A big data-based enterprise data asset auditing method and system

Abstract

The present application belongs to the technical field of data asset management, and discloses an enterprise data asset auditing method and system based on big data, which comprises collecting financial information data, business operation data, management decision data and external industry data of an enterprise; preprocessing the collected financial information data, business operation data, management decision data and external industry data to obtain financial feature data set, business operation feature data set, management decision feature data set and external industry feature data set; extracting features from the financial feature data set, business operation feature data set, management decision feature data set and external industry feature data set and performing weighted fusion to obtain comprehensive feature data set; constructing an auditing risk prediction model, inputting the comprehensive feature data set into the auditing risk prediction model, and predicting the auditing risk coefficient of the enterprise data asset; and providing a more comprehensive perspective, which is helpful for more accurately evaluating the data asset risk of the enterprise.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data asset management, more particularly, the present application relates to an enterprise data asset auditing method and system based on big data. BACKGROUND

[0002] The patent with application publication number CN115375417A discloses a comprehensive financial audit system based on big data, which includes a data collection module, a data analysis module, a learning prediction module, and a database. The data collection module collects structured and semi-structured enterprise intangible asset evaluation data and related industry data based on big data technology, and also collects enterprise intangible asset audit data. The data analysis module analyzes and compares the data collected by the data collection module with historical data in the database to comprehensively analyze the un-audited data of enterprise intangible assets. This comprehensive financial audit system based on big data improves traditional asset evaluation, enhances the reliability of intangible asset evaluation results by combining big data technology, improves audit quality, and reduces evaluation difference risk.

[0003] The existing enterprise data asset auditing method and system have the following main problems:

[0004] The use of cross-validation to evaluate model performance on different data subsets may increase the risk of overfitting or underfitting, affecting the overall performance and reliability of the model. The lack of cross-validation makes the performance evaluation of the model dependent on the division results of a single data set. The number of folds is not reasonably adjusted using the number of folds formula, and for cases with many combinations of hyperparameters, model evaluation may be very time-consuming and waste of computing resources. Excessive computing resources are consumed in unnecessary model training and validation, reducing computing efficiency. The lack of standardized hyperparameter optimization processes such as grid search and cross-validation may result in significant differences in results from different experiments due to different hyperparameter settings.

[0005] The use of a dynamically adjusted threshold formula is not used, and risk assessment usually relies on fixed threshold settings. Such fixed thresholds may not be able to adapt to the risk characteristics of different enterprises and industries, resulting in a lack of flexibility in evaluation results and an inability to accurately reflect the actual risks faced by enterprises. The use of median, standard deviation, and percentile statistics to adjust the threshold is not considered, and the evaluation results are easily affected by outliers or extreme data. Outliers may cause the fixed threshold to be too high or too low, affecting the accuracy and stability of risk assessment.

[0006] Without optimizing the determination of the risk threshold through the coefficient adjustment model, the traditional method relies on simple statistics or a fixed value set by man; it can lead to an inaccurate risk threshold that cannot accurately reflect the actual risk status of the enterprise; without using factors such as the number of features and variance to optimize the adjustment coefficient, the robustness and flexibility of the evaluation model will be reduced; the model can be too sensitive or not sensitive enough to some specific data features, making it difficult to cope with complex and changing risk scenarios, resulting in unstable evaluation results.

[0007] In view of this, the present application provides an enterprise data asset auditing method and system based on big data to solve the above problems. SUMMARY

[0008] In order to overcome the above-mentioned defects of the prior art, in order to achieve the above-mentioned purpose, the present application provides the following technical scheme: an enterprise data asset auditing method based on big data, comprising:

[0009] S1, collecting financial information data, business operation data, management decision data and external industry data of the enterprise;

[0010] S2, preprocessing the collected financial information data, business operation data, management decision data and external industry data to obtain financial feature data set, business operation feature data set, management decision feature data set and external industry feature data set;

[0011] Feature extraction and weighted fusion are performed on the financial feature data set, the business operation feature data set, the management decision feature data set and the external industry feature data set to obtain a comprehensive feature data set;

[0012] S3, constructing an audit risk prediction model, inputting the comprehensive feature data set into the audit risk prediction model, and predicting the audit risk coefficient of the enterprise data asset;

[0013] S4, comparing the predicted audit risk coefficient of the enterprise data asset with the preset enterprise data asset audit risk coefficient threshold, and evaluating the audit risk level faced by the enterprise data asset;

[0014] S5, according to the audit risk level faced by the enterprise data asset, formulating a corresponding asset optimization management strategy.

[0015] Further, the collected financial information data of the enterprise includes income and expenditure data, asset data, liability data, cash flow data and tax data; the business operation data includes production data, inventory data, sales data, supply chain data, customer behavior data and operation performance data;

[0016] The management decision data includes strategic planning data, performance evaluation data, internal audit data, budget and prediction data, and decision support data; and the external industry data includes industry market data, economic environment data, policy and regulation data, and technical trend data.

[0017] Further, the method for preprocessing the collected financial information data, business operation data, management decision data, and external industry data to obtain the financial information dataset, the business operation dataset, the management decision dataset, and the external industry dataset comprises:

[0018] The missing values in the financial information data, the business operation data, the management decision data, and the external industry data are filled using a linear interpolation method; the estimated values of the missing values are calculated using the linear relationship of the non-missing values before and after the missing values, and the original missing values are replaced; the uniqueness of the financial information data, the business operation data, the management decision data, and the external industry data is checked, and the data with repeated records is removed;

[0019] The financial information data, the business operation data, the management decision data, and the external industry data are standardized and normalized to convert the data with different units and dimensions into a standard normal distribution, to obtain the financial information dataset, the business operation dataset, the management decision dataset, and the external industry dataset.

