Enterprise risk monitoring system of financial service platform based on big data analysis

Abstract

The invention discloses an enterprise risk monitoring system of a financial service platform based on big data analysis, which relates to the technical field of enterprise risk monitoring and specifically comprises a financial data feature extraction module, a financial data feature prediction module and an enterprise financial risk assessment module. Wherein the financial data feature extraction module is used for processing structured and unstructured data related to enterprise finance and extracting an enterprise periodic fluctuation index and an enterprise risk fluctuation index, the financial data feature prediction module is used for establishing a dynamic prediction model, the encoder extracts time sequence features through LSTM, and the encoder outputs the time sequence features through LSTM. The decoder dynamically weights key time steps in combination with an attention mechanism and captures a long-period dependency relationship by calculating a final information vector, and the enterprise financial risk assessment module constructs an enterprise risk score by analyzing a current enterprise periodic fluctuation index and an enterprise risk fluctuation index. And determining the prospective risk of the enterprise in combination with the predicted value of the future time point, thereby realizing the full-period monitoring of the enterprise risk.

Description

Technical Field

[0001] The present invention belongs to the technical field of enterprise risk monitoring, and in particular relates to an enterprise risk monitoring system for a financial service platform based on big data analysis. Background Art

[0002] With the rapid development of the financial industry, financial institutions are facing an increasingly complex risk environment. Traditional risk management models can no longer meet the needs of modern financial business. Therefore, financial technology (FinTech) has emerged and has become an important force in promoting changes in the financial industry. FinTech provides more efficient and intelligent solutions for the financial industry by integrating advanced technologies such as big data, artificial intelligence, and blockchain.

[0003] However, current enterprise risk monitoring systems usually rely on a single data source and static models. Data between financial institutions cannot be interoperable due to privacy restrictions, resulting in one-sided risk assessment and difficulty in integrating multi-dimensional information such as supply chain and public opinion. Traditional models rely on financial indicators with fixed time windows (such as current ratios) and cannot dynamically capture cyclical fluctuations and sudden risk events (such as administrative penalties and public opinion crises). Text data (such as news and penalty announcements) require manual annotation, low processing efficiency and difficulty in quantifying risk impacts. Existing technologies such as logistic regression scorecards and isolated forest anomaly detection can partially solve the above problems, but lack deep integration of multimodal data (time series + text + relationship network), and model updates rely on centralized data training, which violates financial data privacy compliance requirements.

[0004] Therefore, there is an urgent need for an enterprise risk monitoring system based on a financial service platform based on big data analysis to solve the technical problems of insufficient utilization of unstructured data, serious data silos and delayed prediction. Summary of the invention

[0005] The purpose of the present invention is to provide an enterprise risk monitoring system for a financial service platform based on big data analysis, which is used to solve the technical problems of insufficient utilization of unstructured data, serious data island problems and delayed prediction in the prior art.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions: The enterprise risk monitoring system of the financial service platform based on big data analysis includes: Financial data feature extraction module, used to process structured and unstructured data related to corporate finance, and extract corporate cyclical fluctuation indicators and corporate risk fluctuation indicators; The financial data feature prediction module is used to establish the LSTM-Attention dynamic prediction model. The encoder extracts time series features through LSTM, and the decoder combines the attention mechanism to dynamically weight key time steps and calculate the final information vector to capture long-term dependencies. The enterprise financial risk assessment module constructs an enterprise risk score by analyzing the current enterprise cyclical fluctuation indicators and enterprise risk fluctuation indicators, and combines the enterprise risk score with the predicted value at a future time point to determine the enterprise's forward-looking risk.

[0007] Furthermore, the specific methods for processing structured and unstructured data related to corporate finance are as follows: Access the financial services platform, use the federated learning framework to perform distributed modeling of sensitive data, integrate corporate structured financial statements and supply chain transaction records, and generate time series data of corporate transaction flows by arranging structured data in time series. By determining the public opinion news data of the company's unstructured text and the administrative penalty data of semi-structured text, text information related to the company can be identified from the public opinion news data. At the same time, the administrative penalty data of semi-structured text can be parsed, and key information can be extracted from data in non-standard formats through intelligent text analysis technology to obtain risk text data about the company.

[0008] Furthermore, the cyclical fluctuation index of enterprises is extracted. The specific method is as follows: Using the formula Represents the cyclical fluctuation index of the enterprise, where It represents the cyclical fluctuation index of the enterprise at time point t, t represents time point t, k represents the kth time point, and K represents a total of K time points. represents the transaction amount at the t-kth time point, Represents the expected benchmark value of the enterprise transaction flow in the current time window, using the formula express, is the time attenuation coefficient, is the trend influence coefficient, and is a constant value, Represents the trend slope at time point t.

