Intelligent accounting data processing method and device, equipment and medium

Abstract

The invention relates to an intelligent accounting data processing method and device, equipment and a medium. The method comprises the following steps: acquiring an accounting data sequence of a target company; analyzing an abnormal condition of the accounting data sequence based on a preset business scene rule and preset reference distribution data to obtain a first abnormal data set; the first abnormal data set at least comprises accounting data, a business scene label of the accounting data and a first abnormal score; based on a preset business rule set, screening the first abnormal data set to obtain a second abnormal data set; the business rule set is used for indicating and determining business demand labels of the accounting data; the second abnormal data set at least comprises abnormal accounting data, a business scene label of the abnormal accounting data and a second abnormal score; and comprehensively analyzing the second abnormal data set based on the business scene rule and the accounting data sequence to obtain an abnormal analysis report. By adopting the method, the accuracy and practicability of accounting data anomaly detection can be improved.

Description

Technical Field

[0001] This invention belongs to the field of intelligent accounting, and in particular relates to an intelligent accounting data processing method, apparatus, equipment and medium. Background Technology

[0002] With the development of intelligent accounting technology, rule-based and statistical accounting data anomaly detection technology has emerged. This technology can screen financial data in batches through preset logical conditions or quantitative models, which in turn has led to the emergence of traditional anomaly detection methods based on fixed threshold detection.

[0003] In traditional anomaly detection methods, the fixed threshold anomaly detection process typically involves the following steps: First, technical personnel set global warning thresholds for key accounting items based on historical experience or general statistical indicators (such as the three-standard-deviation principle). Then, the system compares the value of each transaction in the current period with the corresponding static threshold for that item, automatically triggering an anomaly flag when the transaction amount or balance exceeds the threshold boundary. Finally, the system outputs an anomaly list containing all records exceeding the limits; this list usually only marks the deviation in value and lacks explanations at the business scenario level. The entire process relies on a pre-set, unified standard, using the same sensitivity parameters to judge data generated by all business processes.

[0004] However, the current fixed threshold detection method has obvious adaptability defects. The most critical problem is that it cannot perceive the inherent differences in data fluctuation patterns under different business scenarios, resulting in a large number of false alarms or omissions in the actual application of the system, which directly restricts the accuracy and practicality of accounting data anomaly detection. Summary of the Invention

[0005] Therefore, it is necessary to provide an intelligent accounting data processing method, device, equipment, and medium that can improve the accuracy and practicality of accounting data anomaly detection, addressing the aforementioned technical problems.

[0006] Firstly, this application provides an intelligent accounting data processing method, including:

[0007] Obtain the target company's accounting data sequence; the accounting data sequence should at least include the execution timestamp, execution amount, and execution account type of the accounting data.

[0008] Based on preset business scenario rules and preset benchmark distribution data, anomalies in accounting data sequences are analyzed to obtain a first anomaly dataset; the first anomaly dataset includes at least accounting data, business scenario labels for the accounting data, and a first anomaly score;

[0009] Based on a preset set of business rules, the first abnormal dataset is filtered to obtain a second abnormal dataset; the set of business rules is used to indicate the business requirement labels for determining accounting data; the second abnormal dataset includes at least abnormal accounting data, business scenario labels for the abnormal accounting data, and a second abnormal score.

[0010] Based on business scenario rules and accounting data sequences, a comprehensive analysis of the second abnormal dataset is conducted to obtain an anomaly analysis report; the anomaly analysis report is used to characterize the degree of anomaly in the accounting data sequence.

[0011] Furthermore, the baseline distribution data was obtained through the following method:

[0012] Obtain the preset historical accounting dataset; the historical accounting dataset includes accounting data from the historical accounting data processing process and corresponding business scenario tags;

[0013] Based on historical accounting datasets, the distribution relationship between execution amounts and corresponding business scenario labels is quantified to obtain amount distribution parameters; the amount distribution parameters are used to characterize the probability distribution characteristics of execution amounts under each business scenario label.