[0020] Further, the method for extracting features and weighted fusion of the financial information dataset, the business operation dataset, the management decision dataset, and the external industry dataset to obtain the comprehensive feature dataset comprises:

[0021] The method for extracting features of the financial information dataset is: calculating financial indicators, and performing financial ratio analysis on the financial information data; performing time series analysis on the obtained financial indicators, and extracting trend, seasonal, and periodic features of the time series data; integrating the obtained financial indicators and the trend, seasonal, and periodic features of the data to obtain the financial feature dataset;

[0022] The method for extracting features of the business operation dataset is: performing time series analysis on sales data in the business operation dataset, and extracting seasonal, trend, and periodic features; performing production efficiency analysis on production data, and extracting production rate, equipment utilization rate, and production cycle time features; calculating inventory turnover rate, inventory holding cost, and supply-demand balance features for inventory data; and performing customer feature analysis based on customer behavior data, and the specific method is:

[0023] S41, RFM features of the customer are extracted; R, F, and M indicators are calculated for each customer respectively, and the R, F, and M indicators are scored according to the calculation results;

[0024] S42, based on the R, F, M index score, combining the scores of the customer in the R, F, M three dimensions to obtain the R, F, M score combination, each combination represents a different customer type, and calculating the comprehensive score of each customer;

[0025] The comprehensive score calculation formula is:

[0026] ;

[0027] Wherein, is the total score of the i-th customer; is the R index score of the i-th customer; is the F index score of the i-th customer; is the M index score of the i-th customer; is the R index score weight coefficient of the i-th customer; is the F index score weight coefficient of the i-th customer; is the M index score weight coefficient of the i-th customer; is the index of the customer;

[0028] S43, the customers are preliminarily grouped according to different score combinations, each combination represents a different customer type, and the score combination is limited by the score combination limiting model;

[0029] The score combination limiting model is:

[0030] ;

[0031] Wherein, is the proportion of customers belonging to a certain customer type in the total customer group; is the minimum comprehensive score in the customer group; is the maximum comprehensive score in the customer group; is the proportion of score dimensions to the total number of customers; is the total number of customers; is the type of customer; is the number of RFM score dimensions;

[0032] S44, the R, F, M index score of each customer is standardized; the standardization formula is:

[0033] ;

[0034] Wherein, is the standardized index score;​​​​​​​ R, F, M index scores for the customer

[0035] Further subdivide the customers using the K-means clustering algorithm; the method for feature extraction of the management decision dataset is: SWOT analysis and key performance indicator analysis of the management decision data;

[0036] The method for feature extraction of the external industry dataset is: using time series analysis method to analyze market trends and macroeconomic features;

[0037] Through the above steps, the financial information dataset, business operation dataset, management decision dataset and external industry dataset after feature extraction are obtained; the financial information dataset, business operation dataset, management decision dataset and external industry dataset after feature extraction are weighted and fused to obtain a comprehensive feature dataset.

[0038] Further, the method for weighting and fusing the financial feature dataset, business operation feature dataset, management decision feature dataset and external industry feature dataset after feature extraction to obtain a comprehensive feature dataset comprises:

[0039] The financial feature dataset, business operation feature dataset, management decision feature dataset and external industry feature dataset after feature extraction are fused through a weighted model to form a comprehensive feature dataset, the financial feature dataset is denoted as , the business operation feature dataset is denoted as , the management decision feature dataset is denoted as ; and the external industry feature dataset is denoted as ;

[0040] The weighted model is:

[0041] ;

[0042] wherein, is the weight coefficient of the financial feature dataset; is the weight coefficient of the business operation feature dataset; is the weight coefficient of the management decision feature dataset; is the weight coefficient of the external industry feature dataset.

[0043] Further, the training method of the audit risk prediction model comprises:

[0044] ​​​​The dataset is divided into training, validation, and test sets. The dataset includes historical comprehensive feature datasets and corresponding audit risk coefficients of enterprise asset data. An audit risk prediction model is constructed, which includes an input layer, a GRU layer, a fully connected layer, and an output layer. The input layer of the audit risk prediction model is used to input the historical comprehensive feature dataset, and the number of neurons in the input layer matches the number of features in the historical comprehensive feature dataset.

[0045] GRU layers are used to process historical comprehensive feature datasets, and the number of GRU layers and neurons is adjusted according to task complexity; fully connected layers are used to provide additional nonlinear transformations; the model output layer is used to output network signal performance indicators, and a single neuron outputs the predicted value, using the identity function as the activation function; the audit risk prediction model is a gated recurrent unit model.

[0046] The mean absolute error (MAE) is used as the loss function to measure the error between the model's predicted values ​​and the actual values; the MAE loss function is:

[0047] ;

[0048] in, Mean absolute error; Number of datasets; For the first The true value of the audit risk coefficient for an enterprise's asset data; For the first Predicted values ​​of audit risk coefficients for individual enterprise asset data;

[0049] The model is trained using the training set data, and the loss function is minimized using the Adam optimizer. The model's performance is evaluated using the validation set, and the model's hyperparameters are tuned until the model's performance no longer improves significantly or reaches the preset stopping condition.

[0050] The performance of the model in the prediction task is evaluated using a test set. The current comprehensive feature dataset is input into the trained audit risk prediction model to obtain the audit risk coefficient of the enterprise's data assets.

[0051] Furthermore, the method for tuning the hyperparameters of the model includes:

[0052] S71, There are two preset hyperparameters: learning rate and... and GRU layer number The two hyperparameter combinations are traversed using a grid search, with a preset learning rate. The range of values ​​for is:

[0053] ;

[0054] in, the learning rate the number of values of the GRU layer the value range of the GRU layer

[0055] ;

[0056] wherein, the number of values of the GRU layer ;

[0057] S72, for each hyperparameter combination, calculating the k-fold cross-validation error of the hyperparameter combination by a k-fold cross-validation error formula, wherein the k-fold cross-validation error formula is: ;

[0058] ;

[0059] wherein, is the average validation error obtained by k-fold cross-validation under the hyperparameter combination ; is the value of the i-th learning rate; is the value of the j-th GRU layer; is the number of folds in the k-fold cross-validation; is a loop vector traversing all the folds; is a summation from 1 to ; is the validation error of the model trained using the hyperparameter combination in the i-th fold; ; ;

[0060] S73, limiting the number of folds in the k-fold cross-validation by a k-fold limit formula, wherein the k-fold limit formula is: ;

[0061] ;

[0062] wherein, is the total number of hyperparameter combinations; is the number of types of hyperparameters; is a floor function;

[0063] S74, selecting the hyperparameter combination that makes ​​​​​​​​​The combination of hyperparameters with the minimum cross-validation error is taken as the optimal combination of hyperparameters of the model.