[0009] Furthermore, the enterprise risk volatility index is extracted. The specific method is as follows: For enterprise risk text data, we combined the existing BERT model to extract sentiment polarity and event types. Sentiment polarity is determined by using a preset sentiment vocabulary, using the BERT model to identify words in the text data that belong to the sentiment vocabulary, and according to the preset scores of sentiment vocabulary, determining the sentiment polarity of the text data. After the BERT model classifies and identifies the types of public opinion events, we assign weights, extract the number of penalty records from government public text data, and classify them according to the penalty types. Using the formula The enterprise risk volatility index is constructed by integrating the sentiment polarity of text data, the weight of public opinion event types and the level of penalty event types; Among them, t represents the time point t, the time window is divided into H time lengths, the time point where the current enterprise is located is taken as the end point of the time window, and the time window where the current enterprise is located is determined. represents the enterprise risk volatility index at time point t, represents the type weight of the i-th public opinion event in the time window at time point t, represents the sentiment polarity score of the ith public opinion event in the time window at time point t, represents the difference between the current time and the time when public opinion event i occurs. f represents the time decay coefficient of public opinion event, which is used to control the speed at which the influence of public opinion event decays over time. It represents the severity score of the jth penalty event in the time window at time point t. y represents the penalty event attenuation coefficient, which is used to control the speed at which the influence of the penalty event decays over time. g1 represents the public opinion weight coefficient, and g2 represents the penalty weight coefficient.

[0010] Furthermore, an LSTM-Attention dynamic prediction model is established. The specific method is as follows: Collect historical enterprise risk volatility indicators and enterprise cyclical volatility indicators as prediction data, divide the historical prediction data into input data and control data in chronological order, use the input data as the input training model of the LSTM-Attention model, and record the input data as , each input Different weight matrices obtained through the deep learning process are mapped to different vectors. The vectors obtained through the mapping calculate the attention scores belonging to different weights, which are used to determine which texts of the input data the model pays attention to. The stacked RNN layers are used as encoders and decoders. The encoder combines all input time steps to process the input data to obtain a compact representation of the input data. , which is the compressed format of the input. The decoder receives the context vector and generates output data. In the LSTM-based seq2seq, an Attention Layer is added between the Encoder and the Decoder to help the compact representation A encode the information from all input time steps.

[0011] Furthermore, the final information vector is calculated to capture the long-term dependencies. The specific method is as follows: The vectors of the upper and lower time points are recorded as the state sequence vector s, and the formula is used represents the final information vector, where T represents the length of the selected time step, t represents the time t, represents the state sequence vector at time t-1, is the hidden state of the model, and a(*) indicates that the parameters are normalized by the softmax function.

[0012] Furthermore, the enterprise risk score is constructed by analyzing the current enterprise cyclical fluctuation index and enterprise risk fluctuation index. The specific method is as follows: Using the formula represents the enterprise risk score, where t represents time point t, RS(t) represents the enterprise risk score at time point t, P(t) represents the enterprise cyclical fluctuation index at time point t, and R(t) represents the enterprise risk fluctuation index at time point t. It represents the mean value of the enterprise cyclical fluctuation index in the time window with t as the end point, It represents the standard deviation of the enterprise cyclical fluctuation index in the time window with t as the end point, It represents the mean value of the enterprise risk volatility index in the time window with t as the end point, It represents the standard deviation of the enterprise risk volatility index in the time window with t as the end point, represents the weight coefficient of enterprise cyclical fluctuation, Represents the enterprise risk volatility weight coefficient.

[0013] Furthermore, the enterprise risk score is combined with the predicted value at a future time point to determine the enterprise's forward-looking risk. The specific method is as follows: Based on the financial data characteristics of the enterprise at the current time point, the LSTM-Attention model is used to obtain the predicted value of the financial data characteristics of the enterprise at the future time point. The enterprise risk score at the current time point is combined to determine the enterprise's forward-looking risk probability index. If the enterprise's forward-looking risk probability index at the current time point is greater than the enterprise's forward-looking risk probability index threshold V, it is judged that the current enterprise is in a high-risk time window and the enterprise risk disposal process needs to be initiated. If the enterprise's forward-looking risk probability index at the current time point is less than or equal to the threshold V, it is judged that the current enterprise is not in a high-risk time window and the enterprise risk disposal process does not need to be initiated.