[0014] Based on historical accounting datasets, time-series features are extracted from execution timestamps and corresponding business scenario labels to obtain time distribution parameters. These time distribution parameters are used to characterize the time distribution features of execution timestamps under each business scenario label.

[0015] Based on historical accounting datasets, the relationship between execution account types and corresponding business scenario labels is quantified to obtain account association rules; account association rules are used to characterize the association features between execution account types under each business scenario label;

[0016] Based on the amount distribution parameters, time distribution parameters, and account association rules, the baseline distribution data is obtained.

[0017] Furthermore, based on preset business scenario rules and preset benchmark distribution data, anomalies in the accounting data sequence are analyzed to obtain the first anomaly dataset, including:

[0018] Based on accounting data sequences and business scenario rules, business scenario labels are obtained for each accounting data; business scenario rules are used to characterize the mapping relationship between business scenario labels and execution account types.

[0019] Based on the business scenario labels and the amount distribution parameters in the benchmark distribution data, the abnormality of the execution amount of each accounting data in the accounting database column is quantified to obtain the amount abnormality feature score of each accounting data.

[0020] Based on the time pattern distribution parameters in the business scenario labels and benchmark distribution data, the expected execution timestamp of each accounting data is determined, and the difference between the expected execution timestamp and the execution timestamp is quantified to obtain the time anomaly characteristic score of each accounting data.

[0021] Based on the account association rules in business scenario tags and benchmark distribution data, the abnormal situation of the execution account type of each accounting data in the accounting database column is quantified to obtain the account association abnormal feature score of each accounting data.

[0022] Based on the abnormal characteristic scores of monetary amount, time, and account association, the first abnormality score for each accounting data is calculated using the following formula:

[0023]

[0024] in, It is accounting data The first abnormality score, It is the weight of abnormal amounts. It is accounting data The abnormal feature score of the amount, It is the time anomaly weight. It is accounting data Temporal anomaly feature score, It is the subject association weight. It is accounting data Subject-related abnormal feature scores;

[0025] The accounting data with a first anomaly score greater than a preset first anomaly threshold are selected from the accounting database column to obtain the first anomaly dataset.

[0026] Furthermore, the set of business rules also includes the business data conditions, penalty weights, and importance coefficients for each business requirement tag; business requirement tags include normal business tags and abnormal business tags;

[0027] Based on a preset set of business rules, the first abnormal dataset is filtered to obtain a second abnormal dataset, which includes:

[0028] For each accounting data point in the first abnormal dataset, the following formula is used to match the accounting data with the business data conditions of each normal business label to obtain the matching label value of the accounting data:

[0029]

[0030] in, It is the first Matching tag values ​​for accounting data It is any normal business tag in the set of business rules. Business data conditions, It is a matching indicator function. It is the first One accounting data;

[0031] Based on the preset anomaly matching label threshold, accounting data whose matching label values ​​meet the anomaly matching label threshold are selected from the first anomaly dataset to obtain the business anomaly dataset;

[0032] For each accounting data in the business anomaly dataset, the accounting data is matched with the business data conditions corresponding to each anomaly business tag to obtain the anomaly business matching result for each accounting data; and based on the anomaly business matching result, anomaly business tags are selected to obtain the anomaly business tag set for each accounting data.

[0033] For each accounting data in the business anomaly dataset, based on the abnormal business tag set of the accounting data, the difference between the accounting data and the business data conditions of each abnormal business tag is quantified to obtain the blacklist score of each accounting data.

[0034] For each accounting data in the business anomaly dataset, a second anomaly score is obtained based on the blacklist score and the first anomaly score.

[0035] Based on the business scenario tags of accounting data and the preset second abnormal threshold data, the second abnormal threshold of accounting data is determined, and accounting data with a second abnormal score greater than the second abnormal threshold is identified as abnormal accounting data.

[0036] The second abnormal dataset is obtained based on the second abnormal score of abnormal accounting data, business scenario labels, and abnormal business label set.