[0064] Further, the method of comparing the predicted audit risk coefficient of the enterprise data asset with the preset audit risk coefficient threshold of the enterprise data asset to evaluate the audit risk level faced by the enterprise data asset comprises:

[0065] S81, preset audit risk coefficient threshold, limit the audit risk coefficient threshold through the threshold adjustment formula; the threshold adjustment formula is:

[0066] ;

[0067] Among them, is the audit risk coefficient threshold; is the median of the historical audit risk coefficient; is the standard deviation of the historical audit risk coefficient; is the percentile of the historical audit risk coefficient; is the adjustment coefficient of the difference between the median and the percentile of the historical audit risk coefficient;

[0068] S82, adjust the adjustment coefficient of the difference between the median and the percentile of the historical audit risk coefficient through the coefficient adjustment model, the coefficient adjustment model is:

[0069] ;

[0070] Among them, is the number of features affecting the historical audit risk coefficient; is the number of historical audit risk coefficients; is the variance of the historical audit risk coefficient;

[0071] S83, when the value of the first quartile, the audit risk coefficient threshold is recorded as ; when the value of the third quartile, the audit risk coefficient threshold is recorded as ;

[0072] S84, further divide the audit risk level threshold of the enterprise data asset according to the preset audit risk coefficient threshold; record the predicted audit risk coefficient as ;

[0073] S85, compare the predicted audit risk coefficient with the preset audit risk coefficient threshold to determine the audit risk level faced by the enterprise data asset; if , it is determined that the audit risk level faced by the enterprise data asset is low risk; if If the audit risk level of the enterprise data asset is low, then the audit risk level of the enterprise data asset is determined to be medium risk; if If the audit risk level of the enterprise data asset is low, then the audit risk level of the enterprise data asset is determined to be high risk.

[0074] Further, the method of formulating a corresponding asset optimization management strategy according to the audit risk level of the enterprise data asset comprises:

[0075] For enterprise data assets with low audit risk, the management strategies include data quality inspection, data classification, hierarchical management, setting local data access control permissions, and starting a local data audit tracking mechanism.

[0076] For enterprise data assets with medium audit risk, the management strategies include implementing joint data protection measures, strengthening data access control strategies, conducting comprehensive permission management combined with role and attribute control, establishing data backup and recovery mechanisms, encrypting data storage and transmission, and implementing user behavior analysis and audit to detect abnormal data operations and generate warning information through an enterprise intelligent management terminal.

[0077] For enterprise data assets with high audit risk, comprehensive optimization management strategies are adopted, including comprehensive audit of all data, monitoring of all data operation behaviors, generation of audit reports through an enterprise intelligent management terminal, deployment of DLP data leakage prevention technology, use of high-strength encryption algorithms for data encryption, and automatic interception of unauthorized data transmission through an enterprise intelligent management terminal.

[0078] An enterprise data asset audit system based on big data comprises:

[0079] A data collection module for collecting financial information data, business operation data, management decision data, and external industry data of an enterprise;

[0080] A data processing module for preprocessing the collected financial information data, business operation data, management decision data, and external industry data to obtain financial information data sets, business operation data sets, management decision data sets, and external industry data sets;

[0081] Feature extraction and weighted fusion are performed on the financial information data sets, business operation data sets, management decision data sets, and external industry data sets to obtain a comprehensive feature data set;

[0082] A risk prediction module for constructing an audit risk prediction model, inputting the comprehensive feature data set into the audit risk prediction model, and predicting an audit risk coefficient of the enterprise data asset;

[0083] The risk assessment module is configured to compare the predicted audit risk coefficient of the enterprise data asset with a preset audit risk coefficient threshold of the enterprise data asset, and assess the audit risk level faced by the enterprise data asset.

[0084] The adjustment optimization module is configured to develop a corresponding optimized data management strategy according to the audit risk level faced by the enterprise data asset.

[0085] The technical effects and advantages of the enterprise data asset audit method and system based on big data are as follows:

[0086] Through cross-validation, the performance of the model on different data subsets can be effectively evaluated, ensuring the stability of the training process of the model; the multiple-fold validation can reduce the influence of the randomness of data division on the evaluation results of the model, improving the reliability of the results; the fold limit formula adjusts the number of folds according to the number of hyperparameter combinations, avoiding unnecessary computational overhead; the model training is more efficient, especially in the case of a large number of hyperparameter combinations, which can significantly reduce the computational cost; the standardized process of the grid search and cross-validation method makes the experimental results more reproducible;

[0087] By introducing the threshold adjustment formula, combining the median, standard deviation, and percentile of the historical audit risk coefficient, and dynamically adjusting the audit risk coefficient threshold, the risk assessment method can better adapt to the risk characteristics of different enterprises and industries, avoid the inflexibility caused by a fixed threshold, and make the risk assessment more accurate and adaptable; using the median, standard deviation, and percentile to calculate the threshold of the audit risk coefficient makes the evaluation method more scientific and comprehensive; it can reduce the influence of outliers on the results and provide more stable risk assessment results;

[0088] By adjusting the difference between the median and the percentile of the historical audit risk coefficient using the coefficient adjustment model, and further optimizing the adjustment coefficient using factors such as the number of features and variance, the robustness and flexibility of the model can be enhanced, making the determination of the risk threshold more accurate; by using mathematical models and data analysis methods to determine the audit risk level, the subjectivity in the evaluation process is reduced; the risk assessment process is more standardized, transparent, and interpretable, which helps to improve the objectivity and consistency of audit and management decisions. BRIEF DESCRIPTION OF DRAWINGS

[0089] Figure 1 The figure is a flowchart of an enterprise data asset audit method based on big data according to the present application;

[0090] Figure 2 The figure is a structural diagram of an enterprise data asset audit system based on big data according to the present application. DETAILED DESCRIPTION

[0091] With reference to the accompanying drawings, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts are within the scope of the present application.

[0092] Embodiment 1

[0093] Please refer to Figure 1 The enterprise data asset auditing method based on big data in the embodiment comprises:

[0094] S1, collecting financial information data, business operation data, management decision data and external industry data of an enterprise;

[0095] S2, preprocessing the collected financial information data, business operation data, management decision data and external industry data to obtain a financial feature data set, a business operation feature data set, a management decision feature data set and an external industry feature data set;

[0096] S3, constructing an auditing risk prediction model, inputting the comprehensive feature data set into the auditing risk prediction model, and predicting an auditing risk coefficient of the enterprise data asset;

[0097] S4, comparing the predicted auditing risk coefficient of the enterprise data asset with a preset auditing risk coefficient threshold of the enterprise data asset, and evaluating an auditing risk level faced by the enterprise data asset;

[0098] S5, formulating a corresponding asset optimization management strategy according to the auditing risk level faced by the enterprise data asset.

