Financial risk assessment method and device based on multi-source data fusion and medium
By employing a financial risk assessment method based on multi-source data fusion, and using class confidence values and Euclidean distance to quantify weights, combined with hesitation coefficients and decay constants for adjustment, a neural network architecture is constructed for feature extraction and enhancement. This solves the problem of low accuracy in existing financial risk assessment technologies and achieves more precise risk assessment.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BEIJING YINGDA CHANGAN RISK MANAGEMENT CONSULTING CO LTD
- Filing Date
- 2026-01-27
- Publication Date
- 2026-06-19
Smart Images
Figure CN122243661A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of financial risk prediction, and in particular to financial risk assessment methods, equipment, and media based on multi-source data fusion. Background Technology
[0002] In the fields of business management and financial risk control, financial risk assessment is a core element in ensuring sound business operations and reducing investment decision risks. Currently, corporate financial data is characterized by multiple sources and fragmentation, and existing assessment methods mostly adopt a simple approach of integrating multi-dimensional data to construct models.
[0003] The weight allocation in the aforementioned schemes relies heavily on subjective experience, resulting in low accuracy. Furthermore, the feature extraction process struggles to balance the common characteristics of multi-source data with the unique attributes of each data category, leading to insufficient effectiveness in feature fusion and consequently affecting assessment accuracy. As the market environment becomes increasingly complex, traditional methods can no longer meet the demands for accurate and objective financial risk assessment, necessitating a more accurate approach. Summary of the Invention
[0004] This invention solves the technical problem of low accuracy in financial risk assessment in the prior art by providing a financial risk assessment method, equipment and medium based on multi-source data fusion, and achieves the technical effect of improving the accuracy of financial risk assessment.
[0005] In a first aspect, the present invention provides a financial risk assessment method based on multi-source data fusion, comprising: S11: Obtain the company's historical financial relationship data and classify it, including several major categories, each of which contains several subcategories. For the financial relationship data of each subcategory, execute S110-S112, including: S110, determine the corresponding gain representation, and determine the corresponding time representation based on the gain representation; S111, determine the confidence value of the subclass based on several time representations; S112, based on Euclidean distance and according to the confidence value of subclasses, determine the subclass difference, the output weight of each subclass, and the output weight of each major class; S12, construct the neural network architecture and execute S120-S122, including: S120: First, extract general features from the financial data of each major category, and then deepen the features of each major category to obtain the feature vectors corresponding to each major category. S121, Based on the feature vectors of each major category and the corresponding output weights, the fused features are obtained; S122, based on fusion features, trains the neural network architecture, and when the preset training requirements are met, the neural network architecture is used for the financial risk assessment of the enterprise.
[0006] Further, determining the corresponding gain representation, and determining the corresponding time representation based on the gain representation, includes:
[0007] in, For the first Gain representation of each subclass For the first Relevance tags for each subcategory For the first Discrimination index for each subcategory;
[0008] in, For time representation, For data update interval, This is the time decay coefficient.
[0009] Furthermore, based on several time representations, the confidence values for subcategories are determined, including:
[0010] in, For the first The subclass confidence value of each subclass. The number of effective observations within the time window. For the first The sub-category in the The time representation value on a time slice. For variance operators, It is a positive and stable term.
[0011] Furthermore, based on Euclidean distance and according to the subclass confidence value, the subclass difference, the output weight of each subclass, and the output weight of each major class are determined, including:
[0012] in, For any subclass, the subclass confidence value. For the first The differences between subcategories. This represents the total number of subcategories.
[0013] in, For the first The output weights of each subclass The degree of hesitation is the coefficient. It is the attenuation constant;
[0014] in, For the first The output weights of each major category For the first Among the major categories, the first The output weights of each subclass For the first The number of subcategories within each major category.
[0015] Furthermore, the financial correlation data of each major category is first subjected to general feature extraction, and then feature deepening is performed on the general features of each major category to obtain the feature vectors corresponding to each major category, including: Transform and represent the financial data of each major category using line charts or heatmaps; Based on the hierarchical fusion architecture, the transformed financial correlation data of each major category is input into the block embedding module, and the common feature extraction of each major category is realized based on the shared layer. Based on the unique layer, the common features corresponding to each major category are deepened and fixed, and the feature vectors corresponding to each major category are output.