[0037] Furthermore, the second anomaly threshold data was obtained through the following method:

[0038] Obtain preset historical second anomaly score data; the historical second anomaly score data includes the second anomaly scores of historical accounting data for each business scenario tag;

[0039] For each business scenario tag, the statistical features of the second anomaly score are extracted from the historical accounting data to obtain the statistical feature value of the second anomaly score for each business scenario tag; the statistical feature value of the second anomaly score includes the statistical mean of the second anomaly score and the statistical standard deviation of the second anomaly score.

[0040] For each business scenario tag, the second anomaly score statistical feature value is used to calculate the second anomaly threshold of the business scenario tag using the following formula:

[0041]

[0042] in, It can be a tag for any business scenario. Business scenario tags The second abnormal threshold, Business scenario tags The statistical mean of the second abnormal score, It is an adjustment factor. Business scenario tags The statistical standard deviation of the second outlier score;

[0043] The second anomaly threshold data is obtained based on the second anomaly threshold of each business scenario label.

[0044] Furthermore, based on business scenario rules and accounting data sequences, a comprehensive analysis of the second abnormal dataset is conducted to obtain an anomaly analysis report, including:

[0045] Based on business scenario rules, accounting data sequences, and a second abnormal dataset, the abnormal distribution of each business scenario label is quantified to obtain the risk index of each business scenario label.

[0046] Based on the abnormal business label set in the second abnormal data set, the distribution of each abnormal business label in the abnormal business label set is quantified to obtain the risk score of each abnormal business label.

[0047] Input the abnormal business label set of each accounting data in the second abnormal dataset into the preset response action prediction model to obtain the processing action data of each accounting data.

[0048] An anomaly analysis report is generated based on the business scenario labels, processing action data, second anomaly scores, risk indices of business scenario labels, and risk scores of anomalous business labels for each accounting data in the second anomaly dataset.

[0049] Furthermore, the response action prediction model was obtained through the following method:

[0050] Acquire preset historical response action data; historical response action data includes abnormal business tags of accounting data in historical processing and corresponding processing action data;

[0051] Using the abnormal business tags of accounting data in the historical processing process as feature values ​​and the corresponding processing action data as feature attributes, historical response action mapping data is obtained, and the historical response action mapping data is divided into training sample set and test sample set according to a preset ratio.

[0052] Decision tree models are constructed by randomly selecting feature values ​​and combinations, and a preliminary response action prediction model is built based on multiple decision tree models.

[0053] The initial response action prediction model is trained using a training sample set, and the trained response action prediction model is tested using a test sample set until the response action prediction model meets the preset accuracy requirements, thus obtaining the response action prediction model.

[0054] Secondly, this application also provides an intelligent accounting data processing device, comprising:

[0055] The data acquisition module is used to acquire the target company's accounting data sequence; the accounting data sequence includes at least the execution timestamp, execution amount, and execution account type of the accounting data;

[0056] The first anomaly screening module is used to analyze the anomalies in the accounting data sequence based on preset business scenario rules and preset benchmark distribution data to obtain the first anomaly dataset; the first anomaly dataset includes at least accounting data, business scenario tags for the accounting data, and a first anomaly score;

[0057] The second anomaly filtering module is used to filter the first anomaly dataset based on a preset set of business rules to obtain a second anomaly dataset; the set of business rules is used to indicate the business requirement labels for determining accounting data; the second anomaly dataset includes at least the abnormal accounting data, the business scenario labels of the abnormal accounting data, and the second anomaly score.

[0058] The report generation module is used to comprehensively analyze the second abnormal dataset based on business scenario rules and accounting data sequences to obtain an anomaly analysis report; the anomaly analysis report is used to characterize the degree of anomaly in the accounting data sequence.

[0059] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement any of the intelligent accounting data processing methods described in the first aspect of this application.

[0060] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the intelligent accounting data processing methods described in the first aspect of this application.

