AI intervention correction method in case of financial data input errors

Abstract

The invention discloses an AI intervention correction method for financial data input errors, and relates to the technical field of data management, the input financial data is formatted, and the abnormal level of the financial data is preset in combination with data abnormal factors; evaluating the data anomaly degree based on the interference factors that the abnormal value, the missing value and the data type of the financial data are not matched; and in combination with the comprehensive evaluation result of the data anomaly degree, setting different error levels, and setting correction thresholds for the different error levels. According to the AI intervention correction method for the financial data input errors, by detecting and correcting the errors in the financial data, the accuracy and reliability of the data are improved, an AI system can rapidly recognize the problem that abnormal values, missing values and data types are not matched, correction is automatically conducted according to preset rules and algorithms, and the accuracy and reliability of the data are improved. According to the method, errors caused by human negligence and fatigue are reduced, the accuracy of the financial report is ensured, and meanwhile, the data processing efficiency is greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data management, and in particular to an AI intervention correction method for financial data entry errors. Background Art

[0002] Financial data is an important basis for business operations and decision-making. The accuracy of the data directly affects the company's financial reporting, cost control, budget management, investment analysis and other aspects. In the process of entering financial data into the cloud financial software system, due to human negligence, fatigue, inaccurate understanding of the rules and other reasons, entry errors often occur. Errors may lead to serious consequences such as inconsistent data, report errors, and wrong decisions. Therefore, effective measures need to be taken to reduce and correct existing errors. In recent years, with the rapid development of artificial intelligence technology, AI's capabilities in data processing, analysis, and recognition have been significantly improved. AI technology can be used to realize automatic processing and correction of data.

[0003] In the prior art, due to the deviations in the entered financial data, over-correction and under-correction may occur when AI technology intervenes and corrects the entered financial data. Over-correction may cause data distortion or loss of original meaning, while under-correction may cause errors to still exist or not be completely eliminated. Therefore, how to evaluate the error degree of financial data to improve the accuracy of financial data intervention correction is the problem we need to solve. To this end, an AI intervention correction method for financial data entry errors is proposed. Summary of the invention

[0004] To achieve the above objectives, the present invention is implemented through the following technical solutions: an AI intervention correction method for financial data entry errors, comprising the following steps:

[0005] Step 1: Format the entered financial data and preset the abnormal level of the financial data based on the data abnormality factors;

[0006] Step 2: Evaluate the degree of data anomaly based on the interference factors of outliers, missing values ​​and data type mismatch in financial data;

[0007] Step 3: Combined with the comprehensive evaluation results of the degree of data anomaly, different error levels are set, and correction thresholds are set for different error levels. The correction thresholds are used to determine when to apply correction, the degree of correction, and whether to retain certain features of the original data;

[0008] Step 4: Based on the determined error level, categorize the erroneous financial data, clarify the error handling measures for the financial data, and determine whether the AI ​​system automatically intervenes to correct or manually intervene;

[0009] Step 5: Implement AI intervention and correction strategies for the classified erroneous financial data, and determine whether errors that require manual intervention should be queued for manual intervention and correction;

[0010] Step 6: Monitor the quality of the corrected data, collect user feedback and performance data from actual applications, and optimize the financial data correction process based on the feedback and performance data to improve the accuracy and efficiency of the correction.

[0011] Preferably, in step 1, the process of presetting the abnormal level of financial data includes:

[0012] Identify the data sources for entering financial data, including cloud financial software systems and databases, and export financial data from various data sources to ensure data integrity and consistency. The cloud financial software system is used for daily financial management and account records, and databases include SQL Server and Oracle, which are used to store and manage large amounts of structured data;

[0013] For the entered financial data, invalid or redundant information is removed, and then formatted to convert the time series data into a stationary series. The principal component analysis method is used to reduce the dimension of the data, thereby reducing the complexity and size of the data while maintaining the important features of the data.

[0014] Based on the error causes of financial data entry, including entry errors, calculation errors and system errors, the severity of each error cause and the degree of impact on the financial data are analyzed, and different exception levels are preset, namely minor exceptions, medium exceptions and severe exceptions.