[0016] Furthermore, based on the feature vectors of each major category and their corresponding output weights, the fused features are obtained, including: Based on the feature vectors of each major category and their corresponding output weights, a weighted sum is performed to obtain preliminary fused features; Based on the independent MLP layer, the preliminary fusion features are refined to obtain the fusion features.
[0017] Furthermore, based on the fusion features, the neural network architecture is trained. When the preset training requirements are met, the neural network architecture is used for enterprise financial risk assessment, including: Based on the fusion characteristics, the predicted financial risks corresponding to the fusion characteristics are obtained; Based on the predicted financial risks and corresponding actual financial risks of several enterprises, a neural network architecture is trained. When the preset number of iterations or the error is less than a preset threshold is met, the neural network architecture is used for the financial risk assessment of the enterprise.
[0018] Furthermore, obtain and categorize the company's historical financial relationship data, including: First, break down historical financial data into major categories, including profitability, operational efficiency, earnings quality, and solvency. The subcategories of profitability are: earnings per share, return on equity, and return on assets. The subcategories included in operational capabilities are: fixed asset turnover ratio, current asset turnover ratio, and deposit cycle days. The subcategories included in earnings quality are: operating cash flow and cash ratio; The subcategories of debt repayment ability include: interest coverage ratio, equity ratio, and long-term debt-to-equity ratio.
[0019] In a second aspect, the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to execute a financial risk assessment method based on multi-source data fusion, as provided in the first aspect.
[0020] Thirdly, the present invention provides a non-transitory computer-readable storage medium, wherein when the instructions in the non-transitory computer-readable storage medium are executed by a processor of an electronic device, the electronic device is able to execute a financial risk assessment method based on multi-source data fusion as provided in the first aspect.
[0021] One or more technical solutions provided in this invention have at least the following technical effects or advantages: This invention achieves quantified and hierarchical weight allocation through subclass confidence values and Euclidean distance. This approach identifies reliable and unique subclass indicators in financial risk assessment while effectively suppressing the distortion of weights by extreme differences. Subclass differences measure the relative prominence of indicators, and the weights are adjusted using a hesitation coefficient and a decay constant to ensure the model can robustly allocate contributions even when facing volatile or uncertain data. Simultaneously, major class weights are normalized and aggregated to achieve hierarchical transfer from subclass to major class, clearly reflecting the importance of each financial dimension. This method balances the stability, uniqueness, and overall equilibrium of indicators, improving the scientific rigor and interpretability of weight allocation and providing a reasonable and quantifiable input basis for feature fusion in neural networks, making financial risk assessment more accurate and reliable.
[0022] Based on S120 and S121, this invention employs a mid-level fusion approach. First, it extracts common features from major categories using a hierarchical architecture, then deepens these features in a specific layer, ensuring that each category retains its unique information while reflecting cross-category commonalities. Subsequently, it combines the output weights of each category for weighted summation to generate fused features, amplifying the contributions of important categories and weakening the influence of weakly correlated indicators. This mid-level fusion method preserves the fine-grained information of the original data while achieving a reasonable integration of different financial dimensions, providing high-quality, interpretable input for subsequent MLP refinement and neural network training, thereby improving the accuracy, stability, and reliability of financial risk assessment.
[0023] This invention calculates the credibility and output weight of minor category indicators, then aggregates them to obtain major category weights, which in turn guide feature fusion. The bottom-up hierarchical weighting mechanism can accurately identify the fine-grained indicators that are most discriminative of financial risk and transmit their importance layer by layer to the major category dimension, thereby achieving differentiated focus in the feature fusion stage. This avoids the dilution of key signals caused by simple averaging or splicing, and enhances the model's sensitivity to highly credible and discriminative financial dimensions, ultimately improving the accuracy, interpretability, and robustness of risk prediction. Attached Figure Description
[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0025] Figure 1 This is a flowchart illustrating the financial risk assessment method based on multi-source data fusion provided by the present invention. Detailed Implementation
[0026] The embodiments of the present invention provide a financial risk assessment method based on multi-source data fusion, which solves the technical problem of low accuracy in financial risk assessment in the prior art.
