Financial statement abnormal data detection method

Abstract

The invention relates to the field of data processing, in particular to a financial statement abnormal data detection method. The method comprises the following steps: firstly, carrying out standardization processing on collected financial statement data, calculating a Spearman rank correlation coefficient, and generating an incidence matrix; counting all Spearman rank correlation coefficients in the incidence matrix, and dynamically setting positive and negative correlation thresholds; then, based on a Spearman rank correlation coefficient and positive and negative correlation thresholds, classifying the index relations; based on the standardized financial statement data and the index relationship, introducing a positive and negative and uncorrelated fluctuation feature extraction algorithm, and extracting fluctuation features; on the basis of the Spearman rank correlation coefficient and the fluctuation characteristics, an abnormal score is calculated in a segmentation mode; and finally, comparing the abnormal score with an abnormal threshold value, and judging whether the data is abnormal data or not. The technical problems that the influence of abnormal values on a correlation analysis result is too large, the dependency relationship between financial indexes is neglected, and fluctuation feature extraction is insufficient are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of data processing, in particular to a financial statement abnormal data detection method. BACKGROUND

[0002] In modern enterprise operation activities, financial statements, as the core document reflecting the economic status and operating results of enterprises, are widely used in investment and financing decision-making, corporate governance, risk assessment and compliance supervision. The authenticity and accuracy of financial statements are directly related to the credit of enterprises, the confidence of investors and the stability of capital markets. However, due to the complexity of financial business chain, the variety of transaction types, and the fact that some enterprises may intentionally evade supervision or fail to disclose information completely, abnormal data often lurk in financial statements. Abnormal data may be caused by operational errors, system failures, or intentional evasion of supervision. If not identified in time, it will not only affect the risk management of enterprises themselves, but also may have adverse effects on investors and the overall stability of the market. Therefore, it is of great significance to develop an efficient, intelligent and accurate financial statement abnormal data detection method. SUMMARY

[0003] The present application provides a financial statement abnormal data detection method to solve the technical problems of abnormal values affecting the correlation analysis results too much, ignoring the dependency relationship between financial indicators, and insufficient extraction of fluctuation characteristics.

[0004] The financial statement abnormal data detection method of the present application specifically includes the following technical solutions: A financial statement abnormal data detection method includes the following steps: S1. Collecting financial statement data and performing standardization processing to obtain standardized financial statement data; calculating the Spearman rank correlation coefficient based on the indicators in the standardized financial statement data, generating a correlation matrix; performing statistics on all Spearman rank correlation coefficients in the correlation matrix and setting a dynamic threshold; classifying the relationship of indicators based on the Spearman rank correlation coefficient and the dynamic threshold; S2. Based on the standardized financial statement data and the relationship of indicators, introducing positive and negative and unrelated fluctuation feature extraction algorithm to extract fluctuation characteristics, including positive correlation volatility, negative correlation volatility and unrelated volatility; S3. Based on the Spearman rank correlation coefficient and the fluctuation characteristics, calculating the abnormal score by segmentation; comparing the abnormal score with the abnormal threshold to determine whether it is abnormal data.

[0005] Preferably, the S1 specifically includes: For any two indicators in the standardized financial statement data, in the time window, the Spearman rank correlation coefficient of the two indicators is calculated to generate a correlation matrix, and each element in the correlation matrix represents the Spearman rank correlation coefficient of a pair of indicators.

[0006] Preferably, the S1 specifically includes: The mean and standard deviation of all Spearman rank correlation coefficients in the correlation matrix are calculated, and two dynamic thresholds, i.e., a positive correlation threshold and a negative correlation threshold, are set based on the mean and standard deviation.

[0007] Preferably, the S1 specifically includes: The Spearman rank correlation coefficient of the two indicators is compared with the positive correlation threshold and the negative correlation threshold, and the relationship between the indicators is classified as positive correlation, negative correlation and no correlation.

[0008] Preferably, the S2 specifically includes: In the implementation process of the positive and negative and no correlation volatility feature extraction algorithm, based on the sliding time window, the value of each indicator in the sliding time window is extracted from the standardized financial statement data, the deviation relative to the mean in the sliding time window is calculated and squared, summed and averaged, and then the square root is taken to obtain the volatility.