[0015] Preferably, in step 2, the process of evaluating the degree of data anomaly includes:

[0016] Clarify the objectives and factors of financial data evaluation. The evaluation objective is to evaluate the degree of anomalies in financial data so as to timely discover and correct anomalies and improve data quality. Evaluation factors include outliers, missing values, and data type mismatch. Outliers are data points that deviate significantly from the normal data range. Missing values ​​are data that are missing or unrecorded in the data set. Data type mismatch refers to the inconsistency between the actual type of a data field and the expected type.

[0017] Use the Z-score statistical method to calculate the degree of outliers. The Z-score represents the number of standard deviations between a data point and the mean of a data set. Different levels of outliers are classified according to the degree of outliers. Analyze the impact of missing values ​​on data analysis results. Determine different levels of outliers based on the number and proportion of missing values. Check whether the type of data field is consistent with expectations. Different levels of outliers are classified according to the degree of data type mismatch and the impact on data analysis results.

[0018] Analyze the evaluation results of outliers, missing values, and data type mismatches, determine the type of anomaly for each data point, and match different anomaly levels according to the error degrees of outliers, missing values, and data type mismatches.

[0019] Preferably, for the evaluation of the outliers, set the threshold of the outlier degree as Z > 3 to divide the data points into outliers. Among them, the severe anomaly threshold of the outlier is Z > 5, which is expressed as an extreme outlier; the medium anomaly threshold of the outlier is 4 < Z ≤ 5, which is expressed as a relatively mild outlier; the slight anomaly threshold of the outlier is 3 < Z ≤ 4, which is expressed as an outlier with a low interference degree.

[0020] For the evaluation of the missing values, set the threshold for dividing the influence degree of the missing values. Among them, the severe anomaly threshold of the missing value is that the proportion of the missing values in the data field exceeds 10%; the medium anomaly threshold of the missing value is that the proportion of the missing values is between 5% and 10%; the slight anomaly threshold of the missing value is that the proportion of the missing values is less than 5%.

[0021] For the evaluation of the data type mismatches, set the threshold for the degree of data type mismatches. Among them, the severe anomaly threshold of the data type mismatch is that the proportion of the mismatched records in the total records exceeds 20%; the medium anomaly threshold of the data type mismatch is that the proportion of the mismatched records in the total records is between 5% - 20%; the slight anomaly threshold of the data type mismatch is that the proportion of the mismatched records in the total records is less than 5%.

[0022] Preferably, in the third step, the process of setting the correction threshold includes:

[0023] Integrate the relevant data of outliers, missing values, and data type mismatches, compare the outliers, missing values, and data type mismatches with their respective thresholds, and obtain the outlier comparison difference value, missing value comparison difference value, and mismatch comparison difference value.

[0024] Based on the evaluation results of outliers, missing values, and data type mismatches, assign different weights to different types of anomalies.

[0025] Combine the respective comparison difference values of outliers, missing values, and data type mismatches, and comprehensively calculate the error evaluation coefficient to quantify the severity of the data error.

[0026] Combine the value of the error evaluation coefficient, set different error levels, namely low error level, medium error level, and high error level, and match the corresponding correction thresholds for each error level.

[0027] Preferably, the calculation formula of the error evaluation coefficient is:

[0028]

[0029] Where E is the error estimation coefficient, λ is the outlier comparison difference value, Z is the Z score, ω A is the weight of the outlier comparison difference value, M is the missing value comparison difference value, M t is the total number of missing values, M max is the maximum number of records, ω M is the weight of missing value comparison difference value, D is the weight of mismatch comparison difference value, D t is the total number of data type mismatches, D max is the total number of fields, ω D is the weight of the mismatched alignment difference value;

[0030] The multiple error levels correspond to the multiple correction thresholds, specifically satisfying the following relationship:

[0031] Low error level: 0<E≤Watts;

[0032] Medium error level: E L <E≤E M ;

[0033] High error level: E M <E<1;

[0034] Among them, E is the error assessment coefficient, E L is the upper threshold corresponding to the low error level and the lower threshold corresponding to the medium error level, E M is the upper threshold corresponding to the medium error level and the lower threshold corresponding to the high error level, E L =0.33, E M =0.66.

