Obtaining an initial relationship matrix between each of the core enterprises and each of the chain enterprises, and reconstructing the initial relationship matrix using an arbitrary order proximity network to obtain a reconstructed association relationship matrix; Based on the association relationship matrix and the risk prediction results corresponding to each of the enterprises, a core enterprise risk prediction result set and a chain enterprise risk prediction result set are obtained; The initial relationship matrix is ​​reconstructed using an arbitrary order neighboring network to obtain a reconstructed association relationship matrix, including: Obtaining a Gaussian matrix of core enterprises and a Gaussian matrix of chain enterprises corresponding to the initial relationship matrix; Determine a core enterprise semantic matrix including similarity values ​​between core enterprises, and determine a core enterprise similarity matrix according to the core enterprise semantic matrix and the core enterprise Gaussian matrix; Determine a chain enterprise function similarity matrix including similarity values ​​between each chain enterprise, and determine a chain enterprise similarity matrix according to the chain enterprise function similarity matrix and the chain enterprise Gaussian matrix; Determine a first arbitrary order proximity matrix based on the core enterprise similarity matrix; Performing a descending singular value decomposition on the first arbitrary order neighboring matrix, and selecting the first r eigenvalues ​​of the eigenmatrix after the current singular value decomposition is descended; wherein r is an integer greater than 1; Reconstruct the first principal characteristic matrix of singular value decomposition by using the first r eigenvalues ​​currently selected, and determine the first principal characteristic matrix as the core enterprise neighboring network embedding matrix; Determine a second arbitrary order proximity matrix based on the chain enterprise similarity matrix; Performing a descending singular value decomposition on the second arbitrary order neighboring matrix, and selecting the first r eigenvalues ​​of the eigenvalue matrix after the current singular value decomposition is descended; Reconstruct the second main characteristic matrix of singular decomposition through the first r eigenvalues ​​currently selected, and determine the second main characteristic matrix as the chain enterprise neighboring network embedding matrix; The product result of the core enterprise neighboring network embedding matrix and the chain enterprise neighboring network embedding matrix is ​​calculated, and the product result is determined as the reconstructed association relationship matrix.

2. The method according to claim 1, characterized in that Based on the characteristic data corresponding to the current enterprise and the pre-trained risk prediction model, the risk prediction results corresponding to the current enterprise are obtained, including: Input the characteristic data corresponding to the current enterprise into the 1st to nth sub-models contained in the pre-trained risk prediction model; wherein n is an integer greater than 1; Input the result data outputted by the n sub-models and the characteristic data corresponding to the current enterprise into the n+1th sub-model included in the risk prediction model; Obtain the risk prediction result corresponding to the current enterprise output by the n+1th sub-model.

3. The method according to claim 2, characterized in that Each sub-model in the first to nth sub-models is a logistic regression model, a regression tree model, a support vector machine model, a linear classification model, or a decision tree model; The n+1th model is a random forest model.

4. The method according to claim 2, characterized in that: The training method of the risk prediction model includes: Obtain a training sample set; Using some training samples in the training sample set, training the first to nth sub-models; Another part of the training samples in the training sample set is used to train the risk prediction model including the trained 1st to nth sub-models and the n+1th sub-model to be trained.

5. The method according to claim 1, characterized in that: The first arbitrary order neighbor matrix is ​​determined by the following formula 1, and the second arbitrary order neighbor matrix is ​​determined by the following formula 2: Formula 1: S1=w1A1+w2A1 2 +…+w n A1 n ; Formula 2: S2 = w1A2 + w2A2 2 +…+w n A2 n ; Among them, S1 is the first arbitrary order proximity matrix, S2 is the second arbitrary order proximity matrix, A1 is the core enterprise similarity matrix, A2 is the chain enterprise similarity matrix, A1 i represents the i-th order of A1, A2 i represents the i-th order of A2, w i =β i , β i ∈(0,1), i takes values ​​in [1,n], and n is the preset order.

