Post-loan management method and system based on big data

Abstract

The invention relates to the technical field of post-loan management, and particularly discloses a post-loan management method and system based on big data, and the method comprises the steps: carrying out the semantic analysis of collateral state data and enterprise parameters of a target enterprise through employing an artificial intelligence technology based on deep learning, so as to capture the state characteristics of collateral and the operation state characteristics of the enterprise, and external knowledge is introduced, and fine-grained semantic guidance interactive fusion is performed on the two, so that potential risk associated information of the enterprise is mined, and intelligent assessment of the risk level of the target enterprise is realized. In this way, the accuracy of enterprise post-loan risk assessment can be effectively improved, the maintenance cost is reduced, and the risk management efficiency of financial institutions is improved.

Description

Technical Field

[0001] The present application relates to the technical field of post-loan management, and more specifically, to a post-loan management method and system based on big data. Background Art

[0002] With the increasing complexity of economic activities and the rapid development of financial markets, post-loan management has become an important means for financial institutions to control credit risks and ensure the safety of funds. Post-loan management refers to the process in which financial institutions continuously monitor factors such as the borrower's repayment ability, operating conditions, and market environment in order to ensure the safe recovery of loans after the loan is issued. In this process, accurately assessing and predicting the borrower's default risk is of great significance to reducing the credit losses of financial institutions.

[0003] Existing post-loan management methods mostly rely on traditional statistical models or expert systems. Although these methods can help financial institutions identify and manage risks to a certain extent, they have some limitations. For example, traditional models are often built based on the borrower's historical credit data and may not be able to reflect the borrower's latest operating status changes in a timely manner; in addition, such models usually focus on the quantitative analysis of financial indicators, while ignoring other important factors that may affect the borrower's credit status, such as market trends, industry dynamics, etc.

[0004] In recent years, with the development of big data technology and artificial intelligence, more and more research has begun to explore how to use advanced algorithms and technologies to improve the effectiveness of post-loan management. For example, the invention patent with publication number CN115271914A discloses a post-loan management method, which establishes a risk assessment model based on the industry to which the target enterprise belongs, uses a pre-trained risk assessment model, and assesses the risk probability of the target enterprise based on the status data of the target enterprise's collateral and enterprise parameters.

[0005] Although the above post-loan management methods have improved the accuracy of risk assessment to a certain extent, they still have some limitations. For example, in order to ensure the continued effectiveness of the risk assessment model, it is usually necessary to regularly update the risk assessment model to adapt to market changes and industry trends, which is not only time-consuming and labor-intensive but also costly.

[0006] Therefore, an optimized post-loan management method and system based on big data is expected. Summary of the invention

[0007] The present application provides a post-loan management method and system based on big data, which can effectively improve the accuracy of enterprise post-loan risk assessment, reduce maintenance costs, and improve the risk management efficiency of financial institutions.

[0008] First, a post-loan management method based on big data is provided, including:

[0009] A target enterprise information acquisition module, used to acquire the status data of the collateral of the target enterprise and the enterprise parameters of the target enterprise, wherein the enterprise parameters include conventional parameters and characteristic parameters;

[0010] A semantic embedding coding module, used to perform semantic embedding coding on the state data of the collateral, the conventional parameters and the characteristic parameters respectively to obtain a collateral state semantic embedding coding vector, a conventional parameter semantic embedding coding vector and a characteristic parameter semantic embedding coding vector;

[0011] An enterprise parameter information fusion module, used for performing significant fusion on the conventional parameter semantic embedding coding vector and the characteristic parameter semantic embedding coding vector to obtain an enterprise parameter semantic embedding significant fusion coding vector;

[0012] An information-guided interaction module, used for performing fine-grained feature interaction based on external knowledge guidance on the collateral status semantic embedding coding vector and the enterprise parameter semantic embedding significant fusion coding vector to obtain a collateral status-enterprise parameter fine-grained interaction feature vector;

[0013] The target enterprise risk assessment module is used to determine the risk assessment result of the target enterprise based on the collateral status-enterprise parameter fine-grained interactive feature vector.

[0014] The present application provides a post-loan management method and system based on big data, which uses artificial intelligence technology based on deep learning to perform semantic analysis on the collateral status data and enterprise parameters of the target enterprise to capture the status characteristics of the collateral and the operating status characteristics of the enterprise, and introduces external knowledge to conduct fine-grained semantic guidance interaction and fusion of the two to dig out the potential risk-related information of the enterprise, thereby realizing intelligent assessment of the risk level of the target enterprise. In this way, the accuracy of the enterprise's post-loan risk assessment can be effectively improved, the maintenance cost can be reduced, and the risk management efficiency of financial institutions can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings of the embodiments of the present application are briefly introduced below. Obviously, the drawings described below only relate to some embodiments of the present application, and are not intended to limit the present application.

[0016] Figure 1 This is a schematic block diagram of a post-loan management system based on big data according to an embodiment of the present application.