[0035] Preferably, in step 4, the process of classifying the erroneous financial data includes:

[0036] Specific error handling measures are formulated for each error level according to the different error levels set. Among them, the error level of low error level is relatively minor and has little impact on the overall financial data. The error level of medium error level is moderate and needs to be paid attention to and measures to correct it. The error level of high error level is serious and has a greater impact on the overall financial data and needs to be corrected immediately. For low error levels, automatic corrections are made through data cleaning tools, including removing duplicate values, filling missing values, etc., and regular inspections are carried out to ensure that the problem does not worsen. For medium error levels, manual review is required to confirm the cause of the error. Corrective measures are taken according to the error type, including adjusting data entry rules and optimizing data processing procedures, etc. For high error levels, manual intervention is carried out to analyze the cause of the error in detail, and emergency corrective measures are taken, including re-entering data and adjusting system parameters, etc., and related business processes are comprehensively reviewed and optimized to prevent similar errors from happening again;

[0037] Determine the specific error level of the entered financial data based on the error assessment coefficient and the different error levels set;

[0038] Based on the type and severity of the error, determine whether the AI ​​system automatically intervenes to correct it or manually intervenes, and record detailed information on each error handling, including the error type, error level, and measures taken.

[0039] Preferably, the process of determining whether the correction is automatically intervened by the AI ​​system or manually intervened specifically includes:

[0040] Use AI systems to monitor and detect input financial data in real time to identify potential errors and classify them into different categories based on their specific types and characteristics;

[0041] Combined with the recognition results of the error level, the correction method is determined and analyzed to see whether manual intervention is needed;

[0042] For errors with low error levels that are easy to correct automatically, the AI ​​system automatically intervenes and makes judgments, and simple data format adjustments or missing value fillings are performed by the AI ​​system. For medium error levels, the AI ​​system is used to mark data points for review and send them for manual review. For high error levels, the manual intervention process is automatically triggered.

[0043] Preferably, in step 5, the process of determining whether errors requiring manual intervention are queued includes:

[0044] Develop AI system automatic correction strategies based on error types and historical data, including data cleaning rules, missing value filling algorithms, and outlier detection and adjustment;

[0045] Input the classified erroneous data into the AI ​​system, automatically correct the erroneous data according to the preset strategy, and record the AI ​​correction process and results, including the data comparison before and after correction and the correction time, etc.;

[0046] For erroneous data that the AI ​​system cannot automatically correct or the correction effect is unsatisfactory, manual intervention is performed to mark the data. The marking information includes the error type, error level and recommended correction measures;

[0047] Analyze the error level, and determine whether queuing is needed based on the priority and urgency of the error. For data with medium error levels, queue them by time and wait for manual intervention and correction. For data with high error levels, send them directly for manual intervention and correction. By detecting and correcting errors in financial data, the accuracy and reliability of the data are improved. The AI ​​system can quickly identify outliers, missing values, and data type mismatches, and automatically correct them according to preset rules and algorithms, reducing errors caused by human negligence and fatigue, ensuring the accuracy of financial reports, and greatly improving the efficiency of data processing.

[0048] Preferably, in step 6, the process of optimizing the financial data correction process includes:

[0049] Use automated monitoring tools to monitor the quality of the corrected data, including completeness, accuracy, consistency, uniqueness, and timeliness, and record the results of each inspection, including the items that passed and failed, generate data quality reports, and report the monitoring status to relevant personnel;

[0050] Set up a user feedback mailbox and online form as feedback channels for users to submit feedback, and regularly collect feedback submitted by users, including data errors, system performance issues, and user interface suggestions, etc., and classify and organize the feedback to form a feedback list;

[0051] Use system logs, monitoring tools or performance counters to collect performance data of the financial data correction process, including processing time, resource utilization, and error rate, and analyze the collected performance data to identify performance bottlenecks and potential problems, compare performance data in different time periods, and analyze performance change trends;

[0052] By combining data quality monitoring results, user feedback and performance data analysis, we can identify problems in the financial data correction process, analyze the problems in detail, locate the root causes of the problems, and then formulate specific optimization measures for the identified problems. According to the formulated optimization measures, we can improve and optimize the financial data correction process. By automating the processing of large amounts of data, we can quickly identify and correct errors, greatly shortening the data correction time cycle. This not only improves work efficiency, but also reduces dependence on human resources, saving manpower and time costs. In addition, the AI ​​system can instantly detect and correct errors in the entered financial data, avoiding repeated submission or review processes due to data errors.