[0027] The technical solution of this invention is to solve the above-mentioned technical problems, and the overall idea is as follows: The financial risk assessment method based on multi-source data fusion includes: S11, acquiring and classifying the company's historical financial correlation data, including several major categories, each of which contains several subcategories; for the financial correlation data of each subcategory, executing S110-S112, including: S110, determining the corresponding gain representation and the corresponding time representation based on the gain representation; S111, determining the subcategory confidence value based on the time representations; S112, determining the subcategory gap, the output weight of each subcategory, and the output weight of each major category based on Euclidean distance and the subcategory confidence value; S12, constructing a neural network architecture and executing S120-S122, including: S120, controlling the financial correlation data of each major category to first extract general features, and then deepening the features of each major category to obtain the feature vectors corresponding to each major category; S121, obtaining the fused features based on the feature vectors of each major category and the corresponding output weights; S122, training the neural network architecture based on the fused features, and when the preset training requirements are met, using the neural network architecture for the company's financial risk assessment.
[0028] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.
[0029] First, it should be clarified that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.
[0030] This invention provides, for example Figure 1 The financial risk assessment method based on multi-source data fusion shown includes: S11: Obtain the company's historical financial relationship data and classify it, including several major categories, each of which contains several subcategories. For the financial relationship data of each subcategory, execute S110-S112, including: Obtain and categorize the company's historical financial data, including: first, breaking down the historical financial data into major categories, such as profitability, operational efficiency, earnings quality, and solvency; the subcategories of profitability are: earnings per share, return on equity, and return on assets; the subcategories of operational efficiency are: fixed asset turnover, current asset turnover, and deposit cycle days; the subcategories of earnings quality are: operating cash flow and cash ratio; and the subcategories of solvency are: interest coverage ratio, equity ratio, and long-term debt ratio.
[0031] Profitability reflects a company’s efficiency in generating profits. Its subcategories include earnings per share (which measures the net profit per share of common stock), return on equity (which reflects the return on shareholders’ equity), and return on assets (which reflects a company’s ability to generate net profit using all its assets).
[0032] Operational capabilities focus on the efficiency of a company’s asset utilization and turnover, including fixed asset turnover (which measures a company’s ability to generate revenue using fixed assets) and current asset turnover (which reflects the efficiency of current asset utilization).
[0033] Earnings quality focuses on the quality and sustainability of a company's profits, and is composed of operating cash flow (which reflects the ability of the main business to generate cash) and the cash ratio.
[0034] Debt repayment capacity assesses a company's ability to repay its debt principal and interest, including the interest coverage ratio (which measures a company's ability to pay interest with operating profits), the debt-to-equity ratio (which reflects the ratio of debt to shareholders' equity in the capital structure), and the long-term debt-to-equity ratio (which indicates the proportion of long-term capital financed by debt).
[0035] In addition to the major and minor categories provided by this invention, other major or minor categories can be added, such as risk tolerance, innovation capability, etc. The division of minor categories can be based on the actual situation; the above are just examples.
[0036] In this invention, all calculations are performed using normalization to achieve dimensionless processing.
[0037] S110, determine the corresponding gain representation, and determine the corresponding time representation based on the gain representation.
[0038] Determine the corresponding gain representation, and determine the corresponding time representation based on the gain representation, including:
[0039] in, For the first Gain representation of each subclass For the first Relevance tags for each subcategory For the first Discrimination index for each subcategory;
[0040] in, For time representation, For data update interval, This is the time decay coefficient.
[0041] By constructing gain and time representations, the effectiveness and timeliness of various subcategories of financial indicators in financial risk assessment are uniformly quantified.
[0042] Gain is determined by both a correlation label and a discrimination index, and is used to measure the overall contribution of a sub-category to a company's financial risk. The correlation label reflects the strength of the direct association between the indicator and financial risk, while the discrimination index characterizes the indicator's ability to distinguish between risky and non-risky entities, thus avoiding judgments based solely on a single correlation or volatility. Both the discrimination index and the correlation label can be determined based on historical experience or experiments.