[0099] S5, formulating a corresponding asset optimization management strategy according to the auditing risk level faced by the enterprise data asset.

[0100] The financial information data of the enterprise comprises income and expenditure data, asset data, liability data, cash flow data and tax data; the business operation data comprises production data, inventory data, sales data, supply chain data, customer behavior data and operation performance data;

[0101] The management decision data comprises strategic planning data, performance evaluation data, internal audit data, budget and prediction data and decision support data; the external industry data comprises industry market data, economic environment data, policy and regulation data and technical trend data;

[0102] The collected financial information data, business operation data, management decision data and external industry data are collected through an enterprise internal system, a data integration platform and an external data service platform.

[0103] The collected financial information data, business operation data, management decision data and external industry data are preprocessed to obtain a financial information dataset, a business operation dataset, a management decision dataset and an external industry dataset.

[0104] The missing values in the financial information data, business operation data, management decision data and external industry data are filled using a linear interpolation method; the estimated values of the missing values are calculated using the linear relationship of the non-missing values before and after the missing values, and the original missing values are replaced; the uniqueness of the financial information data, business operation data, management decision data and external industry data is checked, and the data with repeated records is removed.

[0105] The financial information data, business operation data, management decision data and external industry data are standardized and normalized to convert data with different units and dimensions into a standard normal distribution to obtain the financial information dataset, the business operation dataset, the management decision dataset and the external industry dataset.

[0106] The financial information dataset, the business operation dataset, the management decision dataset and the external industry dataset are subjected to feature extraction and weighted fusion to obtain a comprehensive feature dataset, and the method comprises the following steps:

[0107] The method for extracting features from the financial information dataset comprises the following steps: calculating financial indicators, and performing financial ratio analysis on the financial information data; performing time series analysis on the obtained financial indicators to extract trend, seasonal and periodic features of the time series data; and integrating the obtained financial indicators and the trend, seasonal and periodic features of the data to obtain a financial feature dataset.

[0108] The method for extracting features from the business operation dataset comprises the following steps: performing time series analysis on sales data in the business operation dataset to extract seasonal, trend and periodic features; performing production efficiency analysis on production data to extract productivity, equipment utilization rate and production cycle time features; calculating inventory turnover rate, inventory holding cost and supply-demand balance features for inventory data; and performing customer feature analysis based on customer behavior data, specifically as follows:

[0109] S41. Extract the RFM characteristics of customers; calculate the R, F, and M indicators for each customer, and score the R, F, and M indicators based on the calculation results; calculate the RFM value for each customer: R: calculate the time interval between the most recent purchase date p and the analysis date p' for each customer; it can be in units of days, weeks, or months; F: calculate the total number of purchases for each customer within a specified time window t; M: calculate the total purchase amount for each customer within a specified time window t.

[0110] The values ​​of R, F, and M are divided into multiple levels (usually from 1 to 5), and the scoring criteria are as follows:

[0111] The smaller the R value, the more active the customer and the higher the score; the larger the F value, the more frequently the customer makes purchases and the higher the score; the larger the M value, the more money the customer spends and the higher the score.

[0112] S42. Based on the R, F, M indicators, combine the customer's scores in the three dimensions of R, F, and M to obtain the R, F, M score combination. Each combination represents a different customer type, and calculate the comprehensive score for each customer.

[0113] The formula for calculating the overall score is:

[0114] ;

[0115] in, For the first The total score of each customer; For the first R-index score for each customer; For the first F-index score for each customer; For the first M-index score for each customer; For the first Weighting coefficients for each customer's R-index score; For the first The weighting coefficient of the F-index score for each customer; For the first Weighting coefficients for the M-index score of each customer; For customers' index;

[0116] S43. Initially group customers according to different score combinations, with each combination representing a different customer type, and restrict the score combinations using a score combination constraint model;

[0117] The score combination constraint model is as follows:

[0118] ;

[0119] in, The proportion of a particular customer type within the total customer base; The lowest overall score among the customer groups; The highest overall score among the customer groups; The ratio of rating dimensions to the total number of customers; The total number of customers; For different types of customers; The number of rating dimensions for RFM;

[0120] when Increasing the size means that customers are segmented more finely, which can be adjusted... To control the differences in scores for different types of subjects.

[0121] If it is necessary to reduce the number of categories, the impact of the rating dimensions on customers can be appropriately reduced, i.e., decreased. Or increase the total number of customers .

[0122] S44. Standardize the R, F, and M score for each customer; the standardization formula is:

[0123] ;

[0124] in, Scoring of standardized indicators; For customers R, F, M index scores; The average score of the indicator; The standard deviation of the indicator scores; Index for customers;

[0125] The K-means clustering algorithm is used to further segment customers; the method for feature extraction of the management decision dataset is to perform SWOT analysis and key performance indicator analysis on the management decision data.

[0126] The method for feature extraction from external industry datasets is as follows: time series analysis is used to analyze market trends and extract macroeconomic features;

[0127] Through the above steps, we obtain the financial information dataset, business operation dataset, management decision dataset, and external industry dataset after feature extraction. We then perform weighted fusion on these datasets to obtain a comprehensive feature dataset.

[0128] Methods for weighted fusion of the extracted financial feature dataset, business operation feature dataset, management decision feature dataset, and external industry feature dataset to obtain a comprehensive feature dataset include:

[0129] The extracted financial feature dataset, business operation feature dataset, management decision feature dataset, and external industry feature dataset are fused using a weighted model to form a comprehensive feature dataset. The financial feature dataset is denoted as... The business operation feature dataset is denoted as The management decision feature dataset is denoted as External industry feature datasets are denoted as ;

[0130] The weighted model is:

[0131] ;

[0132] in, These are the weighting coefficients for the financial feature dataset; Weight coefficients for the business operation feature dataset; Weight coefficients for the management decision feature dataset; These are the weight coefficients for the external industry feature dataset.