[0053] The present invention provides an AI intervention correction method for financial data entry errors. It has the following beneficial effects:

[0054] 1. The AI ​​intervention correction method for financial data entry errors improves the accuracy and reliability of the data by detecting and correcting errors in the financial data. The AI ​​system can quickly identify outliers, missing values, and data type mismatches, and automatically correct them according to preset rules and algorithms, reducing errors caused by human negligence and fatigue, ensuring the accuracy of financial reports, and at the same time greatly improving the efficiency of data processing.

[0055] 2. The AI ​​intervention correction method for financial data entry errors can quickly identify and correct errors by automatically processing large amounts of data, greatly shortening the time period for data correction. It not only improves work efficiency, but also reduces dependence on human resources, saving manpower and time costs. The AI ​​system can instantly detect and correct errors in entered financial data, avoiding repeated submission or review processes due to data errors. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 A method flow chart of an AI intervention correction method for financial data entry errors according to the present invention;

[0057] Figure 2 Flowchart for setting correction thresholds for the present invention. DETAILED DESCRIPTION

[0058] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. The embodiments of the present invention are provided for the purpose of illustration and description, and are not intended to be exhaustive or to limit the present invention to the disclosed forms. Many modifications and variations will be apparent to those of ordinary skill in the art. The embodiments are selected and described in order to better illustrate the principles and practical applications of the present invention, and to enable those of ordinary skill in the art to understand the present invention and thereby design various embodiments with various modifications suitable for specific uses.

[0059] The first embodiment, as Figure 1 As shown, the present invention provides a technical solution: an AI intervention correction method for financial data entry errors, comprising the following steps:

[0060] Step 1: Format the entered financial data, and combine the data abnormality factors to preset the abnormal level of financial data, clarify the data source of the entered financial data, including the cloud financial software system and database, and export the financial data from each data source to ensure the integrity and consistency of the data. The cloud financial software system is used for daily financial management and account records, and the database includes SQL Server and Oracle, etc., are used to store and manage large amounts of structured data. For the entered financial data, invalid or redundant information is removed, and then formatted, time series data is converted into stationary series, and the principal component analysis method is used to reduce the dimension of the data. Under the premise of maintaining the important characteristics of the data, the complexity and scale of the data are reduced. Combined with the error causes of the entered financial data, including entry errors, calculation errors and system errors, the severity of each error cause and the degree of impact on the financial data are analyzed, and different abnormality levels are preset, namely minor abnormality, medium abnormality and severe abnormality. Among them, entry error refers to inaccurate data entry due to human negligence, calculation error refers to errors occurring during data processing or calculation, and system error refers to data errors caused by software or hardware failure. Minor abnormality refers to abnormality with little impact on financial data, such as decimal point position error or minor format inconsistency, medium abnormality refers to abnormality with certain impact on financial data, such as numerical range error or missing data field, and severe abnormality refers to abnormality with significant impact on financial data, such as complete data missing, abnormal value or severe format error.

[0061] Step 2: Evaluate the degree of data anomalies based on outliers, missing values, and interference factors of data type mismatches in financial data, and clarify the objectives and factors of financial data evaluation. Among them, the evaluation objective is to evaluate the degree of anomalies in financial data to promptly detect and correct anomalies and improve data quality. The evaluation factors include outliers, missing values, and data type mismatches. An outlier is a data point that significantly deviates from the normal data range. A missing value is data that is missing or not recorded in the dataset. A data type mismatch is when the actual type of a data field is inconsistent with the expected type. Use the statistical method of Z-score to calculate the degree of outliers. The Z-score represents the number of standard deviations between a data point and the mean of the dataset. Classify it into different anomaly levels according to the degree of outliers. Analyze the impact degree of missing values on the data analysis results, and determine different anomaly levels in combination with the quantity and proportion of missing values. Check whether the type of the data field is consistent with the expectation, and classify it into different anomaly levels according to the degree of data type mismatch and its impact on the data analysis results. Analyze the evaluation results of outliers, missing values, and data type mismatches, determine the anomaly type of each data point, and match different anomaly levels according to the error degrees of outliers, missing values, and data type mismatches;