[0014] Furthermore, the enterprise forward-looking risk probability index is determined by comprehensively considering the enterprise risk score at the current time point. The specific method is as follows: Using the formula represents the enterprise forward-looking risk probability index, where t represents time point t, Fx(t) represents the enterprise forward-looking risk probability index at time point t, G represents the number of preset simulation time points, i represents the i-th time point, represents the predicted value of the enterprise cyclical fluctuation index at the t+ith time point, represents the predicted value of the enterprise risk volatility index at the t+ith time point, P(t) represents the enterprise cyclical volatility index at the time point t, is the time decay coefficient.

[0015] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: 1. The present invention constructs enterprise cyclical fluctuation indicators and enterprise risk fluctuation indicators. The module can quantify the short-term risk changes of enterprises. The enterprise cyclical fluctuation indicators take into account the cyclical changes of enterprise transaction flows, while the enterprise risk fluctuation indicators combine information from multiple dimensions such as public opinion events, emotional polarity, and penalty events. These indicators provide financial institutions with more accurate risk assessment tools, which help to identify potential risks in a timely manner and take corresponding measures; 2. The present invention realizes dynamic prediction of enterprise financial data by introducing the LSTM-Attention model. Compared with the traditional LSTM model, the LSTM-Attention model can selectively focus on key information in the input data by adding an attention mechanism, thereby improving the prediction accuracy. At the same time, the model can process various types of input data such as text representation, time series data and image pixels, further enriching the information source of the prediction and improving the comprehensiveness of the prediction.

[0016] 3. The present invention calculates the enterprise risk score by integrating the financial data characteristics of the enterprise at the current time, including the enterprise cyclical fluctuation index and the enterprise risk fluctuation index, using a specific formula. This score can accurately quantify the risk level of the enterprise at the current time point, provide an intuitive and quantifiable risk reference for financial institutions or enterprise management, improve the prediction of enterprise financial data characteristics at future time points, and then calculate the enterprise forward-looking risk probability index. This indicator can reveal the risks that the enterprise may face in advance, provide forward-looking risk management suggestions for financial institutions or enterprise management, and help prevent and resolve potential risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 A step diagram of an enterprise risk monitoring method for a financial service platform based on big data analysis is shown; Figure 2 A module diagram of an enterprise risk monitoring system based on a financial service platform for big data analysis is shown; Figure 3 A method step diagram for constructing a forward-looking risk probability indicator for an enterprise is shown. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] like Figure 2 The enterprise risk monitoring system of the financial service platform based on big data analysis shown in the figure specifically includes the following steps: The financial data feature extraction module, by accessing the financial service platform, uses the federated learning framework to perform distributed modeling on sensitive data, avoids cross-domain transmission of original data, and integrates corporate financial structured statements and supply chain transaction records. Corporate financial structured statements include key financial information such as balance sheets, income statements, and cash flow statements, which are an important basis for evaluating the financial health of enterprises. In addition, supply chain transaction records are also included. Supply chain transaction records record in detail the transactions between enterprises and upstream and downstream partners, providing valuable data support for analyzing the operational efficiency and market position of enterprises. By arranging these structured data in time series, it is possible to generate transaction flow data of enterprises arranged in chronological order. By identifying the public opinion news data of the enterprise's unstructured text and the administrative penalty data of the semi-structured text, text information related to the enterprise can be identified from the public opinion news data. This information contains the market's evaluation of the enterprise's operating conditions, product quality, and management changes. At the same time, the administrative penalty data of the semi-structured text is parsed. These data reflect the performance of the enterprise in complying with laws and regulations. Through intelligent text analysis technology, key information is extracted from these non-standard format data to obtain risk text data about the enterprise. The duration between adjacent time points is determined as A. Based on the historical transaction flow data of the enterprise, the cyclical fluctuation index of the current enterprise is determined. The specific formula of the cyclical fluctuation index of the enterprise is as follows: ; in, It represents the cyclical fluctuation index of the enterprise at time point t, t represents time point t, k represents the kth time point, and K represents a total of K time points. represents the transaction amount at the t-kth time point, Represents the expected benchmark value of the enterprise transaction flow in the current time window, using the formula express, is the time attenuation coefficient, is the trend influence coefficient, and is a constant value, It represents the trend slope at time point t. Linear regression is used to fit the enterprise transaction flow amounts at the current time point t and K historical time points. The slopes of the descending line segments of the transaction flow amounts at adjacent time points are determined in chronological order and recorded as slope values. The average of the slope values ​​at the current time point t and K historical time points is calculated to obtain the trend slope at the current time point t.