[0043] Based on this, a time representation is introduced, and the gain representation is time-weighted by the data update time interval and the time decay coefficient, so that the indicators that are recently updated and have high information freshness receive greater weight, while the contribution of indicators that are old and have reduced timeliness is correspondingly decayed.
[0044] Among them, the data update time interval is used to describe the time difference between the current calculation time and the most recent effective update, and the time decay coefficient is used to control the rate and magnitude of gain decay over time. Its value reflects the model's preference for the degree of retention of historical data.
[0045] S111, determine the confidence value of the subclass based on several time representations.
[0046] Based on several time representations, determine the confidence value of the subclass, including:
[0047] in, For the first The subclass confidence value of each subclass. The number of effective observations within the time window. For the first The sub-category in the The time representation value on a time slice. For variance operators, It is a positive and stable term.
[0048] By statistically summarizing the time representations of the same sub-category across multiple time slices, a sub-category credibility value is constructed to quantify the stability and continued effectiveness of the indicator in financial risk assessment.
[0049] Specifically, the time representation of each sub-category in each time slice is summed and averaged within the time window to reflect the overall contribution level of the indicator to financial risk over a period of time. Subsequently, time-represented variance is introduced as a volatility penalty term to suppress minor categories of indicators that change drastically and have poor stability across different time slices, thereby reducing their reliability values accordingly. Among them, the positive stability term is used to avoid numerical instability caused by the variance being zero or at a minimum. The subclass confidence value not only considers the average contribution of the indicator to the risk, but also comprehensively reflects its stability over time.
[0050] S112 determines the subclass difference, the output weight of each subclass, and the output weight of each major class based on Euclidean distance and the subclass confidence value.
[0051] Based on Euclidean distance and the confidence values of subclasses, the subclass differences, the output weights of each subclass, and the output weights of each major class are determined, including:
[0052] in, For any subclass, the subclass confidence value. For the first The differences between subcategories. This represents the total number of subcategories.
[0053] in, For the first The output weights of each subclass The degree of hesitation is the coefficient. It is the attenuation constant;
[0054] in, For the first The output weights of each major category For the first Among the major categories, the first The output weights of each subclass For the first The number of subcategories within each major category.
[0055] This invention establishes the relative importance and weight hierarchy of subclass indicators in risk assessment by calculating Euclidean distance based on subclass confidence values.
[0056] First, the average Euclidean distance between the confidence values of the first subclass and all other subclasses is calculated to quantify the difference or uniqueness of the subclass in the entire indicator set. That is, the larger the difference, the more prominent the confidence value of the subclass is among all subclasses, thus screening out core or unique risk signals.
[0057] Then, the output weights are converted into minor class weights, where the hesitation coefficient ranges from 0 to 1. The hesitation coefficient is used to adjust the model's confidence in the difference (which can be determined based on historical experience). When the model has high uncertainty about the minor class difference, the amplification of the weights can be reduced. The attenuation constant is used to control the overall scaling of the weights, so as to avoid weight distortion due to excessive differences.
[0058] Finally, the output weights of each subclass within the same major category are normalized and summarized to obtain the output weight of the major category, thus realizing the hierarchical weight allocation from the subclass to the major category.
[0059] This invention not only highlights the contribution of highly reliable and unique sub-categories to financial risk assessment, but also adjusts for extreme differences through hesitation and attenuation coefficients, ensuring the rationality and robustness of weight allocation. At the same time, the normalized category weights can clearly show the relative importance of each financial dimension in the overall risk assessment, providing a scientific input basis for the subsequent feature fusion and prediction of the neural network.
[0060] S12, construct the neural network architecture and execute S120-S122, including: S120 controls the financial correlation data of each major category to first extract general features, and then to deepen the features of each major category to obtain the feature vectors corresponding to each major category.