[0062] Further, for the evaluation of outliers, set the threshold of the degree of outliers as Z > 3 to classify data points as outliers. Among them, the severe anomaly threshold of outliers is Z > 5, which is expressed as an extreme outlier. The moderate anomaly threshold of outliers is 4 < Z ≤ 5, which is expressed as a relatively mild outlier. The mild anomaly threshold of outliers is 3 < Z ≤ 4, which is expressed as an outlier with a low interference degree;

[0063] For the evaluation of missing values, set the threshold for classifying the impact degree of missing values. Among them, the severe anomaly threshold of missing values is that the proportion of missing values in the data field exceeds 10%. The moderate anomaly threshold of missing values is that the proportion of missing values is between 5% and 10%. The mild anomaly threshold of missing values is that the proportion of missing values is less than 5%;

[0064] For the evaluation of data type mismatches, set the threshold of the degree of data type mismatches. Among them, the severe anomaly threshold of data type mismatches is that the proportion of mismatched records in the total number of records exceeds 20%. The moderate anomaly threshold of data type mismatches is that the proportion of mismatched records in the total number of records is between 5% - 20%. The mild anomaly threshold of data type mismatches is that the proportion of mismatched records in the total number of records is less than 5%;

[0065] Step 3: Combine the comprehensive evaluation results of the degree of data anomalies, set different error levels, and set correction thresholds for different error levels. Determine when to apply correction, the degree of correction, and whether to retain certain characteristics of the original data through the correction threshold;

[0066] Step 4: Based on the determined error level, categorize the erroneous financial data, clarify the error handling measures for the financial data, and determine whether the AI ​​system automatically intervenes to correct or manually intervene;

[0067] Step 5: Implement AI intervention and correction strategies for the classified erroneous financial data, and determine whether errors that require manual intervention should be queued for manual intervention and correction;

[0068] Step 6: Monitor the quality of corrected data, collect user feedback and performance data in actual applications, and optimize the financial data correction process based on feedback and performance data to improve the accuracy and efficiency of correction. By detecting and correcting errors in financial data, the accuracy and reliability of the data are improved. The AI ​​system can quickly identify outliers, missing values, and data type mismatches, and automatically correct them according to preset rules and algorithms, reducing errors caused by human negligence and fatigue, ensuring the accuracy of financial reports, and greatly improving the efficiency of data processing.

[0069] The second embodiment is based on the first embodiment. Figure 2 As shown, in step 3, the process of setting the correction threshold includes:

[0070] Integrate the relevant data of outliers, missing values ​​and data type mismatches, and compare the outliers, missing values ​​and data type mismatches with their respective thresholds to obtain outlier comparison difference values, missing value comparison difference values ​​and mismatch comparison difference values. Based on the evaluation results of outliers, missing values ​​and data type mismatches, assign different weights to different types of anomalies. Combine the comparison difference values ​​of outliers, missing values ​​and data type mismatches to comprehensively calculate the error assessment coefficient to quantify the severity of data errors. Combine the value of the error assessment coefficient to set different error levels, namely low error level, medium error level and high error level, and match the corresponding correction threshold for each error level.

[0071] Furthermore, the calculation formula of the error assessment coefficient is:

[0072]

[0073] Where E is the error estimation coefficient, λ is the outlier comparison difference value, Z is the Z score, ω A is the weight of the outlier comparison difference value, M is the missing value comparison difference value, M t is the total number of missing values, M max is the maximum number of records, ω M is the weight of missing value comparison difference value, D is the weight of mismatch comparison difference value, Dt is the total number of data type mismatches, D max is the total number of fields, ω D is the weight of the mismatched alignment difference value. The error assessment coefficient E ranges from 0 to 1, where 0 indicates no error and 1 indicates the highest degree of error. As λ, M, or D increases, the value of E also increases, indicating that the severity of data errors increases;

[0074] Multiple error levels correspond to multiple correction thresholds, specifically satisfying the following relationship:

[0075] Low error level: 0<E≤E L ;

[0076] Medium error level: E L <E≤E M ;

[0077] High error level: E M <E<1;

[0078] Among them, E is the error assessment coefficient, E L is the upper threshold corresponding to the low error level and the lower threshold corresponding to the medium error level, E M is the upper threshold corresponding to the medium error level and the lower threshold corresponding to the high error level, E L =0.33, E M =0.66;

[0079] In step 4, the process of classifying erroneous financial data includes:

[0080] Specific error handling measures are formulated for each error level according to the different error levels set. Among them, the error degree of low error level is relatively light and has little impact on the overall financial data. The error degree of medium error level is moderate and needs to be paid attention to and measures to correct it. The error degree of high error level is serious and has a greater impact on the overall financial data and needs to be corrected immediately. For low error levels, automatic corrections are made through data cleaning tools, including removing duplicate values, filling missing values, etc. Regular inspections are carried out to ensure that the problem does not worsen. For medium error levels, manual review is required to confirm the cause of the error. Corrective measures are taken according to the error type, including adjusting data entry rules and optimizing data processing procedures. For high error levels, manual intervention is carried out, the cause of the error is analyzed in detail, and emergency corrective measures are taken, including re-entering data and adjusting system parameters. Relevant business processes are comprehensively reviewed and optimized to prevent similar errors from happening again. According to the error assessment coefficient and the different error levels set, the specific error level of the entered financial data is determined. According to the error type and severity, it is determined whether the AI ​​system automatically intervenes to correct or manually intervenes. The detailed information of each error handling is recorded, including the error type, error level and measures taken. The AI ​​system is used to monitor and detect the entered financial data in real time to identify potential errors, and classify them into different categories according to the specific type and characteristics of the errors. Combined with the identification results of the error level, the correction method is determined to analyze whether manual intervention is required. For errors with low error levels that are easy to correct automatically, the AI ​​system automatically intervenes and judges, and the AI ​​system automatically intervenes and corrects simple data format adjustments or missing value filling. For medium error levels, the AI ​​system is used to mark the data points to be reviewed and send them for manual review. For high error levels, the manual intervention process is automatically triggered;

[0081] In step 5, the process of determining whether errors requiring manual intervention are queued includes:

[0082] According to the error type and historical data, formulate the AI ​​system's automatic correction strategy, including data cleaning rules, missing value filling algorithm, and outlier detection and adjustment, etc., input the classified erroneous data into the AI ​​system, automatically correct the erroneous data according to the preset strategy, and record the AI ​​correction process and results, including data comparison before and after correction and correction time, etc. For erroneous data that the AI ​​system cannot automatically correct or the correction effect is not ideal, perform manual intervention marking. The marking information includes error type, error level and recommended correction measures. Analyze the error level, and determine whether to queue up based on the priority and urgency of the error. For data with medium error levels, queue them according to time to wait for manual intervention correction. For data with high error levels, send them directly for manual intervention correction. By automatically processing a large amount of data, quickly identify and correct errors, and greatly shorten the time cycle for data correction, it not only improves work efficiency, but also reduces dependence on human resources, saving manpower and time costs. In addition, the AI ​​system can instantly detect and correct errors in the entered financial data, avoiding repeated submission or review processes caused by data errors;

[0083] In step 6, the process of optimizing the financial data correction process includes:

[0084] Use automated monitoring tools to monitor the data quality of the corrected data, including completeness, accuracy, consistency, uniqueness and timeliness, and record the results of each inspection, including passed and failed items, generate data quality reports, report monitoring status to relevant personnel, set up user feedback mailboxes and online forms as feedback channels for users to submit feedback, and regularly collect feedback submitted by users, including data errors, system performance issues and user interface suggestions, etc., classify and organize feedback to form a feedback list, use system logs, monitoring tools or performance counters to collect performance data of the financial data correction process. The collected data includes processing time, resource utilization and error rate, etc., and analyze the collected performance data to identify performance bottlenecks and potential problems, compare performance data in different time periods, and analyze performance trends. Combine data quality monitoring results, user feedback and performance data analysis to identify problems in the financial data correction process, analyze the problems in detail, locate the root causes of the problems, and then formulate specific optimization measures for the identified problems. According to the formulated optimization measures, improve and optimize the financial data correction process.

[0085] Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field and related fields without creative work should fall within the scope of protection of the present invention. The structures, devices and operating methods not specifically described and explained in the present invention are implemented according to the conventional means in the field unless otherwise specified and limited.