[0009] Preferably, the S2 specifically includes: In the implementation process of the positive and negative and no correlation volatility feature extraction algorithm, for the positively correlated indicator pair, a proportion factor and a tolerance factor are introduced, and the difference absolute value of the change amount of the standardized financial statement data of the two indicators that exceeds the tolerance factor is counted into the volatility calculation through the maximum value operation; using the indicator function, the difference value adjusted by the tolerance factor is divided by the geometric mean of the volatility of the two indicators to obtain the positively correlated volatility.

[0010] Preferably, the S2 specifically includes: In the implementation process of the positive and negative and no correlation volatility feature extraction algorithm, for the negatively correlated indicator pair, the change amount of the standardized financial statement data of one indicator is multiplied by the proportion factor and added to the change amount of the standardized financial statement data of the other indicator, and the absolute value is calculated; the tolerance factor is introduced, and through the indicator function, the negatively correlated volatility is obtained by dividing the geometric mean of the volatility of the two indicators.

[0011] Preferably, the S2 specifically includes: In the implementation process of the positive and negative and no correlation volatility feature extraction algorithm, for the no correlation indicator pair, the absolute value of the change amount of the standardized financial statement data of the indicator is calculated, and the no correlation volatility is obtained by dividing the volatility of the indicator.

[0012] Preferably, S3 specifically includes: Based on the positive correlation volatility and the negative correlation volatility, the abnormal scores of the positive correlation part and the negative correlation part are calculated respectively, the abnormal scores of the positive correlation part and the negative correlation part are added, and the abnormal score is obtained; at the same time, the unrelated volatility is directly taken as the abnormal score.

[0013] The technical scheme of the present application has the following beneficial effects: 1. By calculating the Spearman rank correlation coefficient in the time window, a symmetric correlation matrix is generated, and a dynamic threshold is set in combination with the mean and standard deviation, which can adapt to the characteristics of different financial statement data sets, avoid misjudgment caused by fixed threshold, make the index relationship determination more flexible, enhance the capture ability of the real relationship between indexes, that is, the coordinated rise of income and cost can be accurately identified, and the reverse change of assets and liabilities can also be identified, providing more stable and reliable correlation information for subsequent anomaly detection.

[0014] 2. By analyzing the positive, negative and unrelated index pairs respectively, the positive correlation volatility, the negative correlation volatility and the unrelated volatility are extracted, and hierarchical volatility feature analysis is realized, which can capture systematic anomalies such as distortion of income and cost relationship, and find independent anomalies such as sudden volatility of certain expenses, breaking through the limitation of traditional methods which only focus on single index or single correlation relationship, and improving the comprehensiveness and accuracy of anomaly detection.

[0015] 3. By calculating the index volatility rate in the sliding time window and normalizing it, the volatility characteristics of different time periods can be dynamically tracked, ensuring that the anomaly detection can reflect the recent financial situation rather than being masked by long-term volatility, which can quickly respond in the financial risk warning scene, and improve the sensitivity and decision speed of enterprises to sudden anomalies.

[0016] 4. By adjusting the volatility difference between indexes through the proportion factor, and filtering the normal fluctuation range through the tolerance factor, the abnormal judgment is more fair and accurate, avoiding the deviation caused by the difference in index dimension or volatility amplitude, preventing normal small differences from being misjudged as abnormal, reducing false positives, improving the credibility of anomaly detection, and obtaining higher application value in actual auditing or risk control. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 A flowchart of a financial statement anomaly data detection method according to the present application. DETAILED DESCRIPTION

[0018] In order to further clarify the technical means and effects taken by the present application to achieve the predetermined inventive objectives, the technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts fall within the scope of the present application.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0020] The specific scheme of the financial statement abnormal data detection method provided by the present application will be described below in conjunction with the drawings.