[0133] Training methods for audit risk prediction models include:

[0134] The dataset is divided into training, validation, and test sets. The dataset includes historical comprehensive feature datasets and corresponding audit risk coefficients of enterprise asset data. An audit risk prediction model is constructed, which includes an input layer, a GRU layer, a fully connected layer, and an output layer. The input layer of the audit risk prediction model is used to input the historical comprehensive feature dataset, and the number of neurons in the input layer matches the number of features in the historical comprehensive feature dataset.

[0135] GRU layers are used to process historical comprehensive feature datasets, and the number of GRU layers and neurons is adjusted according to task complexity; fully connected layers are used to provide additional nonlinear transformations; the model output layer is used to output network signal performance indicators, and a single neuron outputs the predicted value, using the identity function as the activation function; the audit risk prediction model is a gated recurrent unit model.

[0136] The mean absolute error (MAE) is used as the loss function to measure the error between the model's predicted values ​​and the actual values; the MAE loss function is:

[0137] ;

[0138] in, Mean absolute error; Number of datasets; For the first The true value of the audit risk coefficient for an enterprise's asset data; For the first Predicted values ​​of audit risk coefficients for individual enterprise asset data;

[0139] The model is trained using the training set data, and the loss function is minimized using the Adam optimizer. The model's performance is evaluated using the validation set, and the model's hyperparameters are tuned until the model's performance no longer improves significantly or reaches the preset stopping condition.

[0140] The performance of the model in the prediction task is evaluated using a test set. The current comprehensive feature dataset is input into the trained audit risk prediction model to obtain the audit risk coefficient of the enterprise's data assets.

[0141] Methods for tuning the hyperparameters of a model include:

[0142] S71, There are two preset hyperparameters: learning rate and... and GRU layer number The two hyperparameter combinations are traversed using a grid search, with a preset learning rate. The range of values ​​for is:

[0143] ;

[0144] in, For learning rate Number of possible values; Preset number of GRU layers The range of values ​​for is:

[0145] ;

[0146] in, GRU layer number The number of possible values;

[0147] S72. For each combination of hyperparameters, through... The cross-validation error formula is used to calculate the hyperparameter combination. Cross-validation error The formula for the cross-validation error is:

[0148] ;

[0149] in, In hyperparameter combination Next, through The average validation error obtained from cross-validation; For the first The learning rate can take any value; For the first The possible values ​​for the number of GRU layers; for Number of folds in cross-validation; For a traversal of all A cyclic vector; To From 1 to Summation, accumulating all The verification error of the fold is used to calculate its average value; In the first Using hyperparameter combinations as a compromise Validation error after model training;

[0150] S73, Through Folding Restriction Formula Number of folds in cross-validation To impose restrictions, The formula for limiting the fold is:

[0151] ;

[0152] in, The total number of hyperparameter combinations; The number of hyperparameter types; This is the floor function;

[0153] S74, Selecting makes The combination of hyperparameters that minimizes the cross-validation error is taken as the optimal combination of hyperparameters for the model.

[0154] For example, set the range of hyperparameter values: the learning rate ranges to 0.001, 0.01, and 0.1; the number of GRU layers ranges to 1, 2, and 3; use a grid search method to traverse all possible combinations of hyperparameters: combination 1: learning rate = 0.001, number of GRU layers = 1;

[0155] Combination 2: Learning rate = 0.001, number of GRU layers = 2;

[0156] Combination 3: Learning rate = 0.001, number of GRU layers = 3;

[0157] Combination 4: Learning rate = 0.01, number of GRU layers = 1;

[0158] Combination 5: Learning rate = 0.01, number of GRU layers = 2;

[0159] Combination 6: Learning rate = 0.01, number of GRU layers = 3;

[0160] Combination 7: Learning rate = 0.1, number of GRU layers = 1;

[0161] Combination 8: Learning rate = 0.1, number of GRU layers = 2;

[0162] Combination 9: Learning rate = 0.1, number of GRU layers = 3;

[0163] Suppose we choose to use 5-fold cross-validation to evaluate the performance of each hyperparameter combination: For each hyperparameter combination, perform the following steps: Divide the dataset into 5 folds, for each fold, use 4 folds as the training set and the remaining 1 fold as the validation set; train the model and evaluate the error on the validation set;

[0164] Repeat the above process 5 times, selecting a different fold as the validation set each time, and calculate the average of the validation errors for each iteration to obtain the fold cross-validation error for this hyperparameter combination. Assume that when using combination 1, the validation errors obtained for each iteration are as follows:

[0165] Verification error 1: 0.15;

[0166] Verification error 2: 0.14;

[0167] Verification error 3: 0.16;

[0168] Verification error 4: 0.13;

[0169] Verification error 5: 0.15;

[0170] The cross-validation error for combination 1 is:

[0171] The cross-validation error is 0.15 + 0.14 + 0.16 + 0.13 + 0.15 / 5 = 0.146; if the total number of hyperparameter combinations is 9 and the number of hyperparameter types is 2 (learning rate and GRU layer number), then... The folding constraint formula is used to determine a reasonable number of folds; The formula for limiting the fold is:

[0172] ;

[0173] The fold limit here indicates that 5-fold cross-validation can be selected to reduce computation. Suppose that after grid search and fold cross-validation, the following fold cross-validation error is obtained:

[0174] The error of combination 1 is 0.146;

[0175] The error of combination 2 is 0.143;

[0176] The error of combination 3 is 0.147.

[0177] The error of combination 4 is 0.141;

[0178] The error of combination 5 is 0.139;

[0179] The error of combination 6 is 0.144;

[0180] The error of combination 7 is 0.148.

[0181] The error of combination 8 is 0.146;

[0182] The error of combination 9 is 0.142.

[0183] The hyperparameter combination with the smallest error is selected as the optimal hyperparameter combination. In this example, combination 5 (learning rate = 0.01, number of GRU layers = 2) has the smallest error of 0.139, so it is the optimal hyperparameter combination ultimately chosen.