Claims

1. An AI intervention correction method for financial data entry errors, characterized in that: It includes the following steps: Step 1: Format the input financial data, and preset the abnormal levels of the financial data in combination with data abnormal factors. Step 2: Evaluate the data abnormality degree based on the abnormal values, missing values and interference factors of data type mismatch in the financial data. Step 3: Set different error levels in combination with the comprehensive evaluation results of the data abnormality degree, and set correction thresholds for different error levels. Step 4: Classify the incorrect financial data based on the determined error levels, clarify the error handling measures for the financial data, and judge whether to perform automatic intervention correction by the AI system or manual intervention. Step 5: Implement the corresponding correction strategy of AI intervention correction for the classified incorrect financial data, and determine whether to queue the errors that require manual intervention for manual intervention correction. Step 6: Monitor the data quality after correction, collect user feedback and performance data in actual applications, and optimize the financial data correction process according to the feedback and performance data.

2. The AI ​​intervention correction method for financial data entry errors according to claim 1, characterized in that: In the above Step 1, the process of presetting the abnormal levels of the financial data includes: Clarify the data sources of the input financial data, including cloud financial software systems and databases, and export the financial data from each data source. For the input financial data, remove invalid or redundant information, and then perform formatting processing to convert time series data into stationary series. Combined with the error causes existing in the input financial data, including input errors, calculation errors and system errors, analyze the severity of each error cause and the impact on the financial data, and preset different abnormal levels, namely slight abnormality, medium abnormality and severe abnormality.

3. The AI ​​intervention correction method for financial data entry errors according to claim 2 is characterized by: In the above Step 2, the process of evaluating the data abnormality degree includes: Clarify the evaluation objectives and factors of the financial data. Among them, the evaluation objective is to evaluate the degree of abnormality in the financial data, and the evaluation factors include abnormal values, missing values and data type mismatch. An abnormal value is a data point that significantly deviates from the normal data range, a missing value is data that is missing or not recorded in the data set, and data type mismatch means that the actual type of the data field is inconsistent with the expected type. Use the statistical method of Z-score to calculate the degree of abnormal values. The Z-score represents the number of standard deviations between a data point and the average value of the data set. Divide it into different abnormal levels according to the degree of abnormal values, analyze the impact degree of missing values on the data analysis results, determine different abnormal levels in combination with the quantity and proportion of missing values, check whether the type of the data field is consistent with the expectation, and divide it into different abnormal levels according to the degree of data type mismatch and the impact on the data analysis results. Analyze the evaluation results of abnormal values, missing values and data type mismatch, determine the abnormal type of each data point, and match different abnormal levels according to the error degrees of abnormal values, missing values and data type mismatch.

4. The AI ​​intervention correction method for financial data entry errors according to claim 3 is characterized by: For the evaluation of the abnormal values, set the threshold of the abnormal value degree as Z>3 to divide the data points into abnormal values. Among them, the severe abnormal threshold of the abnormal value is Z>5, which is expressed as an extreme abnormal value; the medium abnormal threshold of the abnormal value is 4<Z≤5, which is expressed as a relatively mild abnormal value; the slight abnormal threshold of the abnormal value is 3<Z≤4, which is expressed as an abnormal value with a low interference degree. For the evaluation of the missing values, thresholds for classifying the degree of impact of the missing values ​​are set, wherein the severe abnormal threshold for missing values ​​is when the proportion of missing values ​​in the data field exceeds 10%, the moderate abnormal threshold for missing values ​​is when the proportion of missing values ​​is between 5% and 10%, and the slight abnormal threshold for missing values ​​is when the proportion of missing values ​​is less than 5%; For the evaluation of the data type mismatch, thresholds for the degree of data type mismatch are set, wherein the severe abnormality threshold for data type mismatch is that the proportion of the number of unmatched records to the total number of records exceeds 20%, the moderate abnormality threshold for data type mismatch is that the proportion of the number of unmatched records to the total number of records is between 5% and 20%, and the slight abnormality threshold for data type mismatch is that the proportion of the number of unmatched records to the total number of records is less than 5%.

5. The AI ​​intervention correction method for financial data entry errors according to claim 4 is characterized by: In step 3, the process of setting the correction threshold includes: Integrate the relevant data of outliers, missing values ​​and data type mismatches, and compare the outliers, missing values ​​and data type mismatches with their respective thresholds to obtain the outlier comparison difference values, missing value comparison difference values ​​and mismatch comparison difference values; Based on the evaluation results of outliers, missing values, and data type mismatch, different weights are assigned to different types of anomalies; The error assessment coefficient is calculated by combining the comparison difference values ​​of outliers, missing values, and data type mismatches to quantify the severity of data errors; In combination with the value of the error assessment coefficient, different error levels are set, namely a low error level, a medium error level and a high error level, and a corresponding correction threshold is matched for each error level.