[0021] Referring to the drawings Figure 1 , it shows a flow chart of a financial statement abnormal data detection method provided by an embodiment of the present application, and the method comprises the following steps: S1, collecting financial statement data and performing standardization processing to obtain standardized financial statement data; based on the indicators in the standardized financial statement data, calculating the Spearman rank correlation coefficient to generate a correlation matrix; performing statistics on all Spearman rank correlation coefficients in the correlation matrix and setting a dynamic threshold; based on the Spearman rank correlation coefficient and the dynamic threshold, classifying the relationship of the indicators; Collecting financial statement data containing time dimension from the internal financial system of an enterprise, including indicators such as assets, liabilities, income, cost and cash flow, performing standardization processing on each collected financial statement data, such as Z-score standardization, to eliminate the difference in dimensions of different indicators, and obtaining standardized financial statement data; For any two indicators in the standardized financial statement data, calculating the Spearman rank correlation coefficient in a selected time window, such as the last 12 months, to generate a symmetric correlation matrix, and each element in the correlation matrix represents the Spearman rank correlation coefficient of a pair of indicators, and the calculation formula is as follows: , Wherein, represents the Spearman rank correlation coefficient of the i-th indicator and the j-th indicator in the time window T; represents the sum operation of the rank difference squares of the i-th indicator and the j-th indicator at all time points t in the time window T; ​​​​​​​represents a time window, containing a series of continuous time points, which can be set according to specific implementation scenarios; represents a time window represents a single time point in the time window represents the standardized financial statement data of the th indicator at the time point ; represents the standardized financial statement data of the th indicator at the time point ; represents the ranking of the standardized financial statement data of the th indicator at the time point , i.e., the rank of the th indicator, which can be directly sorted according to the size of the standardized financial statement data to obtain the ranking; represents the standardized financial statement data of the th indicator at the time point ; represents the ranking of the standardized financial statement data of the th indicator at the time point ;

[0022] represents the length of the time window, i.e., the number of time points.

[0023] Compared with the traditional Pearson correlation coefficient, the Spearman rank correlation coefficient is not sensitive to outliers, which is suitable for scenarios where there are outliers in the financial statement data, thereby improving the stability of the correlation matrix. , wherein, represents the positive correlation threshold; represents the mean of the Spearman rank correlation coefficients between all pairs of indicators within the time window ; represents a global sensitivity parameter, which is used to adjust the strictness of the positive and negative correlation thresholds, and is determined by expert experience method; represents the standard deviation of the Spearman rank correlation coefficients between all pairs of indicators within the time window ; represents the negative correlation threshold, which is equal to the mean minus the same adjustable parameter times the standard deviation as the positive correlation threshold, and is used to identify strong negative correlation, the calculation formula is: , wherein, represents a negative correlation threshold value; The dynamic threshold value is driven by data to adapt to the characteristics of different financial statement data, avoiding misjudgment caused by using a fixed threshold value.

[0024] If the Spearman rank correlation coefficient of a pair of indicators is greater than or equal to the positive correlation threshold value, it is considered that the two indicators have strong positive correlation, indicating that the indicators change in the same direction, such as simultaneous growth of income and cost, and the two indicators are added to the positive correlation neighbor set; if the Spearman rank correlation coefficient of a pair of indicators is less than or equal to the negative correlation threshold value, it is considered that the two indicators have strong negative correlation, indicating that the indicators change in opposite directions, such as an increase in assets accompanied by a decrease in liabilities, and the two indicators are added to the negative correlation neighbor set; if the Spearman rank correlation coefficient of a pair of indicators is between the negative correlation threshold value and the positive correlation threshold value, it is considered that the two indicators have no significant correlation, and they are added to the unrelated neighbor set, indicating that the fluctuations of the indicators are relatively independent.

[0025] By calculating the Spearman rank correlation coefficient, an association matrix is generated, the positive and negative correlation threshold values are dynamically set, the indicator relationships are classified into positive correlation, negative correlation and no correlation, the positive correlation neighbor set reflects the coordinated changes of the indicators, the negative correlation neighbor set reflects the opposite changes of the indicators, and the unrelated neighbor set can identify independent fluctuation indicators, providing dynamic relationship information between indicators for subsequent anomaly detection, and ensuring that the anomaly score can comprehensively consider the dependence and independence between indicators.

[0026] S2, based on the relationship between the standardized financial statement data and the indicators, introducing a positive and negative and unrelated fluctuation feature extraction algorithm to extract fluctuation features, including positive correlation fluctuation, negative correlation fluctuation and unrelated fluctuation; In order to extract the fluctuation features related to anomaly detection from the standardized financial statement data, a positive and negative and unrelated fluctuation feature extraction algorithm is introduced, which analyzes the positive and negative correlation relationship between indicators and the independent fluctuation characteristics of unrelated indicators, calculates the feature quantity that can reflect the anomaly degree of the indicators, including positive correlation fluctuation, negative correlation fluctuation and unrelated fluctuation; In order to quantify the fluctuation amplitude of each indicator in a specific time window, the positive and negative and unrelated fluctuation feature extraction algorithm first selects a sliding time window to ensure that the analysis can reflect the dynamic changes of the standardized financial statement data in the recent period, extracts the values in the sliding time window from the standardized financial statement data for each indicator, calculates the deviation relative to the mean value in the sliding time window and squares it, takes the average after summation, and then takes the square root to obtain the volatility rate, and the calculation formula is: , wherein, represents the Each indicator at a given time point The volatility of standardized financial statement data; This indicates the sliding time window, which can be set according to the specific implementation scenario, such as 6 months; It is the index of a time point within the sliding time window; Indicates the sliding time window All time points within The The summation of the squared deviations of the mean of the standardized financial statement data for each indicator. Indicates the first Each indicator at a given time point Standardized financial statement data; Indicates the first Each indicator in the sliding time window The mean of the standardized financial statement data within the country.