[0061] Specifically, this includes: converting and representing the financial correlation data of each major category into line charts or heatmaps; based on a hierarchical fusion architecture, inputting the converted financial correlation data of each major category into a block embedding module, and extracting the common features corresponding to each major category based on the shared layer; and based on the unique layer, deepening and fixing the common features corresponding to each major category, and outputting the feature vectors corresponding to each major category.
[0062] First, the financial correlation data of each category is converted into line charts or heatmaps, so that the original time series or numerical matrix data are presented in the form of two-dimensional images. This preserves the time trend, periodicity and relationship information between different indicators, providing a processable structured input for subsequent deep learning processing.
[0063] Subsequently, the converted image data is input into the block embedding module in the hierarchical fusion architecture. This module divides each major category of data into several local regions, extracts the common features of each major category of data through a shared layer, ensures that different categories of indicators are uniformly encoded in the same feature space, and extracts common information across indicators and time slices.
[0064] Specifically, it can be implemented based on View-wise Squeeze-and-Excitation Vision Transformer. Line graphs or heatmaps of major categories are passed in parallel through the same shared block embedding module and segmented to obtain several sequences. Then, a learnable classification token is pre-applied to each sequence and added to the shared position code to preserve spatial information. All sequences are subjected to a common shared layer to achieve general feature capture.
[0065] General features capture the overall behavioral patterns and potential correlations of major categories of indicators, providing a basic representation for risk assessment.
[0066] Based on the unique layer, the common features of each major category are further deepened and fixed to generate category-specific feature vectors. This preserves the unique information and key differences of each major category, so that each feature vector reflects both the general pattern and the category features. Ultimately, this provides high-quality, interpretable input for the fusion of features from multiple categories and the training of neural networks, enabling refined modeling and prediction of corporate financial risks.
[0067] Each Transformer Block consists of a multi-head self-attention module and a feedforward network, and is trained stably using layer normalization and residual connections.
[0068] Finally, the token states in the output sequences of each unique layer are normalized by their respective LayerNorm layers and used as feature vectors for each major category.
[0069] S121, based on the feature vectors of each major category and the corresponding output weights, obtain the fused features.
[0070] Specifically, this includes: performing weighted summation based on the feature vectors of each major category and their corresponding output weights to obtain preliminary fused features; and refining the preliminary fused features based on independent MLP layers to obtain fused features.
[0071] Specifically, the feature vectors of each major category are weighted and summed according to their corresponding output weights to obtain preliminary fusion features. This can reasonably reflect the importance of different financial dimensions, making the core major category contribute more to the final fusion features, while the influence of weakly correlated or unstable major categories is relatively weakened. This ensures that the fusion features take into account the differences in indicators and the weight distribution when reflecting the overall financial status of the enterprise.
[0072] Subsequently, the preliminary fused features are input into independent multilayer perceptron (MLP) layers for refinement. Through nonlinear mapping and inter-layer weight learning, the potential complex interaction relationships between features are further extracted, redundant information is removed, and the discriminative and expressive capabilities of the fused features are enhanced.
[0073] The resulting fusion features retain important information from all major feature categories and optimize the overall representation through weighted and nonlinear mapping, enabling them to serve as high-quality inputs to neural networks for more accurate and interpretable corporate financial risk assessment.
[0074] Based on S120 and S121, this invention employs a mid-level fusion approach. First, it extracts common features from major categories using a hierarchical architecture, then deepens these features in a specific layer, ensuring that each category retains its unique information while reflecting cross-category commonalities. Subsequently, it combines the output weights of each category for weighted summation to generate fused features, amplifying the contributions of important categories and weakening the influence of weakly correlated indicators. This mid-level fusion method preserves the fine-grained information of the original data while achieving a reasonable integration of different financial dimensions, providing high-quality, interpretable input for subsequent MLP refinement and neural network training, thereby improving the accuracy, stability, and reliability of financial risk assessment.
[0075] S122, based on fusion features, trains the neural network architecture, and when the preset training requirements are met, the neural network architecture is used for the financial risk assessment of the enterprise.