6. The AI ​​intervention correction method for financial data entry errors according to claim 5, characterized in that: The calculation formula of the error assessment coefficient is: Among them, E is the error evaluation coefficient, A is the difference value of the outlier comparison, Z is the Z score, ω A is the weight of the outlier comparison difference value, M is the missing value comparison difference value, M t is the total number of missing values, M max is the maximum number of records, ω M is the weight of missing value comparison difference value, D is the weight of mismatch comparison difference value, D t is the total number of data type mismatches, D max is the total number of fields, ω D is the weight of the mismatched alignment difference value; The multiple error levels correspond to the multiple correction thresholds, specifically satisfying the following relationship: Low error level: 0 <E≤E L ; Medium error level: E L <E≤E M ; High error level: E M <E<1; Among them, E is the error assessment coefficient, E L is the upper threshold corresponding to the low error level and the lower threshold corresponding to the medium error level, E M is the upper threshold corresponding to the medium error level and the lower threshold corresponding to the high error level, E L =0.33, E M =0.

66.

7. The AI ​​intervention correction method for financial data entry errors according to claim 6, characterized in that: In step 4, the process of classifying the erroneous financial data includes: According to the different error levels set, formulate specific error handling measures for each error level; Determine the specific error level of the entered financial data based on the error assessment coefficient and the different error levels set; Based on the type and severity of the error, determine whether the AI ​​system automatically intervenes to correct it or manually intervenes, and record detailed information on each error handling, including the error type, error level, and measures taken.

8. The AI ​​intervention correction method for financial data entry errors according to claim 7, characterized in that: The process of determining whether the AI ​​system automatically intervenes in the correction or manually intervenes specifically includes: Use AI systems to monitor and detect input financial data in real time to identify potential errors and classify them into different categories based on their specific types and characteristics; Combined with the recognition results of the error level, the correction method is determined and analyzed to see whether manual intervention is needed; For errors with low error levels that are easy to correct automatically, the AI ​​system automatically intervenes and makes judgments, and simple data format adjustments or missing value fillings are performed by the AI ​​system. For medium error levels, the AI ​​system is used to mark data points for review and send them for manual review. For high error levels, the manual intervention process is automatically triggered.

9. The AI ​​intervention correction method for financial data entry errors according to claim 8, characterized in that: In step 5, the process of determining whether errors requiring manual intervention are queued includes: Develop AI system automatic correction strategies based on error types and historical data, including data cleaning rules, missing value filling algorithms, and outlier detection and adjustment; Input the classified erroneous data into the AI ​​system, automatically correct the erroneous data according to the preset strategy, and record the AI ​​correction process and results, including the data comparison before and after correction and the correction time; For erroneous data that the AI ​​system cannot automatically correct or the correction effect is unsatisfactory, manual intervention is performed to mark the data. The marking information includes the error type, error level and recommended correction measures; Analyze the error level and determine whether queuing is needed based on the error priority and urgency. For data with medium error levels, queue them according to time to wait for manual intervention and correction. For data with high error levels, send them directly for manual intervention and correction.

10. The AI ​​intervention correction method for financial data entry errors according to claim 9, characterized in that: In step 6, the process of optimizing the financial data correction process includes: Use automated monitoring tools to monitor the quality of the corrected data, including completeness, accuracy, consistency, uniqueness, and timeliness, and record the results of each inspection, including the items that passed and failed, generate data quality reports, and report the monitoring status to relevant personnel; Set up a user feedback mailbox and online form as feedback channels, and regularly collect feedback submitted by users, including data errors, system performance issues, and user interface suggestions, and classify and organize the feedback to form a feedback list; Use system logs, monitoring tools or performance counters to collect performance data of the financial data correction process, analyze the collected performance data, identify performance bottlenecks and potential problems, compare performance data in different time periods, and analyze performance change trends; Combine data quality monitoring results, user feedback and performance data analysis to identify problems in the financial data correction process, conduct detailed analysis of the problems, locate the root causes of the problems, and then formulate specific optimization measures for the identified problems. According to the formulated optimization measures, the financial data correction process is improved and optimized.

Images