[0017] Figure 2 This is a data flow diagram of the big data-based post-loan management system of an embodiment of the present application.

[0018] Figure 3This is a schematic block diagram of the information guidance interaction module in the big data-based post-loan management system of an embodiment of the present application.

[0019] Figure 4 This is a schematic block diagram of a feature-guided modulation and fusion unit in a big data-based post-loan management system according to an embodiment of the present application.

[0020] Figure 5 This is a schematic flow chart of the post-loan management method based on big data according to an embodiment of the present application. DETAILED DESCRIPTION

[0021] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work also fall within the scope of protection of the present application.

[0022] In response to the above technical problems, the technical concept of this application is: to use artificial intelligence technology based on deep learning to perform semantic analysis on the collateral status data and enterprise parameters of the target enterprise to capture the status characteristics of the collateral and the operating status characteristics of the enterprise, and to introduce external knowledge to conduct fine-grained semantic guidance interaction and fusion of the two to dig out the potential risk-related information of the enterprise, thereby realizing intelligent assessment of the risk level of the target enterprise. In this way, the accuracy of the enterprise's post-loan risk assessment can be effectively improved, and the maintenance cost can be reduced, thereby improving the risk management efficiency of financial institutions.

[0023] like Figure 1 and Figure 2As shown, the post-loan management system based on big data includes: a target enterprise information acquisition module 10, which is used to obtain the status data of the collateral of the target enterprise, and the enterprise parameters of the target enterprise, wherein the enterprise parameters include conventional parameters and characteristic parameters; a semantic embedding coding module 20, which is used to perform semantic embedding coding on the status data of the collateral, the conventional parameters and the characteristic parameters respectively to obtain a collateral status semantic embedding coding vector, a conventional parameter semantic embedding coding vector and a characteristic parameter semantic embedding coding vector; an enterprise parameter information fusion module 30, which is used to perform significant fusion on the conventional parameter semantic embedding coding vector and the characteristic parameter semantic embedding coding vector to obtain an enterprise parameter semantic embedding significant fusion coding vector; an information guided interaction module 40, which is used to perform fine-grained feature interaction based on external knowledge guidance on the collateral status semantic embedding coding vector and the enterprise parameter semantic embedding significant fusion coding vector to obtain a collateral status-enterprise parameter fine-grained interaction feature vector; a target enterprise risk assessment module 50, which is used to determine the risk assessment result of the target enterprise based on the collateral status-enterprise parameter fine-grained interaction feature vector.

[0024] Exemplarily, in the target enterprise information acquisition module 10, the status data of the collateral of the target enterprise and the enterprise parameters of the target enterprise are obtained, and the enterprise parameters include conventional parameters and characteristic parameters. It should be understood that the value of the collateral can be used as a guarantee for the loan. Therefore, by collecting the status data of the collateral of the target enterprise, the recovery risk of the loan can be effectively evaluated. In the embodiment of the present application, for real estate collateral, the status data of the collateral includes but is not limited to the location, area, usage, property status, etc. of the collateral; for movable collateral, it includes but is not limited to the type, quantity, quality, market value, etc. of the collateral. At the same time, the operating status of the enterprise is another important factor affecting the risk of loan recovery. The conventional parameters of the enterprise, such as financial statement data, tax information and social insurance payment status, can fully reflect the financial health of the enterprise; and the specific parameters are determined according to the characteristics of the industry to which the target enterprise belongs. For example, if the target enterprise belongs to the manufacturing industry, its characteristic parameters may include production scale, material cost expenditure, etc. If the target enterprise belongs to the retail industry, its characteristic parameters may include inventory turnover rate, sales growth rate, etc. Comprehensive analysis of a company's conventional and characteristic parameters helps to more comprehensively assess the company's operating conditions and repayment capacity.

[0025] Exemplarily, in the semantic embedding coding module 20, the state data of the collateral, the conventional parameters and the characteristic parameters are semantically embedded and coded respectively to obtain the collateral state semantic embedding coding vector, the conventional parameter semantic embedding coding vector and the characteristic parameter semantic embedding coding vector. It should be understood that considering that the state data of the collateral and the enterprise parameters contain a large amount of unstructured information with different data types and formats, it is difficult to be directly used for risk assessment. Therefore, in order to unify the data format and extract useful information from it, the present application adopts the semantic embedding coding technology to process the state data of the collateral, the conventional parameters and the characteristic parameters respectively to capture the semantic features of the original data and map it into a vector representation in a high-dimensional semantic feature space, thereby obtaining the collateral state semantic embedding coding vector, the conventional parameter semantic embedding coding vector and the characteristic parameter semantic embedding coding vector. In a specific example of the present application, the Word2Vec model is used to train and generate the collateral state data embedding coding matrix and the parameter embedding coding matrix to perform the above-mentioned semantic embedding coding task, thereby realizing a deep semantic understanding of the collateral state and enterprise parameters.