[0061] The aforementioned intelligent accounting data processing method, apparatus, equipment, and medium acquire the accounting data sequence of the target company. The accounting data sequence includes at least the execution timestamp, execution amount, and execution account type of the accounting data. Based on preset business scenario rules and preset benchmark distribution data, anomalies in the accounting data sequence are analyzed to obtain a first abnormal dataset. The first abnormal dataset includes at least accounting data, business scenario tags for the accounting data, and a first anomaly score. Based on a preset set of business rules, the first abnormal dataset is filtered to obtain a second abnormal dataset. The business rule set is used to indicate and determine the business requirement tags for the accounting data. The second abnormal dataset includes at least abnormal accounting data, business scenario tags for the abnormal accounting data, and a second anomaly score. Based on the business scenario rules and the accounting data sequence, the second abnormal dataset is comprehensively analyzed to obtain an anomaly analysis report. The anomaly analysis report is used to characterize the degree of anomaly in the accounting data sequence. This successfully transforms the lagging detection mode of traditional accounting auditing, which relies on fixed thresholds and human experience, into an intelligent detection paradigm that dynamically perceives business context, integrates statistical learning and symbolic rules, and possesses continuous learning capabilities. This improves the accuracy and practicality of accounting data anomaly detection in complex and ever-changing business environments. Attached Figure Description

[0062] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0063] Figure 1 A flowchart illustrating an intelligent accounting data processing method provided in one embodiment of this application;

[0064] Figure 2 This is a schematic diagram of the structure of an intelligent accounting data processing device provided in one embodiment of this application;

[0065] Figure 3 This is a schematic diagram of the structure of a computer device for an intelligent accounting data processing method provided in one embodiment of this application. Detailed Implementation

[0066] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0067] In one embodiment, such as Figure 1As shown, an intelligent accounting data processing method is provided. This embodiment illustrates the method applied to a terminal, but it is understood that the method can also be applied to a server, or to a system including both a terminal and a server, and is implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following S101-S104, wherein:

[0068] S101, Obtain the target company's accounting data sequence; the accounting data sequence shall include at least the execution timestamp, execution amount, and execution account type of the accounting data.

[0069] Specifically, the terminal acquires the target company's accounting data sequence, which includes at least the execution timestamp, execution amount, and execution account type. The execution timestamp indicates the point in time when the accounting data occurred; the execution amount indicates the transaction value of the accounting data; and the execution account type indicates the classification account of the accounting data, which can be an account code according to accounting standards. For example, the accounting data sequence can be in the form of… ,in, It is an accounting data sequence. It is the number of the accounting data. It is a number The execution timestamp of the accounting data. It is a number The execution amount of the accounting data. It is a number The type of accounting account used for accounting data execution. It is the total number of accounting data included in the accounting data sequence.

[0070] S102, based on preset business scenario rules and preset benchmark distribution data, analyze the abnormal situations of the accounting data sequence to obtain the first abnormal dataset; the first abnormal dataset includes at least accounting data, business scenario labels of the accounting data, and a first abnormal score.

[0071] Specifically, the preset business scenario rules are a set of mapping rules used to associate accounting data with specific business scenario labels. The benchmark distribution data, on the other hand, is a statistical model built through offline analysis of a large number of historical normal transactions, used to characterize the fluctuation patterns of execution amount, execution timestamp, and execution account type under each first anomaly score. The terminal compares each accounting data in the current accounting data sequence with the benchmark distribution data of the corresponding business scenario label, calculates the degree of deviation, and generates a first anomaly score. This score quantifies the probability of anomalies recorded at the statistical level. Based on the first anomaly score, the terminal selects accounting data and the corresponding first anomaly score from the current accounting data sequence to obtain the first anomaly dataset. For example, the preset business scenario rules can take the form of... ,in These are preset business scenario rules. It is the type of account to be executed. It is the number of the business scenario tag. It is a number The system includes business scenario tags, and the preset business scenario rules can be set according to the accounting scenario classification rules in actual work. This represents the total number of business scenario tags.