6. The method according to claim 1, characterized in that Based on the association relationship matrix and the risk prediction results corresponding to each of the enterprises, a core enterprise risk prediction result set and a chain enterprise risk prediction result set are obtained, including: According to the risk prediction results corresponding to each of the enterprises, an initial core enterprise risk vector and an initial chain enterprise risk vector are obtained; wherein the initial core enterprise risk vector includes the risk prediction results of each core enterprise, and the initial chain enterprise risk vector includes the risk prediction results of each chain enterprise; Based on the association relationship matrix, the initial core enterprise risk vector and the initial chain enterprise risk vector, a final core enterprise risk vector and a final chain enterprise risk vector are determined respectively.

7. The method according to claim 6, characterized in that The final core enterprise risk vector is determined by the following formula 3, and the final chain enterprise risk vector is determined by the following formula 4: Formula 3: SCD z =UD ⊙ CU+CD; Formula 4: SUD z =CD ⊙ CU- 1 +UD; Among them, SCD z For the final core enterprise risk vector, SUD z is the final chain enterprise risk vector, UD is the initial chain enterprise risk vector, CD is the initial core enterprise risk vector, and CU is the association relationship matrix.

8. The method according to any one of claims 1 to 7, characterized in that The characteristic data of the enterprise includes: at least one of the enterprise's capital trend data, scale data, financial indicator data, and compliance indicator data; and / or, The element value in the i-th row and j-th column of the initial relationship matrix indicates whether there is an association relationship between the i-th core enterprise and the j-th chain enterprise.

9. An enterprise risk prediction device, characterized in that: include: A feature data acquisition module, used to acquire feature data corresponding to each enterprise in the enterprise set; wherein the enterprise set includes at least one core enterprise and at least one chain enterprise; A first risk result obtaining module is used to obtain, for each of the enterprises, a risk prediction result corresponding to the current enterprise based on the characteristic data corresponding to the current enterprise and a pre-trained risk prediction model, wherein the risk prediction result represents the probability that the current enterprise is a risky enterprise; A relationship matrix acquisition module is used to obtain the initial relationship matrix between each of the core enterprises and each of the chain enterprises, and reconstruct the initial relationship matrix using an arbitrary order neighboring network to obtain a reconstructed correlation relationship matrix; A second risk result acquisition module is used to obtain a core enterprise risk prediction result set and a chain enterprise risk prediction result set based on the association relationship matrix and the risk prediction results corresponding to each of the enterprises; The relationship matrix acquisition module is specifically used to obtain the core enterprise Gaussian matrix and the chain enterprise Gaussian matrix corresponding to the initial relationship matrix; determine the core enterprise semantic matrix containing similarity values ​​between each core enterprise, and determine the core enterprise similarity matrix according to the core enterprise semantic matrix and the core enterprise Gaussian matrix; determine the chain enterprise functional similarity matrix containing similarity values ​​between each chain enterprise, and determine the chain enterprise similarity matrix according to the chain enterprise functional similarity matrix and the chain enterprise Gaussian matrix; determine the first arbitrary order neighbor matrix based on the core enterprise similarity matrix; perform singular value decomposition descending on the first arbitrary order neighbor matrix, and select the first r eigenvalues ​​in the characteristic matrix after the current singular value decomposition descending; wherein r is an integer greater than 1; reconstructing the first principal characteristic matrix of singular value decomposition through the first r eigenvalues ​​currently selected, and determining the first principal characteristic matrix as the core enterprise neighbor network embedding matrix; determining the second arbitrary order neighbor matrix based on the chain enterprise similarity matrix; performing singular value decomposition in descending order on the second arbitrary order neighbor matrix, and selecting the first r eigenvalues ​​in the characteristic matrix after the current singular value decomposition in descending order; reconstructing the second principal characteristic matrix of singular decomposition through the first r eigenvalues ​​currently selected, and determining the second principal characteristic matrix as the chain enterprise neighbor network embedding matrix; calculating the product result of the core enterprise neighbor network embedding matrix and the chain enterprise neighbor network embedding matrix, and determining the product result as the reconstructed association relationship matrix.

10. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the enterprise risk prediction method as described in any one of claims 1-8 is implemented.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the enterprise risk prediction method as described in any one of claims 1-8 is implemented.

Images