[0021] For enterprise risk text data, the existing BERT model is combined to extract sentiment polarity and event types. Sentiment polarity is determined by using a preset sentiment word library and using the BERT model to identify words in the text data that belong to the sentiment word library. The sentiment polarity of the text data is determined based on the preset score of the sentiment word library. The type of public opinion event is classified and identified by the BERT model and then weighted. The number of penalty records is extracted from the government's public text data and graded according to the penalty type. For example, warning events are graded as level 1, and the fine time is graded by amount. If it is less than 5% of the enterprise's monthly turnover amount, it is graded as level 1, and it is upgraded by 1 level for every 5% increase. Litigation events are graded as level 3. The sentiment polarity of the text data, the weight of the public opinion event type, and the level of the penalty event type are combined to construct an enterprise risk volatility index to quantify the short-term risk changes of the enterprise. The specific formula of the enterprise risk volatility index is as follows: ; Among them, t represents the time point t, the time window is divided into H time lengths, the time point where the current enterprise is located is taken as the end point of the time window, and the time window where the current enterprise is located is determined. represents the enterprise risk volatility index at time point t, represents the type weight of the i-th public opinion event in the time window at time point t, represents the sentiment polarity score of the ith public opinion event in the time window at time point t, represents the difference between the current time and the time when public opinion event i occurs. f represents the time decay coefficient of public opinion event, which is used to control the speed at which the influence of public opinion event decays over time. It represents the severity score of the jth penalty event in the time window at time point t. y represents the penalty event attenuation coefficient, which is used to control the speed at which the influence of the penalty event decays over time. g1 represents the public opinion weight coefficient, and g2 represents the penalty weight coefficient.

[0022] The enterprise risk volatility index and the enterprise cyclical volatility index are marked as the financial data characteristics of the enterprise.

[0023] The financial data feature prediction module introduces the LSTM-Attention model to build a dynamic prediction model for corporate financial data. It collects historical corporate risk volatility indicators and corporate cyclical volatility indicators as prediction data, divides the historical prediction data into input data and control data in chronological order, and uses the input data as the input training model of the LSTM-Attention model. Different from the traditional LSTM model, the LSTM-Attention model achieves selective attention to the input data by adding an attention mechanism, and records the input data as , the input data includes text representation, time series data and image pixels. Each input Different weight matrices obtained through the deep learning process are mapped to different vectors. The vectors obtained through the mapping calculate the attention scores belonging to different weights, which are used to determine which texts of the input data the model pays attention to. The stacked RNN layers are used as encoders and decoders. The encoder combines all input time steps to process the input data to obtain a compact representation of the input data. , which is the compressed format of the input. The decoder receives the context vector and generates output data. In the LSTM-based seq2seq, an Attention Layer is added between the Encoder and the Decoder to help the compact representation A encode the information from all input time steps. By calculating the final information vector, the vectors of the upper and lower time points are recorded as the state sequence vector s. The calculation formula of the final information vector is as follows: ; Where T represents the length of the selected time step, t represents the time t, represents the state sequence vector at time t-1, is the hidden state of the model, and a(*) indicates that the parameters are normalized by the softmax function.

[0024] The LSTM-Attention model is used to predict the input data of a time period to obtain the prediction component of a time period. The predicted value of the model is compared with the true value in the control set of the corresponding time period. The prediction ability of the prediction model is measured by calculating the absolute value of the error between the predicted value and the true value at the corresponding time. The model accuracy threshold R is set, and the prediction value of I time points is predicted by the prediction model. The target square error between I predicted values ​​and the true value at the corresponding time points is calculated. When the target square errors calculated I times are all less than or equal to the threshold R, it proves that the LSTM-Attention model training is completed. On the contrary, when the target square error of at least one time point in the target square error calculated I times is greater than the threshold R, it proves that the LSTM-Attention model training is not completed and needs to be retrained by adjusting the model parameters.

[0025] The enterprise financial risk assessment module determines the enterprise risk score at the current time based on the financial data characteristics of the enterprise at the current time. The specific formula for the enterprise risk score is as follows: ; Among them, t represents time point t, RS(t) represents the enterprise risk score at time point t, P(t) represents the enterprise cyclical fluctuation index at time point t, and R(t) represents the enterprise risk fluctuation index at time point t. It represents the mean value of the enterprise cyclical fluctuation index in the time window with t as the end point, It represents the standard deviation of the enterprise cyclical fluctuation index in the time window with t as the end point, It represents the mean value of the enterprise risk volatility index in the time window with t as the end point, It represents the standard deviation of the enterprise risk volatility index in the time window with t as the end point, represents the weight coefficient of enterprise cyclical fluctuation, Represents the enterprise risk volatility weight coefficient.