[0184] Methods for assessing the audit risk level of enterprise data assets by comparing the predicted audit risk coefficient with the preset audit risk coefficient threshold include:

[0185] S81. Preset an audit risk coefficient threshold, and limit the audit risk coefficient threshold using a threshold adjustment formula; the threshold adjustment formula is:

[0186] ;

[0187] in, This is the threshold for the audit risk coefficient; This represents the median of historical audit risk coefficients. The standard deviation of historical audit risk coefficients; Percentile of historical audit risk coefficient; This is an adjustment factor for the difference between the median and percentile of historical audit risk coefficients;

[0188] S82. Adjustment coefficient for the difference between the median and percentile of historical audit risk coefficients using a coefficient adjustment model. Adjustments are made, and the coefficient adjustment model is as follows:

[0189] ;

[0190] in, The number of features that affect the historical audit risk coefficient; This represents the number of historical audit risk coefficients; The variance of the historical audit risk coefficient;

[0191] For example, the number of features that affect the historical audit risk coefficient. The number of historical audit risk coefficients is 8. Given a median of 2 and a variance of 1 for the historical audit risk coefficient, the adjustment factor for the difference between the median and percentile of the historical audit risk coefficient is... Approximately 0.5;

[0192] S83, when When the value is the first quartile, the audit risk coefficient threshold is denoted as... ;when When the value is the third quartile, the audit risk coefficient threshold is recorded as... ;

[0193] S84. Based on the preset audit risk coefficient threshold, further classify the audit risk level thresholds faced by the enterprise's data assets; denot the predicted audit risk coefficient as... ;

[0194] S85. Compare the predicted audit risk coefficient with the preset audit risk coefficient threshold to determine the audit risk level faced by the enterprise's data assets; if If so, the audit risk level of the enterprise's data assets is determined to be low; if If so, the audit risk level of the enterprise's data assets is determined to be medium risk; if If so, the audit risk level of the enterprise's data assets is determined to be high.

[0195] For example, suppose a set of enterprise data assets has an audit risk coefficient of 0.7; based on the dataset, the first quartile (Q1) is calculated to be 0.4; the third quartile (Q3) is 0.8; therefore:

[0196] T1 = 0.4 (first quartile); T2 = 0.8 (third quartile);

[0197] Now assume the predicted audit risk coefficient R is 0.7; according to the above rule: 0.4 < 0.7 <= 0.8, therefore, the audit risk level faced by the enterprise's data assets is medium risk.

[0198] Methods for developing corresponding asset optimization management strategies based on the audit risk level faced by enterprise data assets include:

[0199] For enterprise data assets with low audit risk, the management strategies include conducting data quality checks, data classification, hierarchical management, setting local data access control permissions, and initiating local data audit tracking mechanisms.

[0200] For enterprise data assets with moderate audit risk, the management strategies adopted include implementing joint data protection measures, strengthening data access control strategies, conducting comprehensive access control in combination with role and attribute control, establishing data backup and recovery mechanisms, encrypting data storage and transmission, implementing user behavior analysis and auditing, detecting abnormal data operations, and generating early warning information through enterprise intelligent management terminals.

[0201] For enterprise data assets with high audit risks, a comprehensive optimization management strategy is adopted to conduct a full audit of all data, monitor all data operation behaviors, and generate audit reports through the enterprise intelligent management terminal; DLP data leakage prevention technology is deployed, which uses high-strength encryption algorithms to encrypt data and issues interception commands through the enterprise intelligent management terminal to automatically intercept unauthorized data transmission.

[0202] The preset threshold for audit risk coefficient of enterprise data assets is set by staff. Different audit risk coefficients of enterprise data assets are collected through the enterprise intelligent management terminal, and the average value of multiple audit risk coefficients of enterprise data assets is taken as the preset threshold for audit risk coefficient of enterprise data assets.

[0203] In this embodiment, cross-validation effectively evaluates the model's performance on different data subsets, ensuring the stability of the model's training process. Multi-fold validation reduces the impact of randomness in data partitioning on model evaluation results, improving the reliability of the results. The fold limit formula adjusts the fold number based on the number of hyperparameter combinations, avoiding unnecessary computational overhead. This makes model training more efficient, especially when there are many hyperparameter combinations, significantly reducing computational costs. The standardized process of grid search and cross-validation methods ensures higher reproducibility of experimental results.

[0204] By introducing a threshold adjustment formula and combining statistical characteristics such as the median, standard deviation, and percentiles of historical audit risk coefficients, the audit risk coefficient threshold is dynamically adjusted. This better adapts to the risk characteristics of different enterprises and industries, avoiding the inflexibility caused by fixed thresholds, making risk assessment more accurate and adaptable. Using multiple statistical characteristics such as median, standard deviation, and percentiles to calculate the audit risk coefficient threshold makes the assessment method more scientific and comprehensive. It can reduce the impact of outliers on the results and provide more stable risk assessment results.

[0205] By adjusting the difference between the median and percentile of historical audit risk coefficients using a coefficient adjustment model, and further optimizing the adjustment coefficients using factors such as feature quantity and variance, the robustness and flexibility of the model can be enhanced, making the determination of risk thresholds more accurate. By using mathematical models and data analysis methods to determine audit risk levels, the subjectivity in the assessment process is reduced. This makes the risk assessment process more standardized, transparent, and interpretable, which helps to improve the objectivity and consistency of audit and management decisions.

[0206] Example 2

[0207] Please see Figure 2 As shown in the figure, this embodiment of an enterprise data asset auditing system based on big data includes:

[0208] The data acquisition module is used to collect financial information data, business operation data, management decision data, and external industry data of enterprises;

[0209] The data processing module is used to preprocess the collected financial information data, business operation data, management decision data and external industry data to obtain financial information datasets, business operation datasets, management decision datasets and external industry datasets.

[0210] Feature extraction and weighted fusion are performed on financial information datasets, business operation datasets, management decision datasets, and external industry datasets to obtain a comprehensive feature dataset.

[0211] The risk prediction module is used to build an audit risk prediction model. The comprehensive feature dataset is input into the audit risk prediction model to predict the audit risk coefficient of the enterprise's data assets.

[0212] The risk assessment module is used to compare the predicted audit risk coefficient of enterprise data assets with the preset audit risk coefficient threshold of enterprise data assets to assess the audit risk level faced by enterprise data assets.

[0213] The adjustment and optimization module is used to formulate corresponding data management optimization strategies based on the audit risk level faced by the enterprise's data assets.