[0026] In another specific example of the present application, for the status data of the collateral (such as the location, area, usage of real estate, etc. or the type, quantity, market value, etc. of movable property), it is first necessary to convert these unstructured or semi-structured information into a form that can be understood by the machine learning model. For example, an embedding matrix can be generated by Word2Vec model training to represent different types of collateral attributes. Each attribute (such as location, area) will be mapped into a vector, and these vectors together constitute the semantic embedding encoding vector of the collateral status. The conventional parameters of an enterprise usually include financial statement data, tax information, and social insurance payment status. These data also need to be processed before they can be used for subsequent analysis. One method is to encode the text description using a pre-trained language model (such as BERT), or to directly input the numerical data into the neural network layer for encoding after normalization. Each conventional parameter will eventually be converted into a vector, and all these vectors are combined to form a conventional parameter semantic embedding encoding vector. Specific parameters determined according to the characteristics of the industry to which the target enterprise belongs, such as the production scale, material cost expenditure of the manufacturing industry, or the inventory turnover rate and sales growth rate of the retail industry. The processing method for these parameters is similar to that of conventional parameters, but industry-specific factors may need to be considered. In this embodiment, the selected feature extraction method includes: convolutional neural network (CNN) or recurrent neural network (RNN), and these parameters can be converted into characteristic parameter semantic embedding coding vectors through feature extraction.

[0027] Exemplarily, in the enterprise parameter information fusion module 30, the conventional parameter semantic embedding coding vector and the characteristic parameter semantic embedding coding vector are significantly fused to obtain the enterprise parameter semantic embedding significantly fused coding vector. It should be understood that, considering that the conventional parameters and characteristic parameters respectively reflect different aspects of information of the enterprise, in order to more comprehensively evaluate the comprehensive operating status of the enterprise, the present application further semantically fuses the conventional parameter semantic embedding coding vector and the characteristic parameter semantic embedding coding vector to capture the multi-dimensional characteristics of enterprise operation. In one embodiment, the enterprise parameter information fusion module is used to: input the conventional parameter semantic embedding coding vector and the characteristic parameter semantic embedding coding vector into a soft linear weighted fusion unit to obtain the enterprise parameter semantic embedding significantly fused coding vector. That is, a soft linear weighted fusion unit is used to fuse the conventional parameter semantic embedding coding vector and the characteristic parameter semantic embedding coding vector, and by introducing an attention mechanism, adaptive attention is achieved to the characteristics of different enterprise operating states, so as to obtain the enterprise parameter semantic embedding significantly fused coding vector. Specifically, the soft linear weighted fusion unit replaces the Sigmoid function with the Softmax function to gate the input features on the basis of the SiLU activation function, and introduces the weight factor 1-Softmax before another input feature, and uses softmax(·) and 1-softmax(·) as attention weight vectors to better normalize the feature weights and achieve better feature selection and fusion effects. The specific calculation method of the soft linear weighted fusion unit is as follows:

[0028]

[0029] in, is the semantic embedding encoding vector of the conventional parameter, is the semantic embedding coding vector of the characteristic parameter, softmax(·) is the normalized exponential function, and V2 is the significant fusion coding vector of the enterprise parameter semantic embedding.

[0030] Exemplarily, in the information guided interaction module 40, the collateral status semantic embedding coding vector and the enterprise parameter semantic embedding significant fusion coding vector are subjected to fine-grained feature interaction based on external knowledge guidance to obtain the collateral status-enterprise parameter fine-grained interaction feature vector. It should be understood that, considering that in the prior art, in order to ensure the continued effectiveness of the risk assessment model, it is usually necessary to regularly update the risk assessment model to adapt to the market changes and industry trends of the industry to which the target enterprise belongs, which is not only time-consuming and labor-intensive but also costly. To this end, the present application proposes a feature interaction method based on external knowledge guidance, which introduces external knowledge of the industry to which the target enterprise belongs, such as industry reports, market analysis, policies and regulations, etc., to perform fine-grained semantically guided interactive fusion on the collateral status semantic embedding coding vector and the enterprise parameter semantic embedding significant fusion coding vector, so that the feature expression can be dynamically optimized according to the industry knowledge, effectively improving the accuracy of risk assessment.

[0031] In one embodiment, the information-guided interaction module 40 includes: a feature fine-grained interaction optimization unit 41, which is used to perform feature fine-grained interaction optimization on the collateral status semantic embedding coding vector and the enterprise parameter semantic embedding significant fusion coding vector based on external knowledge to obtain an external knowledge optimized collateral status-enterprise parameter feature fine-grained interaction matrix; a feature-guided modulation fusion unit 42, which is used to optimize the collateral status-enterprise parameter feature fine-grained interaction matrix based on the external knowledge, and perform feature-guided modulation fusion on the collateral status semantic embedding coding vector and the enterprise parameter semantic embedding significant fusion coding vector to obtain the collateral status-enterprise parameter fine-grained interaction feature vector.