[0027] The algorithm for extracting positive, negative, and uncorrelated fluctuation features analyzes the degree of abnormality in the relative fluctuations of indicator pairs with positive and negative correlations, dividing them into positively correlated volatility and negatively correlated volatility. For indicator pairs identified as positively correlated, it focuses on whether the direction and magnitude of the changes in the two indicators conform to the expected positive correlation. Specifically, it calculates the correlation between the two indicators at different time points. The changes in standardized financial statement data are analyzed by introducing a proportional factor to ensure fairness in comparison. The absolute value of the difference between the changes in two indicators is calculated to reflect the deviation in their synchronicity. To avoid misjudging normal small deviations as abnormalities, a tolerance factor is introduced, and the maximum value is taken to ensure that only deviations exceeding the tolerance factor are included in volatility calculations. An indicator function ensures that the two indicators change in the same direction, i.e., increasing or decreasing together; otherwise, volatility is zero. After adjusting for the tolerance factor, the difference is normalized by dividing by the geometric mean of the volatility of the two indicators to obtain the positively correlated volatility. The calculation formula is as follows: , in, Indicators indicating positive correlation At the point of time Positive correlation volatility; Indicates the first Each indicator at a given time point The change in financial statement data after standardization. ; Indicates the first Each indicator at a given time point Standardized financial statement data; Represents the scaling factor, used to adjust the first... The first indicator and the first To ensure fair comparison, the fluctuation range differences between standardized financial statement data for each indicator were determined through expert experience. Indicates the first Each indicator at a given time point The change in financial statement data after standardization. ; Indicates the first Each indicator at a given time point Standardized financial statement data; The tolerance factor is defined as follows: ; This indicates a maximum value operation, ensuring that volatility is only included in the calculation when the deviation exceeds the tolerance factor; Indicates an indicator function, when That is, the first The first indicator and the first When all indicators change in the same direction, the value is 1; otherwise, the value is 0. Indicates the first Each indicator at a given time point The volatility of standardized financial statement data; This represents the geometric mean of the volatility of the two indicators, used to normalize positively correlated volatility.

[0028] The algorithm for extracting positive, negative, and uncorrelated fluctuation features analyzes whether negatively correlated indicator pairs exhibit the expected inverse changes. The calculation process is similar to that for positive correlation, but the key difference lies in the handling of the change amount: an indicator at a given time point... The change in standardized financial statement data, multiplied by a scaling factor, is compared with another indicator at a given point in time. The changes in the standardized financial statement data are summed, and the absolute value is calculated to reflect the deviation of the inverse relationship. A tolerance factor is also introduced to eliminate deviations within the normal range. An indicator function is used to ensure that the two indicators change in opposite directions, i.e., one increases and the other decreases. Dividing by the geometric mean of volatility yields the negative correlation volatility, calculated using the following formula: , in, Indicators of negative correlation At the point of time Negative correlation volatility; Indicates an indicator function, when That is, the first The first indicator and the first When the changes of an indicator are in opposite directions, the value is 1; otherwise, the value is 0. The algorithm for extracting positive and negative, and unrelated fluctuation features independently analyzes the fluctuation intensity of unrelated indicator pairs and calculates the indicator at a given time point. The change of the standardized financial statement data is taken as the absolute value to measure the fluctuation range, and divided by the volatility of the index at the time point The normalized uncorrelated volatility is suitable for detecting abnormal changes that are not constrained by other indicators, and the calculation formula is: , Wherein, represents the uncorrelated volatility of the standardized financial statement data of the th index at the time point .