[0214] Since the electronic device described in this embodiment is the one used to implement the enterprise data asset auditing method and system based on big data in this application embodiment, those skilled in the art can understand the specific implementation method and various variations of the electronic device in this embodiment based on the enterprise data asset auditing method and system based on big data in this application embodiment. Therefore, how the electronic device implements the method in this application embodiment will not be described in detail here. Any electronic device used by those skilled in the art to implement the enterprise data asset auditing method and system based on big data in this application embodiment falls within the scope of protection of this application.

[0215] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0216] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for users of ordinary technical skills, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A big data based enterprise data asset auditing method, characterized in that, The application relates to an audit risk prediction method and device. The method comprises the following steps: S1, collecting financial information data, business operation data, management decision data and external industry data of an enterprise; S2, preprocessing the collected financial information data, business operation data, management decision data and external industry data to obtain a financial information dataset, a business operation dataset, a management decision dataset and an external industry dataset; S3, constructing an audit risk prediction model, inputting the comprehensive feature dataset into the audit risk prediction model, and predicting an audit risk coefficient of enterprise data assets; The training method of the audit risk prediction model comprises the following steps: dividing the dataset into a training set, a validation set and a test set, wherein the dataset comprises a historical comprehensive feature dataset and an audit risk coefficient of corresponding enterprise asset data; constructing an audit risk prediction model, wherein the model comprises an input layer, a GRU layer, a full connection layer and an output layer; the input layer of the audit risk prediction model is used for inputting the historical comprehensive feature dataset, and the number of neurons of the input layer is matched with the number of features of the historical comprehensive feature dataset; processing the historical comprehensive feature dataset by using the GRU layer, adjusting the number and the number of neurons of the GRU layer according to the task complexity; and providing additional nonlinear transformation through the full connection layer; the output layer of the model is used for outputting the audit risk coefficient, and a neuron is used to output a predicted value, and an identity function is used as an activation function; the audit risk prediction model is a gated recurrent unit model; using the mean absolute error as a loss function to measure the error between the predicted value and the actual value of the model; using the training set data to train the model, minimizing the loss function by using an Adam optimizer; using the validation set to evaluate the performance of the model, and optimizing the hyperparameters of the model until the performance of the model no longer improves significantly or the preset stopping condition is reached; using the test set to evaluate the performance of the model in the prediction task, inputting the current comprehensive feature dataset into the trained audit risk prediction model to obtain the audit risk coefficient of the enterprise data assets; S4, comparing the predicted audit risk coefficient of the enterprise data assets with a preset audit risk coefficient threshold of the enterprise data assets, and evaluating the audit risk level faced by the enterprise data assets; The specific method comprises the following steps: presetting the audit risk coefficient threshold, and limiting the audit risk coefficient threshold by using a threshold adjustment formula; the threshold adjustment formula is as follows: ； wherein, is an audit risk factor threshold value; is a median of historical audit risk factors; is a standard deviation of historical audit risk factors; is a percentile of historical audit risk factors; is an adjustment factor for the difference between the median and the percentile of historical audit risk factors; a coefficient adjustment model for adjusting the difference between the median and the percentile of the historical audit risk coefficients adjusting, the coefficient adjustment model being: ； wherein, is a number of characteristics that influence the historical audit risk factor; is a number of historical audit risk factors; is a variance of historical audit risk factors; S5, formulating a corresponding asset optimization management strategy according to the audit risk level faced by the enterprise data assets.

2. The big data based enterprise data asset auditing method of claim 1, wherein, The collected financial information data of the enterprise comprises income and expenditure data, asset data, liability data, cash flow data and tax data; the business operation data comprises production data, inventory data, sales data, supply chain data, customer behavior data and operation performance data; The management decision data comprises strategic planning data, performance evaluation data, internal audit data, budget and prediction data and decision support data; the external industry data comprises industry market data, economic environment data, policy and regulation data and technical trend data.

3. The big data based enterprise data asset auditing method of claim 2, wherein, The method for preprocessing the collected financial information data, business operation data, management decision data and external industry data to obtain the financial information dataset, the business operation dataset, the management decision dataset and the external industry dataset comprises: The missing values in the financial information data, the business operation data, the management decision data and the external industry data are filled by using a linear interpolation method; the estimated value of the missing value is calculated by using the linear relationship of the non-missing values before and after the missing value, and the original missing value is replaced; the uniqueness of the financial information data, the business operation data, the management decision data and the external industry data is checked, and the data with repeated records is removed; The financial information data, the business operation data, the management decision data and the external industry data are standardized and normalized to convert data with different units and dimensions into a standard normal distribution to obtain the financial information dataset, the business operation dataset, the management decision dataset and the external industry dataset.

4. The big data based enterprise data asset auditing method of claim 3, wherein, The method for extracting features and weighted fusion of the financial information dataset, the business operation dataset, the management decision dataset and the external industry dataset to obtain a comprehensive feature dataset comprises: The method for extracting features from the financial information dataset is: calculating financial indicators, and performing financial ratio analysis on the financial information data; performing time series analysis on the obtained financial indicators to extract trend, seasonal and periodic features of the time series data; integrating the obtained financial indicators and the trend, seasonal and periodic features of the data to obtain a financial feature dataset; The method for extracting features from the business operation dataset is: performing time series analysis on sales data in the business operation dataset to extract seasonal, trend and periodic features; performing production efficiency analysis on production data to extract production rate, equipment utilization rate and production cycle time features; calculating inventory turnover rate, inventory holding cost and supply-demand balance features for inventory data; and performing customer feature analysis based on customer behavior data, specifically: S41, RFM features of customers are extracted; R, F and M indicators are calculated for each customer respectively, and the R, F and M indicators are scored according to the calculation results; S42, based on the R, F and M indicator scores, the scores of customers in the R, F and M dimensions are combined to obtain R, F and M score combinations, each combination representing a different customer type, and a comprehensive score of each customer is calculated; The comprehensive score calculation formula is: ； wherein, is the total score for the th customer is the R-index score for the th customer; is the F-index score for the th customer; is the M-index score for the th customer; is the R-index score weight coefficient for the th customer; is the F-index score weight coefficient for the th customer; is the M-index score weight coefficient for the th customer; is the index of the customer; S43, the customers are preliminarily grouped according to different score combinations, each combination representing a different customer type, and the score combinations are limited by a score combination limiting model; The score combination limiting model is: ； wherein is the proportion of the total customer population that belongs to a certain customer type; is the minimum composite score in the customer population; is the maximum composite score in the customer population; is the proportion of the scoring dimensions to the total number of customers; is the total number of customers; is the kind of customers; is the number of scoring dimensions of RFM; S44, the R, F and M indicator scores of each customer are standardized; the standardization formula is: ； in, Scoring of standardized indicators; For customers R, F, M index scores; The average score of the indicator; The standard deviation of the indicator scores; Index for customers; K-means clustering algorithm is used to further subdivide the customers; the method for extracting features from the management decision dataset is: performing SWOT analysis and key performance indicator analysis on the management decision data; The method for extracting features from the external industry dataset is: using time series analysis method to analyze market trends and extract macroeconomic features; The financial feature data set, the business operation feature data set, the management decision feature data set and the external industry feature data set after feature extraction are fused by weighting to obtain a comprehensive feature data set.