[0032] Exemplarily, in the feature fine-grained interactive optimization unit 41, based on external knowledge, the collateral status semantic embedding encoding vector and the enterprise parameter semantic embedding significant fusion encoding vector are subjected to feature fine-grained interactive optimization to obtain an external knowledge optimized collateral status-enterprise parameter feature fine-grained interactive matrix. It should be understood that external knowledge can help the model to more accurately reflect the current market conditions and industry trends, so that the risk assessment model can be dynamically adjusted as the external environment changes, thereby maintaining its continued effectiveness. By introducing external knowledge, the model can capture other potential influencing factors beyond traditional financial indicators, which may not be directly reflected in the financial statements of the enterprise, but have an important impact on the operating conditions and repayment ability of the enterprise. Therefore, combining external knowledge can help the model to more comprehensively understand the risk status of the enterprise, thereby improving the accuracy of risk assessment. External knowledge helps to reveal the potential correlation between the collateral status and the operating status of the enterprise, which may not be intuitively obvious. For example, policy changes in a specific industry may indirectly affect the cash flow status or asset value of a certain enterprise, thereby affecting the risk level of the enterprise's loan. By utilizing external knowledge, the model can better identify such indirect but important connections. Incorporating external knowledge into the feature interaction process can also help reduce the risk of model overfitting, because it is equivalent to increasing the amount of information in the training data, making the model more robust and reliable when facing new situations.

[0033] In one embodiment, the feature fine-grained interaction optimization unit is used to: input the collateral status semantic embedding encoding vector and the enterprise parameter semantic embedding significant fusion encoding vector into the fine-grained feature interaction network to obtain the collateral status-enterprise parameter feature fine-grained interaction matrix; input the collateral status-enterprise parameter feature fine-grained interaction matrix into the external knowledge-based attention unit to obtain the external knowledge-optimized collateral status-enterprise parameter feature fine-grained interaction matrix. Specifically, the process can be expressed as:

[0034]

[0035] Where V1 represents the collateral status semantic embedding encoding vector, V2 represents the enterprise parameter semantic embedding saliency fusion encoding vector, (·) T represents the transpose of a vector, Represents matrix multiplication operation, M q represents the collateral status-enterprise parameter feature fine-grained interaction matrix, M k and M v represents the learnable memory parameter matrix of the attention unit based on external knowledge, norm(·) represents the normalization function, M y Representing external knowledge to optimize the collateral state-enterprise parameter feature fine-grained interaction matrix.

[0036] Specifically, first, the collateral status semantic embedding coding vector and the enterprise parameter semantic embedding significant fusion coding vector are input into the fine-grained feature interaction network to understand the correlation between the two from a micro level, capture the fine-grained interaction information between the two, and generate a collateral status-enterprise parameter feature fine-grained interaction matrix. Subsequently, the generated collateral status-enterprise parameter feature fine-grained interaction matrix is ​​sent to the attention unit based on external knowledge, and the external knowledge is used to further optimize the collateral status-enterprise parameter feature fine-grained interaction matrix to enhance the model's ability to understand enterprise risks. In specific implementation, the parameter matrix of the attention unit can be trained by constructing an industry knowledge base and extracting industry knowledge related to the target enterprise from the industry knowledge base, thereby helping the model to more accurately understand the risk status of the target enterprise, capture potential risks in enterprise operations in a larger range, and achieve feature optimization expression of the associated interaction information between the collateral status and enterprise parameters.

[0037] In one embodiment, the feature-guided modulation fusion unit 42 includes: a feature modulation subunit 421, which is used to optimize the collateral status-enterprise parameter feature fine-grained interaction matrix based on the external knowledge, and perform feature modulation optimization on the collateral status semantic embedding coding vector and the enterprise parameter semantic embedding significant fusion coding vector respectively to obtain the optimized collateral status semantic embedding coding vector and the optimized enterprise parameter semantic embedding significant fusion coding vector; a position-by-position semantic interaction fusion subunit 422, which is used to perform position-by-position semantic interaction fusion on the optimized collateral status semantic embedding coding vector and the optimized enterprise parameter semantic embedding significant fusion coding vector to obtain the collateral status-enterprise parameter fine-grained interaction feature vector.