[0072] S103, based on a preset set of business rules, the first abnormal dataset is filtered to obtain a second abnormal dataset; the set of business rules is used to indicate the business requirement labels for determining accounting data; the second abnormal dataset includes at least abnormal accounting data, business scenario labels for the abnormal accounting data, and a second abnormal score.

[0073] Specifically, a preset set of business rules is used to indicate the business requirement tags for determining accounting data. This includes business data conditions expressed in logical terms (e.g., "travel expenses exceeding 5000 yuan require additional approval"), and is associated with business requirement tags (e.g., "requires special approval," "compliant," etc.). These business requirement tags also include normal business tags and abnormal business tags. The set of business rules represents the mapping relationship from data features to business semantics. The terminal, using the preset set of business rules, excludes accounting data belonging to normal business tags from the first abnormal dataset, identifies the remaining accounting data as abnormal accounting data, and quantifies the association between the abnormal accounting data and the business data conditions corresponding to the abnormal business tags, obtaining a second abnormal score for each abnormal accounting data. The terminal integrates all abnormal accounting data, the business scenario tags of the abnormal accounting data, and the second abnormal scores to obtain a second abnormal dataset. For example, the second abnormal dataset may also include a set of abnormal business tags for the abnormal accounting data, which includes the abnormal business tags to which the abnormal accounting data belongs. Optionally, the preset set of business rules can take the form of… ,in, It is a pre-defined set of business rules. It is the first Business data conditions for each business requirement tag It is the first Individual business requirement tags The label indicating this business requirement is a normal business label, used to indicate that accounting data meeting the data conditions for this business is considered normal data; when At this time, the label indicating that the business requirement is an abnormal business label is used to indicate that accounting data that meets the business data conditions is abnormal data, such as "over-budget expenditure" or "abnormal supplier qualifications". The preset set of business rules can be set according to the relationship between accounting data and business requirements in actual work.

[0074] S104. Based on business scenario rules and accounting data sequences, a comprehensive analysis of the second abnormal dataset is conducted to obtain an anomaly analysis report. The anomaly analysis report is used to characterize the degree of anomaly in the accounting data sequence.

[0075] Specifically, the terminal performs risk aggregation on each accounting data in each second abnormal dataset based on the accounting data and the business scenario labels of the accounting data, combining business scenario rules and accounting data sequences, and from the perspective of business scenario label distribution, calculates the overall accounting data quality score, and generates an anomaly analysis report based on the quality score.

[0076] This embodiment provides an intelligent accounting data processing method. It obtains the execution timestamp, execution amount, and execution account type of accounting data; utilizes integrated business scenario knowledge and historical statistical patterns to achieve efficient and adaptive preliminary screening, obtaining a first abnormal dataset; and refines the screening results semantically by introducing explicit business rules to remove false positives and improve accuracy, obtaining a second abnormal dataset. Furthermore, it aggregates point-like anomalies into a surface-like risk view and outputs an anomaly analysis report that can directly support management decisions. This successfully transforms the traditional accounting auditing method, which relies on fixed thresholds and human experience for lagging detection, into an intelligent detection paradigm that dynamically perceives business context, integrates statistical learning and symbolic rules, and possesses continuous learning capabilities. This improves the accuracy and practicality of accounting data anomaly detection in complex and ever-changing business environments.

[0077] In one embodiment, the baseline distribution data is obtained through the following method:

[0078] S201, Obtain the preset historical accounting dataset; the historical accounting dataset includes accounting data from the historical accounting data processing process and corresponding business scenario labels.

[0079] Specifically, the terminal acquires a preset historical accounting dataset; the historical accounting dataset includes accounting data from historical accounting data processing and corresponding business scenario tags. The historical accounting dataset can be in the following format: ,in It is a historical accounting dataset. It is the number of accounting data in the historical accounting data processing process. It is a number The execution timestamp of the accounting data. It is a number The execution amount of the accounting data. It is a number The type of accounting account used for accounting data execution. It is a number Business scenario tags for accounting data This refers to the total number of accounting data included in the accounting data sequence. The preset historical accounting dataset can be obtained from historical accounting data in actual work. This embodiment does not further limit the setting of this preset historical accounting dataset.