[0076] The process includes: obtaining the predicted financial risk corresponding to the fusion features based on the fusion features; training a neural network architecture based on the predicted financial risks and corresponding actual financial risks of several enterprises; and using the neural network architecture for enterprise financial risk assessment when the preset number of iterations or the error is less than a preset threshold is met.
[0077] The system can input fused features into a preset neural network architecture to obtain the predicted financial risk of the corresponding enterprise. The preset neural network architecture can be trained sequentially for several enterprises using steps S11-12. When the preset number of iterations or the error is less than a preset threshold is met, the neural network architecture can be used for the financial risk assessment of the enterprise.
[0078] In summary, this invention achieves quantified and hierarchical weight allocation through subclass confidence values and Euclidean distance. This approach not only identifies reliable and unique subclass indicators in financial risk assessment but also effectively suppresses the distortion of weights by extreme differences. Subclass differences measure the relative prominence of indicators, and the weights are adjusted using a hesitation coefficient and a decay constant to ensure the model can robustly allocate contributions even when facing volatile or uncertain data. Simultaneously, major class weights are normalized and aggregated to achieve hierarchical transfer from subclass to major class, clearly reflecting the importance of each financial dimension. This method balances the stability, uniqueness, and overall equilibrium of indicators, improving the scientific rigor and interpretability of weight allocation and providing a reasonable and quantifiable input basis for feature fusion in neural networks, thus making financial risk assessment more accurate and reliable.
[0079] Based on S120 and S121, this invention employs a mid-level fusion approach. First, it extracts common features from major categories using a hierarchical architecture, then deepens these features in a specific layer, ensuring that each category retains its unique information while reflecting cross-category commonalities. Subsequently, it combines the output weights of each category for weighted summation to generate fused features, amplifying the contributions of important categories and weakening the influence of weakly correlated indicators. This mid-level fusion method preserves the fine-grained information of the original data while achieving a reasonable integration of different financial dimensions, providing high-quality, interpretable input for subsequent MLP refinement and neural network training, thereby improving the accuracy, stability, and reliability of financial risk assessment.
[0080] This invention calculates the credibility and output weight of minor category indicators, then aggregates them to obtain major category weights, which in turn guide feature fusion. The bottom-up hierarchical weighting mechanism can accurately identify the fine-grained indicators that are most discriminative of financial risk and transmit their importance layer by layer to the major category dimension, thereby achieving differentiated focus in the feature fusion stage. This avoids the dilution of key signals caused by simple averaging or splicing, and enhances the model's sensitivity to highly credible and discriminative financial dimensions, ultimately improving the accuracy, interpretability, and robustness of risk prediction.
[0081] Based on the same inventive concept, an electronic device is also provided, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to execute a financial risk assessment method based on multi-source data fusion as described above.
[0082] Based on the same inventive concept, this invention also provides a non-transitory computer-readable storage medium that, when the instructions in the storage medium are executed by the processor of an electronic device, enables the electronic device to perform the financial risk assessment method based on multi-source data fusion as described above.
[0083] Since the electronic device described in this embodiment is an electronic device used to implement the information processing method in the embodiments of the present invention, those skilled in the art can understand the specific implementation methods and various variations of the electronic device in this embodiment based on the information processing method described in the embodiments of the present invention. Therefore, how the electronic device implements the method in the embodiments of the present invention will not be described in detail here. Any electronic device used by those skilled in the art to implement the information processing method in the embodiments of the present invention falls within the scope of protection of the present invention.
[0084] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0085] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0086] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0087] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0088] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.
[0089] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.
Claims
1. A financial risk assessment method based on multi-source data fusion, characterized in that, include: S11: Obtain the company's historical financial relationship data and classify it, including several major categories, each of which contains several subcategories. For the financial relationship data of each subcategory, execute S110-S112, including: S110, determine the corresponding gain representation, and determine the corresponding time representation based on the gain representation; S111, determine the confidence value of the subclass based on several time representations; S112, based on Euclidean distance and according to the confidence value of subclasses, determine the subclass difference, the output weight of each subclass, and the output weight of each major class; S12, construct the neural network architecture and execute S120-S122, including: S120: First, extract general features from the financial data of each major category, and then deepen the features of each major category to obtain the feature vectors corresponding to each major category. S121, Based on the feature vectors of each major category and the corresponding output weights, the fused features are obtained; S122, Based on the fusion features, the neural network architecture is trained, and when the preset training requirements are met, the neural network architecture is used for the financial risk assessment of the enterprise.