[0038] In one embodiment, the feature modulation subunit 421 is used to: perform a linear transformation on the collateral status semantic embedding coding vector to obtain a first query feature vector and a first value feature vector and use the external knowledge optimized collateral status-enterprise parameter feature fine-grained interaction matrix as a key matrix, input the first query feature vector, the first value feature vector and the key matrix into a fine-grained modulation module based on a Transformer structure to obtain the optimized collateral status semantic embedding coding vector; perform a linear transformation on the enterprise parameter semantic embedding significant fusion coding vector to obtain a second query feature vector and a second value feature vector and use the external knowledge optimized fine-grained feature interaction matrix as a key matrix, input the second query feature vector, the second value feature vector and the key matrix into the fine-grained modulation module based on a Transformer structure to obtain the optimized enterprise parameter semantic embedding significant fusion coding vector. Specifically, the process can be expressed as follows:

[0039]

[0040] Among them, W 1q and W 1v denote the first query embedding matrix and the first value embedding matrix respectively, V 1q and V 1v denote the first query feature vector and the first value feature vector respectively, W 2q and W 2v denote the second query embedding matrix and the second value embedding matrix, V 2q and V 2v denote the second query feature vector and the second value feature vector, respectively, and b 1q 、b 1v 、b 2q and b 2v denote different bias terms respectively, softmax(·) denotes a normalized exponential function, d denotes the characteristic scale value of the external knowledge optimized fine-grained feature interaction matrix, V′1 and V′2 denote the optimized collateral status semantic embedding coding vector and the optimized enterprise parameter semantic embedding significant fusion coding vector respectively.

[0041] Specifically, the collateral status-enterprise parameter feature fine-grained interaction matrix optimized with external knowledge is used as the key matrix, and a first query feature vector and a first value feature vector are constructed based on the collateral status semantic embedding coding vector, and a second query feature vector and a second value feature vector are constructed based on the enterprise parameter semantic embedding significant fusion coding vector. The self-attention mechanism of the Transformer structure is used to realize information exchange and integration between internal and external knowledge of the features, thereby ensuring that the collateral status semantic embedding coding vector and the enterprise parameter semantic embedding significant fusion coding vector can obtain a deeper semantic understanding from external knowledge to improve the quality of their feature expression.

[0042] In one embodiment, the position-by-position semantic interaction fusion subunit 422 is used to calculate the position-by-position point division between the optimized collateral status semantic embedding coding vector and the optimized enterprise parameter semantic embedding significant fusion coding vector to obtain the collateral status-enterprise parameter fine-grained interaction feature vector. Specifically, the process can be expressed as:

[0043]

[0044] Among them, V′1 and V′2 represent the optimized collateral status semantic embedding encoding vector and the optimized enterprise parameter semantic embedding saliency fusion encoding vector, respectively. i Represents the collateral status-enterprise parameter fine-grained interaction feature vector.

[0045] That is, by performing semantic interaction fusion encoding on the optimized collateral status semantic embedding coding vector and the optimized enterprise parameter semantic embedding saliency fusion coding vector by positional division, a collateral status-enterprise parameter fine-grained interaction feature vector is obtained, thereby achieving a comprehensive representation of the collateral status characteristics and enterprise parameter characteristics.

[0046] Exemplarily, in the target enterprise risk assessment module 50, the risk assessment result of the target enterprise is determined based on the collateral status-enterprise parameter fine-grained interactive feature vector. In one embodiment, the target enterprise risk assessment module is used to: input the collateral status-enterprise parameter fine-grained interactive feature vector into the risk assessment module based on the classifier to obtain the risk assessment result, and the risk assessment result is a risk probability level label. It should be understood that the collateral status-enterprise parameter fine-grained interactive feature vector is input into the risk assessment module based on the classifier to obtain the risk assessment result, and a softmax multi-classifier is used, whose main function is to convert the complex features extracted in the previous step into a specific risk probability level label. This is done to intuitively quantify the default risk of the target enterprise and provide a standardized result to facilitate financial institutions to make decisions quickly. In the encoding process, first, the collateral status-enterprise parameter fine-grained interactive feature vector is linearly transformed through the fully connected layer to generate an unnormalized score vector. This process involves multiplying the collateral status-enterprise parameter fine-grained interactive feature vector with the weight matrix and adding a bias term to form a score for each possible risk level. Next, the softmax function is applied to these scores to convert them into probability distributions. The softmax function ensures that the sum of all output probabilities is equal to 1, and each probability value represents the likelihood of the corresponding risk level. Finally, the risk assessment result, i.e., the risk probability level label, is determined based on the category corresponding to the maximum probability output by the softmax function. For example, if the probability distribution output by the softmax function indicates that the probability of "medium risk" is the highest, the risk assessment result of the enterprise is labeled as "medium risk".

[0047] In the technical solution of the present application, the collateral status semantic embedding coding vector and the enterprise parameter semantic embedding significant fusion coding vector respectively represent the collateral status semantic embedding coding features and the semantic joint coding features of enterprise conventional parameters and enterprise characteristic parameters. In this way, when the fine-grained interaction of features based on external knowledge modulation is performed, the introduction of external knowledge will optimize the semantic fine-grained interaction quality between the collateral status semantic embedding coding vector and the enterprise parameter semantic embedding significant fusion coding vector, but may introduce additional noise knowledge and cause the overall feature manifold representation of the collateral status-enterprise parameter fine-grained interaction feature vector in the high-dimensional feature space to have fine-grained feature structure holes. The existence of fine-grained feature structure holes will not only make the features of the collateral status-enterprise parameter fine-grained interaction feature vector corresponding to the semantic coverage of the class probability label insufficient, but also cause the outlier class regression inference mapping to be off-target, affecting the accuracy of the risk assessment result obtained by inputting the collateral status-enterprise parameter fine-grained interaction feature vector into the risk assessment module based on the classifier.