5. The big data based enterprise data asset auditing method of claim 4, wherein, The method for fusing the financial feature data set, the business operation feature data set, the management decision feature data set and the external industry feature data set after feature extraction by weighting to obtain a comprehensive feature data set comprises: The financial feature data set, the business operation feature data set, the management decision feature data set and the external industry feature data set after feature extraction are fused through a weighted model to form a comprehensive feature data set, the financial feature data set is denoted as , the business operation feature data set is denoted as , the management decision feature data set is denoted as , and the external industry feature data set is denoted as ; The weighting model is: ； wherein, is a weight coefficient for the financial feature dataset; is a weight coefficient for the business operations feature dataset; is a weight coefficient for the management decision feature dataset; is a weight coefficient for the external industry feature dataset.

6. The big data based enterprise data asset auditing method of claim 5, wherein, The method for optimizing the hyperparameters of the model comprises: S71, two hyperparameters are preset respectively as learning rate and GRU layer number , the two hyperparameter combinations are traversed using grid search, and the preset learning rate value range is: ； wherein, is a learning rate a number of values of a value range of ； wherein, the number of GRU layers the number of values; S72、for each hyperparameter combination, by the fold cross-validation error formula calculates the fold cross-validation error for the hyperparameter combination the fold cross-validation error, which is calculated by the fold cross-validation error formula the fold cross-validation error formula is: ； wherein, is the validation error of the model trained with the hyperparameter combination below, by k-fold cross-validation; is the value of the k-th learning rate; is the value of the k-th GRU layer; is the value of the k-th GRU layer; is the value of the k-th GRU layer; is the value of the k-th GRU layer; is the number of folds in the k-fold cross-validation; is a loop vector that iterates through all folds; is the sum from 1 to ; and ; and is the validation error of the model trained with the hyperparameter combination in the k-th fold; and is the validation error of the model trained with the hyperparameter combination S73, by fold restriction formula for number of folds in cross-validation restrictions, said the fold restriction formula is: ； wherein, is the total number of hyperparameter combinations; is the number of types of hyperparameters; is a floor function; S74, selecting such that The combination of hyperparameters that minimizes the cross-validation error is selected as the optimal combination of hyperparameters for the model.

7. The big data based enterprise data asset auditing method of claim 6, wherein, The method for comparing the predicted audit risk coefficient of the enterprise data asset with a preset audit risk coefficient threshold of the enterprise data asset and evaluating the audit risk level faced by the enterprise data asset comprises: S81、when the value of the first quartile, the audit risk coefficient threshold is recorded as ; when the value of the third quartile, the audit risk coefficient threshold is recorded as ; S82, according to the preset audit risk coefficient threshold, further divide the audit risk level threshold faced by the enterprise data asset; the predicted audit risk coefficient is denoted as ; S83、Compare the predicted audit risk coefficient with the preset audit risk coefficient threshold to determine the audit risk level faced by the enterprise data assets; if , it is determined that the audit risk level faced by the enterprise data assets is low risk; if , it is determined that the audit risk level faced by the enterprise data assets is medium risk; if , it is determined that the audit risk level faced by the enterprise data assets is high risk.

8. The big data based enterprise data asset auditing method of claim 7, wherein, The method for formulating a corresponding asset optimization management strategy according to the audit risk level faced by the enterprise data asset comprises: For the enterprise data asset with a low audit risk, the management strategy comprises data quality inspection, data classification, hierarchical management, setting of local data access control permission and starting of a local data audit tracking mechanism; For the enterprise data asset with a medium audit risk, the management strategy comprises implementation of joint data protection measures, strengthening of a data access control strategy, comprehensive permission management in combination with role and attribute control, establishment of a data backup and recovery mechanism, encrypted storage and transmission of data, implementation of user behavior analysis and audit, detection of abnormal data operation and generation of early warning information through an enterprise intelligent management terminal; For the enterprise data asset with a high audit risk, a comprehensive optimization management strategy is adopted, all data are comprehensively audited, all data operation behaviors are monitored, an audit report is generated through an enterprise intelligent management terminal; a DLP data leakage prevention technology is deployed, a high-strength encryption algorithm is used to encrypt data, an intercepting instruction is issued through an enterprise intelligent management terminal, and unauthorized data transmission is automatically intercepted.

9. A big data based enterprise data asset auditing system for implementing the big data based enterprise data asset auditing method of any one of claims 1 to 8, characterized in that, Comprise: The data acquisition module is configured to acquire financial information data, business operation data, management decision data and external industry data of the enterprise; The data processing module is configured to pre-process the acquired financial information data, business operation data, management decision data and external industry data to obtain a financial information data set, a business operation data set, a management decision data set and an external industry data set; The financial information data set, the business operation data set, the management decision data set and the external industry data set are subjected to feature extraction and weighted fusion to obtain a comprehensive feature data set; The risk prediction module is configured to construct an audit risk prediction model, input the comprehensive feature data set into the audit risk prediction model, and predict an audit risk coefficient of the enterprise data asset; The risk evaluation module is configured to compare the predicted audit risk coefficient of the enterprise data asset with a preset audit risk coefficient threshold of the enterprise data asset, and evaluate an audit risk level faced by the enterprise data asset. An adjustment optimization module is configured to formulate an optimized data management strategy according to the audit risk level faced by the enterprise data asset; the modules are connected through wired and / or wireless means.

Images