[0026] Based on the financial data characteristics of the enterprise at the current time point, the LSTM-Attention model is used to obtain the predicted value of the financial data characteristics of the enterprise at the future time point, and the enterprise forward-looking risk probability index is obtained comprehensively. The specific calculation formula of the enterprise forward-looking risk probability index is as follows: ; Where t represents time point t, Fx(t) represents the enterprise forward-looking risk probability index at time point t, G represents the number of preset simulation time points, i represents the i-th time point, represents the predicted value of the enterprise cyclical fluctuation index at the t+ith time point, represents the predicted value of the enterprise risk volatility index at the t+ith time point, P(t) represents the enterprise cyclical volatility index at the time point t, is the time decay coefficient.

[0027] A threshold value V of the enterprise's forward-looking risk probability index is preset. If the enterprise's forward-looking risk probability index at the current time point is greater than the threshold value V, it is judged that the current enterprise is in a high-risk time window and the enterprise risk disposal process needs to be initiated. If the enterprise's forward-looking risk probability index at the current time point is less than or equal to the threshold value V, it is judged that the current enterprise is not in a high-risk time window and the enterprise risk disposal process does not need to be initiated.

[0028] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

[0029] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to only specific implementation methods. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. An enterprise risk monitoring system for a financial service platform based on big data analysis, characterized in that: include: Financial data feature extraction module, used to process structured and unstructured data related to corporate finance, and extract corporate cyclical fluctuation indicators and corporate risk fluctuation indicators; The financial data feature prediction module is used to establish the LSTM-Attention dynamic prediction model. The encoder extracts time series features through LSTM, and the decoder combines the attention mechanism to dynamically weight key time steps and calculate the final information vector to capture long-term dependencies. The enterprise financial risk assessment module constructs an enterprise risk score by analyzing the current enterprise cyclical fluctuation indicators and enterprise risk fluctuation indicators, and combines the enterprise risk score with the predicted value at a future time point to determine the enterprise's forward-looking risk.

2. The enterprise risk monitoring system of the financial service platform based on big data analysis according to claim 1 is characterized in that: Processing structured and unstructured data related to corporate finance, the specific methods are: Access the financial services platform, use the federated learning framework to perform distributed modeling of sensitive data, integrate corporate structured financial statements and supply chain transaction records, and generate time series data of corporate transaction flows by arranging structured data in time series. By determining the public opinion news data of the company's unstructured text and the administrative penalty data of semi-structured text, text information related to the company can be identified from the public opinion news data. At the same time, the administrative penalty data of semi-structured text can be parsed, and key information can be extracted from data in non-standard formats through intelligent text analysis technology to obtain risk text data about the company.

3. The enterprise risk monitoring system based on the financial service platform of big data analysis according to claim 1 is characterized in that: Extract the enterprise cyclical fluctuation index, the specific method is: Using the formula Represents the enterprise cyclical fluctuation index, where: It represents the cyclical fluctuation index of the enterprise at time point t, t represents time point t, k represents the kth time point, and K represents a total of K time points. represents the transaction amount at the t-kth time point, Represents the expected benchmark value of the enterprise transaction flow in the current time window, using the formula express, is the time attenuation coefficient, is the trend influence coefficient, and is a constant value, Represents the trend slope at time point t.

4. The enterprise risk monitoring system based on the financial service platform of big data analysis according to claim 1 is characterized in that: Extract enterprise risk volatility index, the specific method is: For enterprise risk text data, we combined the existing BERT model to extract sentiment polarity and event types. Sentiment polarity is determined by using a preset sentiment vocabulary, using the BERT model to identify words in the text data that belong to the sentiment vocabulary, and according to the preset scores of sentiment vocabulary, determining the sentiment polarity of the text data. After the BERT model classifies and identifies the types of public opinion events, we assign weights, extract the number of penalty records from government public text data, and classify them according to the penalty types. Using the formula The enterprise risk volatility index is constructed by integrating the sentiment polarity of text data, the weight of public opinion event types and the level of penalty event types; Among them, t represents the time point t, the time window is divided into H time lengths, the time point where the current enterprise is located is taken as the end point of the time window, and the time window where the current enterprise is located is determined. represents the enterprise risk volatility index at time point t, represents the type weight of the i-th public opinion event in the time window at time point t, represents the sentiment polarity score of the ith public opinion event in the time window at time point t, represents the difference between the current time and the time when public opinion event i occurs. f represents the time decay coefficient of public opinion event, which is used to control the speed at which the influence of public opinion event decays over time. It represents the severity score of the jth penalty event in the time window at time point t. y represents the penalty event attenuation coefficient, which is used to control the speed at which the influence of the penalty event decays over time. g1 represents the public opinion weight coefficient, and g2 represents the penalty weight coefficient.