[0048] In the technical solution of the present application, the collateral status-enterprise parameter fine-grained interactive feature vector is input into a risk assessment module based on a classifier to obtain a risk assessment result, including:

[0049] Input the collateral status-enterprise parameter fine-grained interaction feature vector into a pre-classifier based on a Softmax function to obtain a collateral status-enterprise parameter semantic interaction fine-grained class probability label vector;

[0050] Calculate the product between the collateral status-enterprise parameter semantic interaction fine-grained class probability label vector and its transposed vector to obtain the collateral status-enterprise parameter semantic interaction fine-grained full label domain modulation matrix;

[0051] Dividing the eigenvalues ​​of each position in the collateral status-enterprise parameter semantic interaction fine-grained full label domain modulation matrix by the square root of the scale value of the collateral status-enterprise parameter fine-grained interaction eigenvector to obtain the collateral status-enterprise parameter semantic interaction fine-grained full label domain scale modulation matrix;

[0052] Input the collateral status-enterprise parameter semantic interaction fine-grained full-label domain scale modulation matrix into a class probability domain attention sparse module based on a multi-level gating function to obtain a sparse collateral status-enterprise parameter semantic interaction fine-grained full-label domain scale modulation matrix;

[0053] Taking the collateral status-enterprise parameter fine-grained interaction feature vector as a query feature vector, matrix-multiplying the sparse collateral status-enterprise parameter semantic interaction fine-grained full label domain scale modulation matrix with the collateral status-enterprise parameter fine-grained interaction feature vector to obtain an optimized collateral status-enterprise parameter fine-grained interaction feature vector;

[0054] The optimized collateral status-enterprise parameter fine-grained interactive feature vector is input into the classifier-based risk assessment module to obtain the risk assessment result.

[0055] The optimization process of the collateral status-enterprise parameter fine-grained interactive feature vector is expressed as follows:

[0056]

[0057] λ≥1

[0058]

[0059] Among them, v i represents the eigenvalue of the ith position of the collateral status-enterprise parameter fine-grained interactive feature vector, L represents the scale value of the collateral status-enterprise parameter fine-grained interactive feature vector, The eigenvalue of the ith position of the fine-grained class probability label vector representing the collateral status-enterprise parameter semantic interaction, V p Represents the collateral status-enterprise parameter semantic interaction fine-grained class probability label vector, represents matrix multiplication operation, T represents the transpose of the vector, M p represents the collateral status-enterprise parameter semantic interaction fine-grained full label domain scale modulation matrix, θ represents the predetermined threshold, represents the eigenvalue of the ith position of the fine-grained full-label domain scale modulation matrix of the collateral status-enterprise parameter semantic interaction, λ represents a predetermined hyperparameter, mask represents a mask function, Represents the eigenvalue of the i-th position of the fine-grained full-label domain scale modulation matrix of the sparse collateral state-enterprise parameter semantic interaction, M p ' represents the sparse collateral status-enterprise parameter semantic interaction fine-grained full label domain scale modulation matrix, and V′ represents the optimized collateral status-enterprise parameter fine-grained interaction feature vector.

[0060] Therefore, by adding class probability domain-level optimizable disturbances to the original feature vector to strengthen the dependency entanglement between different variables in the original feature vector based on label domain modulation, the source domain feature vector can be optimized based on class probability query scale modulation. In this way, the significant inherent modal information in the source domain features can be more effectively retained, and the interference components in the feature distribution can be effectively masked, thus improving the adversarial robustness of the manifold expression of the feature vector. In this way, the accuracy of the risk assessment results obtained by the risk assessment module based on the classifier input is improved.

[0061] In summary, the post-loan management method based on big data according to the embodiment of the present application is explained, which uses artificial intelligence technology based on deep learning to perform semantic analysis on the collateral status data and enterprise parameters of the target enterprise to capture the status characteristics of the collateral and the operating status characteristics of the enterprise, and introduces external knowledge to perform fine-grained semantic guidance interaction and fusion of the two to mine the potential risk-related information of the enterprise, thereby realizing intelligent assessment of the risk level of the target enterprise. In this way, the accuracy of the enterprise's post-loan risk assessment can be effectively improved, and the maintenance cost can be reduced, thereby improving the risk management efficiency of financial institutions.