2. The financial risk assessment method based on multi-source data fusion as described in claim 1, characterized in that, Determine the corresponding gain representation, and determine the corresponding time representation based on the gain representation, including: in, For the first Gain representation of each subclass For the first Relevance tags for each subcategory For the first Discrimination index for each subcategory; in, For time representation, For data update interval, This is the time decay coefficient.
3. The financial risk assessment method based on multi-source data fusion as described in claim 2, characterized in that, Based on several time representations, determine the confidence value of the subclass, including: in, For the first The subclass confidence value of each subclass. The number of effective observations within the time window. For the first The sub-category in the The time representation value on a time slice. For variance operators, It is a positive and stable term.
4. The financial risk assessment method based on multi-source data fusion as described in claim 1, characterized in that, Based on Euclidean distance and the confidence values of subclasses, the subclass differences, the output weights of each subclass, and the output weights of each major class are determined, including: in, For any subclass, the subclass confidence value. For the first The differences between subcategories. This represents the total number of subcategories. in, For the first The output weights of each subclass The degree of hesitation is the coefficient. It is the attenuation constant; in, For the first The output weights of each major category For the first Among the major categories, the first The output weights of each subclass For the first The number of subcategories within each major category.
5. The financial risk assessment method based on multi-source data fusion as described in claim 1, characterized in that, First, general features are extracted from the financial correlation data of each major category. Then, feature refinement is performed on the general features of each major category to obtain the feature vectors corresponding to each major category, including: Transform and represent the financial data of each major category using line charts or heatmaps; Based on the hierarchical fusion architecture, the transformed financial correlation data of each major category is input into the block embedding module, and the common feature extraction of each major category is realized based on the shared layer. Based on the unique layer, the common features corresponding to each major category are deepened and fixed, and the feature vectors corresponding to each major category are output.
6. The financial risk assessment method based on multi-source data fusion as described in claim 5, characterized in that, Based on the feature vectors of each major category and their corresponding output weights, the fused features are obtained, including: Based on the feature vectors of each major category and their corresponding output weights, a weighted sum is performed to obtain preliminary fused features; Based on the independent MLP layer, the preliminary fusion features are refined to obtain the fusion features.
7. The financial risk assessment method based on multi-source data fusion as described in claim 1, characterized in that, Based on the fusion features, the neural network architecture is trained. When preset training requirements are met, the neural network architecture is used for enterprise financial risk assessment, including: Based on the fusion characteristics, the predicted financial risks corresponding to the fusion characteristics are obtained; Based on the predicted financial risks and corresponding actual financial risks of several enterprises, the neural network architecture is trained. When the preset number of iterations or the error is less than a preset threshold is met, the neural network architecture is used for the financial risk assessment of the enterprise.
8. The financial risk assessment method based on multi-source data fusion as described in claim 1, characterized in that, Obtain and categorize the company's historical financial data, including: First, break down historical financial data into major categories, including profitability, operational efficiency, earnings quality, and solvency. The subcategories of profitability are: earnings per share, return on equity, and return on assets. The subcategories included in operational capabilities are: fixed asset turnover ratio, current asset turnover ratio, and deposit cycle days. The subcategories included in earnings quality are: operating cash flow and cash ratio; The subcategories of debt repayment ability include: interest coverage ratio, equity ratio, and long-term debt-to-equity ratio.
9. An electronic device, characterized in that, include: processor; Memory used to store the processor's executable instructions; The processor is configured to execute the financial risk assessment method based on multi-source data fusion as described in any one of claims 1 to 8.
10. A non-transitory computer-readable storage medium, characterized in that, When the instructions in the non-transitory computer-readable storage medium are executed by the processor of the electronic device, the electronic device is able to perform the financial risk assessment method based on multi-source data fusion as described in any one of claims 1 to 8.