[0029] By extracting the fluctuation characteristics in layers, namely positive correlation volatility, negative correlation volatility and uncorrelated volatility, the problem of insufficient analysis of positive and negative correlation and uncorrelated indicators in traditional methods is solved. In the calculation of positive and negative correlation volatility, the expected relationship between indicators, i.e. positive synchronization or reverse change, is used to detect deviation, which is suitable for finding systematic anomalies, such as abnormal synchronization of income and cost. In the calculation of uncorrelated volatility, the independent change of the index is focused on, which is suitable for detecting anomalies independent of other indicators, such as sudden fluctuations in advertising expenses.

[0030] S3, based on the Spearman rank correlation coefficient and the fluctuation characteristics, the abnormal score is calculated by segmenting; compare the abnormal score with the abnormal threshold to determine whether it is abnormal data.

[0031] In order to quantify the abnormal degree of each index in the standardized financial statement data at a specific time point, the fluctuation difference of positive and negative correlation indicators and the independent fluctuation characteristics of uncorrelated indicators are considered comprehensively by segmenting, aiming to capture complex abnormal patterns.

[0032] If the index has positive and negative correlation neighbors, for positive correlation neighbors, multiply the positive correlation volatility by the positive correlation weight to reflect the influence of the positive correlation strength between the index pairs, and sum up the contributions of all positive correlation neighbors to get the abnormal score of the positive correlation part; for negative correlation neighbors, multiply the negative correlation volatility by the negative correlation weight to reflect the strength of the negative correlation between the index pairs, and sum up the contributions of all negative correlation neighbors to get the abnormal score of the negative correlation part; add the abnormal scores of the positive and negative correlation parts to get the total abnormal score of the index; For uncorrelated indicators, the uncorrelated volatility is directly used as the abnormal score; The calculation formula of the abnormal score is: , Wherein, represents the abnormal score of the standardized financial statement data of the th index at the time point . represents the abnormal score of the standardized financial statement data of the The abnormal score is based on the volatility of the positive and negative correlation indicators, while considering the correlation weight; An index representing the neighbor indicator of the first An index representing the neighbor indicator of the first The sum of the weighted fluctuation deviations of the normalized financial statement data of the first The sum of the weighted fluctuation deviations of the normalized financial statement data of the first The sum of the weighted fluctuation deviations of the normalized financial statement data of the first The sum of all neighbor indicators in the positive correlation neighbor set of the first The sum of all neighbor indicators in the positive correlation neighbor set of the first The sum of all neighbor indicators in the positive correlation neighbor set of the first The sum of all neighbor indicators in the positive correlation neighbor set of the first The positive correlation weight, equal to the Spearman rank correlation coefficient of the first The positive correlation weight, equal to the Spearman rank correlation coefficient of the first The sum of the weighted fluctuation deviations of the normalized financial statement data of the first The sum of the weighted fluctuation deviations of the normalized financial statement data of the first The sum of the weighted fluctuation deviations of the normalized financial statement data of the first The sum of the weighted fluctuation deviations of the normalized financial statement data of the first The sum of all neighbor indicators in the negative correlation neighbor set of the first The sum of all neighbor indicators in the negative correlation neighbor set of the first The sum of all neighbor indicators in the negative correlation neighbor set of the first The sum of all neighbor indicators in the negative correlation neighbor set of the first The sum of all neighbor indicators in the negative correlation neighbor set of the first The negative correlation weight, equal to the negative value of the Spearman rank correlation coefficient of the first The negative correlation weight, equal to the negative value of the Spearman rank correlation coefficient of the first The sum of the weighted fluctuation deviations of the normalized financial statement data of the first The sum of the weighted fluctuation deviations of the normalized financial statement data of the first When the first When the first

[0033] The abnormal threshold is determined according to expert experience method, and the abnormal score is compared with the abnormal threshold to determine whether it is abnormal data; if the abnormal score is greater than or equal to the abnormal threshold, it is abnormal data, otherwise it is normal data.

[0034] Through the above segmented scoring mechanism, the synchronous or reverse fluctuation deviation of the positive and negative correlation indicators and the independent abnormal fluctuation of the irrelevant indicators are captured, overcoming the neglect of traditional methods such as Z-score and IQR for complex correlation and irrelevant indicators, enhancing objectivity, suitable for financial statement analysis, audit compliance and risk warning, providing reliable abnormal detection tools for enterprises, reducing financial risk and improving decision-making efficiency. In summary, a financial statement abnormal data detection method is completed.