[0062] Figure 5 Schematic flow chart of the post-loan management method based on big data in the embodiment of the present application. Figure 5 As shown, the post-loan management method based on big data includes: S1, obtaining the status data of the collateral of the target enterprise, and the enterprise parameters of the target enterprise, wherein the enterprise parameters include conventional parameters and characteristic parameters; S2, performing semantic embedding coding on the status data of the collateral, the conventional parameters and the characteristic parameters respectively to obtain the collateral status semantic embedding coding vector, the conventional parameter semantic embedding coding vector and the characteristic parameter semantic embedding coding vector; S3, performing significant fusion on the conventional parameter semantic embedding coding vector and the characteristic parameter semantic embedding coding vector to obtain the enterprise parameter semantic embedding significant fusion coding vector; S4, performing fine-grained feature interaction based on external knowledge guidance on the collateral status semantic embedding coding vector and the enterprise parameter semantic embedding significant fusion coding vector to obtain the collateral status-enterprise parameter fine-grained interaction feature vector; S5, determining the risk assessment result of the target enterprise based on the collateral status-enterprise parameter fine-grained interaction feature vector.

[0063] In one embodiment, the conventional parameter semantic embedding coding vector and the characteristic parameter semantic embedding coding vector are significantly fused to obtain the enterprise parameter semantic embedding significantly fused coding vector, including: inputting the conventional parameter semantic embedding coding vector and the characteristic parameter semantic embedding coding vector into a soft linear weighted fusion unit to obtain the enterprise parameter semantic embedding significantly fused coding vector.

[0064] Here, those skilled in the art can understand that the specific operations of each step in the above-mentioned post-loan management method based on big data have been referred to above. Figures 1 to 4 The description of the post-loan management system based on big data has been introduced in detail, and therefore, its repeated description will be omitted.

[0065] An embodiment of the present application further provides a computer program product, which includes a computer program code. When the computer program code runs on a computer, the computer implements the methods in the above embodiments of the present application.

[0066] An embodiment of the present application further provides a computer-readable storage medium, which stores computer instructions. When the computer instructions are executed on a computer, the computer implements the methods in the above embodiments of the present application.

[0067] An embodiment of the present application also provides a chip, including a circuit, for executing the methods in the above embodiments of the present application.

[0068] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0069] In the description of the embodiments of the present application, unless otherwise specified, " / " means or, for example, A / B can mean A or B; "and / or" in this article is a kind of association relationship that describes associated objects, indicating that three relationships can exist, for example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In this application, "at least one" means one or more, and "multiple" means two or more. "At least one of the following items" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b, or c can mean: a, b, c, ab, ac, bc, or abc, where a, b, c can be single or multiple.

[0070] The prefixes such as "first" and "second" used in the embodiments of the present application are only used to distinguish different description objects, and have no limiting effect on the position, order, priority, quantity or content of the described objects. The use of prefixes such as ordinal numbers used to distinguish description objects in the embodiments of the present application does not constitute a limitation on the described objects. For the statement of the described objects, please refer to the description in the context of the claims or embodiments, and the use of such prefixes should not constitute an unnecessary limitation.

[0071] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0072] In the various embodiments of the present application, unless otherwise specified or logically conflicting, the terms and / or descriptions between the various embodiments are consistent and can be referenced to each other, and the technical features in different embodiments can be combined to form new embodiments according to their inherent logical relationships.

[0073] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0074] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0075] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A post-loan management system based on big data, characterized in that: include: A target enterprise information acquisition module, used to acquire the status data of the collateral of the target enterprise and the enterprise parameters of the target enterprise, wherein the enterprise parameters include conventional parameters and characteristic parameters; A semantic embedding coding module, used to perform semantic embedding coding on the state data of the collateral, the conventional parameters and the characteristic parameters respectively to obtain a collateral state semantic embedding coding vector, a conventional parameter semantic embedding coding vector and a characteristic parameter semantic embedding coding vector; An enterprise parameter information fusion module, used for performing significant fusion on the conventional parameter semantic embedding coding vector and the characteristic parameter semantic embedding coding vector to obtain an enterprise parameter semantic embedding significant fusion coding vector; An information-guided interaction module, used for performing fine-grained feature interaction based on external knowledge guidance on the collateral status semantic embedding coding vector and the enterprise parameter semantic embedding significant fusion coding vector to obtain a collateral status-enterprise parameter fine-grained interaction feature vector; The target enterprise risk assessment module is used to determine the risk assessment result of the target enterprise based on the collateral status-enterprise parameter fine-grained interactive feature vector.

2. The post-loan management system based on big data according to claim 1 is characterized in that: The enterprise parameter information fusion module is used to: The conventional parameter semantic embedding coding vector and the characteristic parameter semantic embedding coding vector are input into a soft linear weighted fusion unit to obtain the enterprise parameter semantic embedding significant fusion coding vector.