5. The enterprise risk monitoring system based on the financial service platform of big data analysis according to claim 1 is characterized in that: Establish the LSTM-Attention dynamic prediction model. The specific method is as follows: Collect historical enterprise risk volatility indicators and enterprise cyclical volatility indicators as prediction data, divide the historical prediction data into input data and control data in chronological order, use the input data as the input training model of the LSTM-Attention model, and record the input data as , each input Different weight matrices obtained through the deep learning process are mapped to different vectors. The vectors obtained through the mapping calculate the attention scores belonging to different weights, which are used to determine which texts of the input data the model pays attention to. The stacked RNN layers are used as encoders and decoders. The encoder combines all input time steps to process the input data to obtain a compact representation of the input data. , which is the compressed format of the input. The decoder receives the context vector and generates output data. In the LSTM-based seq2seq, an Attention Layer is added between the Encoder and the Decoder to help the compact representation A encode the information from all input time steps.

6. The enterprise risk monitoring system based on the financial service platform of big data analysis according to claim 1 is characterized in that: The final information vector is calculated to capture long-term dependencies as follows: The vectors of the upper and lower time points are recorded as the state sequence vector s, and the formula is used represents the final information vector, where T represents the length of the selected time step, t represents the time t, represents the state sequence vector at time t-1, is the hidden state of the model, and a(*) indicates that the parameters are normalized by the softmax function.

7. The enterprise risk monitoring system of the financial service platform based on big data analysis according to claim 1 is characterized in that: The enterprise risk score is constructed by analyzing the current enterprise cyclical fluctuation index and enterprise risk fluctuation index. The specific method is as follows: Using the formula represents the enterprise risk score, where t represents time point t, RS(t) represents the enterprise risk score at time point t, P(t) represents the enterprise cyclical fluctuation index at time point t, and R(t) represents the enterprise risk fluctuation index at time point t. It represents the mean value of the enterprise cyclical fluctuation index in the time window with t as the end point, It represents the standard deviation of the enterprise cyclical fluctuation index in the time window with t as the end point, It represents the mean value of the enterprise risk volatility index in the time window with t as the end point, It represents the standard deviation of the enterprise risk volatility index in the time window with t as the end point, represents the weight coefficient of enterprise cyclical fluctuation, Represents the enterprise risk volatility weight coefficient.

8. The enterprise risk monitoring system based on the financial service platform of big data analysis according to claim 1 is characterized in that: The enterprise risk score is combined with the predicted value at a future time point to determine the enterprise's forward-looking risk. The specific method is as follows: Based on the financial data characteristics of the enterprise at the current time point, the LSTM-Attention model is used to obtain the predicted value of the financial data characteristics of the enterprise at the future time point. The enterprise risk score at the current time point is combined to determine the enterprise's forward-looking risk probability index. If the enterprise's forward-looking risk probability index at the current time point is greater than the enterprise's forward-looking risk probability index threshold V, it is judged that the current enterprise is in a high-risk time window and the enterprise risk disposal process needs to be initiated. If the enterprise's forward-looking risk probability index at the current time point is less than or equal to the threshold V, it is judged that the current enterprise is not in a high-risk time window and the enterprise risk disposal process does not need to be initiated.

9. The enterprise risk monitoring system of the financial service platform based on big data analysis according to claim 8 is characterized in that: The enterprise risk score at the current time point is comprehensively used to determine the enterprise forward-looking risk probability index. The specific method is as follows: Using the formula represents the enterprise forward-looking risk probability index, where t represents time point t, Fx(t) represents the enterprise forward-looking risk probability index at time point t, G represents the number of preset simulation time points, i represents the i-th time point, represents the predicted value of the enterprise cyclical fluctuation index at the t+ith time point, represents the predicted value of the enterprise risk volatility index at the t+ith time point, P(t) represents the enterprise cyclical volatility index at the time point t, is the time decay coefficient.

Images