[0035] The progressive nature of the discussion in this specification should not be construed as a requirement that the application be practiced in a progressive or sequential manner. Although each embodiment described in this specification can stand on its own, the embodiments can be utilized in various combinations.

[0036] Each of the various embodiments described in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. Each embodiment can be combined with other embodiments.

[0037] The above embodiments are only used to illustrate the technical solutions of the present application, but not limit the present application; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions recorded in the foregoing embodiments can be modified, or some technical features can be replaced by equivalent features; and these modifications or replacements do not make the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A method of detecting abnormal data in a financial statement, characterized by, The method comprises the following steps: S1. Collecting financial statement data and performing standardization processing to obtain standardized financial statement data; calculating Spearman rank correlation coefficients based on the indicators in the standardized financial statement data, generating a correlation matrix; performing statistics on all Spearman rank correlation coefficients in the correlation matrix and setting a dynamic threshold; classifying the relationships of the indicators based on the Spearman rank correlation coefficients and the dynamic threshold; S2. Based on the standardized financial statement data and the relationships of the indicators, introducing a positive and negative and unrelated fluctuation feature extraction algorithm to extract fluctuation features, including positive correlation volatility, negative correlation volatility and unrelated volatility; S3. Based on the Spearman rank correlation coefficients and the fluctuation features, calculating abnormal scores by segmenting; comparing the abnormal scores with an abnormal threshold to determine whether the data is abnormal.

2. The method of claim 1, wherein, The S1 specifically comprises: For any two indicators in the standardized financial statement data, the Spearman rank correlation coefficient of the two indicators is calculated in a time window to generate a correlation matrix, and each element in the correlation matrix represents the Spearman rank correlation coefficient of a pair of indicators.

3. The method of claim 2, wherein, The S1 specifically comprises: The statistics of all Spearman rank correlation coefficients in the correlation matrix are calculated to calculate the mean and standard deviation; based on the mean and standard deviation, two dynamic thresholds, i.e. positive correlation threshold and negative correlation threshold, are set.

4. The method of claim 3, wherein, The S1 specifically comprises: The Spearman rank correlation coefficients of the two indicators are compared with the positive correlation threshold and the negative correlation threshold, and the relationships of the indicators are classified as positive correlation, negative correlation and unrelated.

5. The method of claim 1, wherein, The S2 specifically comprises: In the implementation process of the positive and negative and unrelated fluctuation feature extraction algorithm, based on a sliding time window, the values of each indicator in the sliding time window are extracted from the standardized financial statement data, the deviation from the mean in the sliding time window is calculated and squared, the sum is averaged, and the square root is taken to obtain the volatility.

6. The method of claim 5, wherein, The S2 specifically comprises: In the implementation process of the positive and negative and unrelated fluctuation feature extraction algorithm, for the positively correlated indicator pair, a proportion factor and a tolerance factor are introduced, the difference between the change amounts of the two indicators that exceed the tolerance factor is counted into the volatility calculation through the maximum value operation; the difference adjusted by the tolerance factor is normalized by dividing the geometric mean of the volatility rates of the two indicators using an indicator function to obtain the positive correlation volatility.

7. The method of claim 5, wherein the step of identifying the abnormal data in the financial statement comprises the steps of: The S2 specifically comprises: ​ In the implementation process of the positive and negative and unrelated fluctuation feature extraction algorithm, for the negatively correlated indicator pair, the change amount of one indicator is multiplied by the proportion factor and added to the change amount of the other indicator to calculate the absolute value; a tolerance factor is introduced, and the negative correlation volatility is obtained by dividing the geometric mean of the volatility rates of the two indicators through an indicator function.

8. The method of claim 5, wherein the step of identifying the abnormal data in the financial statement comprises the steps of: The S2 specifically comprises: ​ In the implementation process of the positive and negative and unrelated fluctuation feature extraction algorithm, for the unrelated indicator pair, the absolute value of the change amount of the standardized financial statement data of the indicators is calculated and divided by the volatility rate of the indicators to obtain the unrelated volatility.

9. The method of claim 1, wherein, The S3 specifically comprises: Based on the positive correlation volatility and the negative correlation volatility, the abnormal scores of the positive correlation part and the negative correlation part are calculated respectively, the abnormal scores of the positive correlation part and the negative correlation part are added, and the abnormal score is obtained; meanwhile, the irrelevant volatility is directly taken as the abnormal score.

Images