[0080] S202, based on historical accounting datasets, quantifies the distribution relationship between execution amount and corresponding business scenario labels to obtain amount distribution parameters; amount distribution parameters are used to characterize the probability distribution characteristics of execution amount under each business scenario label.

[0081] Specifically, the terminal traverses the historical accounting dataset, calculates the execution amount corresponding to each business scenario label, and forms a subset of amounts for each business scenario label. The terminal uses a Gaussian mixture model to fit the probability distribution of the amount subset for each business scenario label, establishing parameters for the normal fluctuation range of the amount values, thus obtaining the amount distribution parameters. The Gaussian mixture model is a clustering algorithm based on probability density estimation. It fits complex data distribution patterns through a linear combination of multiple Gaussian distributions. Based on the amount subsets of the same business scenario label, the terminal uses the Bayesian information criterion to decompose the execution amount of the same business scenario label into several Gaussian components, and estimates the weight, mean, and standard deviation of each Gaussian component, thereby obtaining a complete probability distribution representation of the amount under that business scenario label. For example, the form of the amount distribution parameters can be... ,in, It is a parameter for the distribution of monetary amounts. It can be a tag for any business scenario. This represents the total number of business scenario tags. It is a business scenario label in historical accounting data. The number of Gaussian components, Business scenario tags The Middle The mixing weights of Gaussian components, Business scenario tags No. The mean of the Gaussian components, Business scenario tags No. The standard deviation of each Gaussian component.

[0082] S203, based on historical accounting datasets, extracts time-series features from execution timestamps and corresponding business scenario labels to obtain time distribution parameters; the time distribution parameters are used to characterize the time distribution features of execution timestamps under each business scenario label.

[0083] Specifically, the terminal iterates through the historical accounting dataset, counting the execution timestamps corresponding to each business scenario label, forming a subset of timestamps for each business scenario label. For each subset of timestamps for a business scenario label, the terminal uses a time series decomposition method to extract the periodicity and trend characteristics of the execution timestamps under the same business scenario label, obtaining time distribution parameters. Time series decomposition is a statistical method that breaks down time-series data into trend, seasonal, and residual components. The terminal uses a time series decomposition algorithm to detect inherent periods through execution timestamps, extracts trend and seasonal periodic terms through local weighted regression, and models the distribution characteristics of the residual sequence to obtain time distribution parameters. For example, the time distribution parameters can take the form of... ,in, Business scenario tags The trend component function, obtained by multinomial fitting, is used to describe the long-term variation of business scenario labels with execution timestamps. Business scenario tags The seasonal component function is used to reflect the periodic occurrence pattern of business scenario tags; Business scenario tags The normal distribution parameters of the residual components are used to characterize the random fluctuation range of the execution timestamp after removing trend and seasonal factors.

[0084] S204, based on historical accounting datasets, quantifies the association between execution account types and corresponding business scenario labels to obtain account association rules; account association rules are used to characterize the association features between execution account types under each business scenario label.

[0085] Specifically, the terminal employs the FP-Growth algorithm to mine association rules for execution subject types under each business scenario label, identifying frequently occurring subject sets and their association strengths to obtain subject association rules. The FP-Growth algorithm is a highly efficient association rule mining algorithm based on frequent pattern trees. It avoids generating candidate sets by constructing a compressed tree structure, thereby quickly discovering frequent subject combinations. The terminal uses the FP-Growth algorithm to statistically analyze the support of execution subject types appearing under each business scenario label and generate subject association rules. For example, the form of the subject association rules can be... .in, The business scenario is tagged as Subject association rules, It is the first Rule antecedents for each associated feature, It is the first The rule consequent of each associated feature, It is the first The support of each associated feature is used to characterize the frequency of the combination in historical accounting datasets. It is the first The confidence level of each associated feature is used to characterize when... When it appears The conditional probability of occurrence. At this point, the... The meaning of each associated feature is that the label in the business scenario is... Under these circumstances, 85% of historical transactions involve all three execution account types (raw materials, accounts payable, and input tax), and when the "raw materials" account appears, there is a 92% probability that the "accounts payable" and "input tax" execution account types will also appear.