3. The post-loan management system based on big data according to claim 2 is characterized in that: The information guidance interaction module includes: A feature fine-grained interactive optimization unit, used to perform feature fine-grained interactive optimization on the collateral status semantic embedding coding vector and the enterprise parameter semantic embedding significant fusion coding vector based on external knowledge to obtain an external knowledge optimized collateral status-enterprise parameter feature fine-grained interactive matrix; A feature-guided modulation fusion unit is used to optimize the collateral status-enterprise parameter feature fine-grained interaction matrix based on the external knowledge, and perform feature-guided modulation fusion on the collateral status semantic embedding coding vector and the enterprise parameter semantic embedding significant fusion coding vector to obtain the collateral status-enterprise parameter fine-grained interaction feature vector.

4. The post-loan management system based on big data according to claim 3 is characterized in that: The feature fine-grained interactive optimization unit is used to: Inputting the collateral status semantic embedding encoding vector and the enterprise parameter semantic embedding significant fusion encoding vector into a fine-grained feature interaction network to obtain a collateral status-enterprise parameter feature fine-grained interaction matrix; The collateral status-enterprise parameter feature fine-grained interaction matrix is ​​input into an external knowledge-based attention unit to obtain the external knowledge-optimized collateral status-enterprise parameter feature fine-grained interaction matrix.

5. The post-loan management system based on big data according to claim 4 is characterized in that: The feature-guided modulation fusion unit comprises: A feature modulation subunit is used to optimize the collateral status-enterprise parameter feature fine-grained interaction matrix based on the external knowledge, and perform feature modulation optimization on the collateral status semantic embedding coding vector and the enterprise parameter semantic embedding significant fusion coding vector to obtain an optimized collateral status semantic embedding coding vector and an optimized enterprise parameter semantic embedding significant fusion coding vector; The position-by-position semantic interaction fusion subunit is used to perform position-by-position semantic interaction fusion on the optimized collateral status semantic embedding coding vector and the optimized enterprise parameter semantic embedding significant fusion coding vector to obtain the collateral status-enterprise parameter fine-grained interaction feature vector.

6. The post-loan management system based on big data according to claim 5 is characterized in that: The characteristic modulation subunit is used for: Performing a linear transformation on the collateral status semantic embedding coding vector to obtain a first query feature vector and a first value feature vector, and using the external knowledge to optimize the collateral status-enterprise parameter feature fine-grained interaction matrix as a key matrix, inputting the first query feature vector, the first value feature vector and the key matrix into a fine-grained modulation module based on a Transformer structure to obtain the optimized collateral status semantic embedding coding vector; The enterprise parameter semantic embedding significant fusion coding vector is linearly transformed to obtain a second query feature vector and a second value feature vector, and the external knowledge optimized fine-grained feature interaction matrix is ​​used as a key matrix. The second query feature vector, the second value feature vector and the key matrix are input into the Transformer-based fine-grained modulation module to obtain the optimized enterprise parameter semantic embedding significant fusion coding vector.

7. The post-loan management system based on big data according to claim 6 is characterized in that: The position-by-position semantic interaction fusion subunit is used to: The collateral status-enterprise parameter fine-grained interaction feature vector is obtained by calculating the point-by-point division between the optimized collateral status semantic embedding coding vector and the optimized enterprise parameter semantic embedding significant fusion coding vector.

8. The post-loan management system based on big data according to claim 7 is characterized in that: The target enterprise risk assessment module is used to: The collateral status-enterprise parameter fine-grained interactive feature vector is input into a classifier-based risk assessment module to obtain the risk assessment result, which is a risk probability level label.

9. A post-loan management method based on big data, characterized in that: include: Acquire status data of collateral of the target enterprise and enterprise parameters of the target enterprise, wherein the enterprise parameters include conventional parameters and characteristic parameters; Performing semantic embedding coding on the state data of the collateral, the conventional parameters and the characteristic parameters respectively to obtain a collateral state semantic embedding coding vector, a conventional parameter semantic embedding coding vector and a characteristic parameter semantic embedding coding vector; Performing significant fusion on the conventional parameter semantic embedding coding vector and the characteristic parameter semantic embedding coding vector to obtain the enterprise parameter semantic embedding significant fusion coding vector; Performing fine-grained feature interaction based on external knowledge guidance on the collateral status semantic embedding coding vector and the enterprise parameter semantic embedding significant fusion coding vector to obtain a collateral status-enterprise parameter fine-grained interaction feature vector; Based on the collateral status-enterprise parameter fine-grained interactive feature vector, a risk assessment result of the target enterprise is determined.

10. The post-loan management method based on big data according to claim 9, characterized in that: The conventional parameter semantic embedding coding vector and the characteristic parameter semantic embedding coding vector are significantly fused to obtain the enterprise parameter semantic embedding significant fusion coding vector, including: The conventional parameter semantic embedding coding vector and the characteristic parameter semantic embedding coding vector are input into a soft linear weighted fusion unit to obtain the enterprise parameter semantic embedding significant fusion coding vector.

Images