[0086] S205, based on the amount distribution parameters, time distribution parameters and account association rules, obtains the baseline distribution data.

[0087] Specifically, the terminal concatenates the amount distribution parameters, time distribution parameters, and subject association rules to obtain the baseline distribution data.

[0088] This embodiment provides an intelligent accounting data processing method. By acquiring historical accounting datasets and employing probabilistic models, time-series analysis algorithms, and rule mining algorithms, it captures the complex patterns of amount distribution, the temporal regularities of business activities, and the inherent logical relationships between execution account types. This yields amount distribution parameters, time distribution parameters, and account association rules, which are then integrated into baseline distribution data. This successfully transforms scattered historical transaction data into a structured business knowledge base, providing dynamic and multi-dimensional benchmark data for real-time anomaly detection, thus improving the accuracy and reliability of accounting data anomaly detection in different business scenarios.

[0089] In one embodiment, based on preset business scenario rules and preset benchmark distribution data, anomalies in the accounting data sequence are analyzed to obtain a first abnormal dataset, including:

[0090] S301, based on the accounting data sequence and business scenario rules, obtain the business scenario label for each accounting data; the business scenario rules are used to characterize the mapping relationship between the business scenario label and the execution account type.

[0091] Specifically, the terminal assigns a corresponding business scenario label to each accounting data point in the accounting data sequence based on preset business scenario rules, thus obtaining the business scenario label for each accounting data point. For example, one of the business scenario rules is... When there is an accounting data in the accounting data sequence When the execution account type is travel expenses, travel reimbursement should be used as the accounting data. Business scenario tags .

[0092] S302, based on the business scenario labels and the amount distribution parameters in the benchmark distribution data, quantifies the abnormality of the execution amount of each accounting data in the accounting database column, and obtains the amount abnormality feature score of each accounting data.

[0093] Specifically, for each accounting data in the accounting database column, the terminal matches its business scenario label with the amount distribution parameters to determine the mixed weight, mean, and standard deviation parameters of the accounting data, and uses the following formula based on the Gaussian mixture model parameters: The abnormality score of the amount in the accounting data is calculated. It is the first The score of abnormal characteristics of the amount in each accounting data point It is accounting data Business scenario tags, It is the first The execution amount of each accounting data point Business scenario tags The number of Gaussian components, Business scenario tags The Middle The mixing weights of Gaussian components, Business scenario tags The Middle The mean of the Gaussian components, Business scenario tags The Middle The standard deviation of each Gaussian component It is a Gaussian distribution function.

[0094] S303, based on the time pattern distribution parameters in the business scenario labels and benchmark distribution data, determines the expected execution timestamp of each accounting data, quantifies the difference between the expected execution timestamp and the execution timestamp, and obtains the time anomaly characteristic score of each accounting data.

[0095] Specifically, for each accounting data in the accounting database column, the terminal matches its business scenario label with time pattern distribution parameters, uses the trend component function and seasonal component function in the time distribution parameters to predict the normal transaction time that should exist under that business scenario label, obtains the expected execution timestamp, and calculates it using the following formula: The time anomaly score of the accounting data is calculated. It is the first The time anomaly score of each accounting data point It is the first The execution timestamp of each accounting data point It is the first Business scenario tags for accounting data Business scenario tags The trend component function, Business scenario tags The seasonal component function, Business scenario tags The standard deviation of the residuals can be obtained from the time-varying distribution parameters.

[0096] S304. Based on the subject association rules in the business scenario labels and benchmark distribution data, quantify the abnormal situation of the execution subject type of each accounting data in the accounting database column, and obtain the subject association abnormality feature score of each accounting data.

[0097] Specifically, for each accounting data in the accounting database column, the terminal matches its business scenario label with time pattern distribution parameters, compares the matching degree of the current record's account combination with the historical frequent account combination patterns under the corresponding business scenario, and identifies and quantifies the degree of structural anomalies that violate the conventional accounting account combination logic, thus obtaining the account association anomaly feature score for that accounting data. The account association anomaly feature score can be calculated based on the confidence level of the association rules using the following formula: , calculated to be. Where, It is the first The scores of abnormal characteristics associated with individual accounting data items It is the first Business scenario tags for accounting data Business scenario tags Subject association rules, Business scenario tags Any one of the subject association rules, It is any one of the subject association rules. The confidence level represents the probability that the rule holds true in historical data.

[0098] S305, based on the abnormal characteristic scores of amount, time, and account association, the first abnormality score of each accounting data is calculated using the following formula:

[0099]

[0100] in, It is accounting data The first abnormality score, It is the weight of abnormal amounts. It is accounting data The abnormal feature score of the amount, It is the time anomaly weight. It is accounting data Temporal anomaly feature score, It is the subject association weight. It is accounting data Subject-related abnormal feature scores.

[0101] Specifically, the terminal weights and combines the abnormal characteristic scores of amount, time, and account association for each accounting data point to obtain a first abnormality score for each data point. Among these, the accounting data... The abnormal amount feature score database is obtained from S302, accounting data. The time anomaly characteristic score can be obtained from S303, accounting data. The subject-related anomaly score can be obtained from S304. Amount anomaly weight. Time anomaly weight is used to characterize the impact of monetary anomaly feature scores on the first anomaly score. Subject-related weights are used to characterize the impact of time-related anomaly feature scores on the first anomaly score. This is used to characterize the impact of the subject association anomaly feature score on the first anomaly score. The sum of the amount anomaly weight, the amount anomaly weight, and the subject association weight is 1, which can be set according to actual work.

[0102] S306, Select accounting data from the accounting database column whose first anomaly score is greater than the preset first anomaly threshold to obtain the first anomaly dataset.

[0103] Specifically, the terminal selects accounting data from the accounting database column whose first anomaly score is greater than a preset first anomaly threshold, and integrates the aforementioned accounting data and the corresponding first anomaly scores to obtain a first anomaly dataset. The preset first anomaly threshold can be determined based on the distribution of anomaly scores in historical normal data; this embodiment does not further limit this preset first anomaly threshold.

[0104] This embodiment provides an intelligent accounting data processing method that determines the business scenario labels of each accounting data item in an accounting database column, and then quantifies abnormal signals from three key business dimensions: amount, time, and account association, comprehensively obtaining a first anomaly score for each accounting data item. Finally, based on a preset first anomaly threshold, a high-quality candidate anomaly set is output. This elevates traditional single-dimensional anomaly detection to multi-dimensional collaborative analysis based on business scenario awareness, significantly reducing the false positive rate while maintaining a high recall rate, thereby improving the accuracy and reliability of accounting data anomaly detection in different business scenarios.

[0105] In one embodiment, the set of business rules also includes business data conditions, penalty weights, and importance coefficients for each business requirement tag; the business requirement tags include normal business tags and abnormal business tags;

[0106] Based on a preset set of business rules, the first abnormal dataset is filtered to obtain a second abnormal dataset, which includes:

[0107] S401, for each accounting data in the first abnormal dataset, use the following formula to match the accounting data with the business data conditions of each normal business label to obtain the matching label value of the accounting data:

[0108]

[0109] in, It is the first Matching tag values ​​for accounting data It is any normal business tag in the set of business rules. Business data conditions, It is a matching indicator function. It is the first Accounting data.

[0110] Specifically, the matching indicator function Used to indicate matching feedback results, when accounting data Business data conditions that satisfy any normal business tag in the business rule set hour, The value is 1 if the condition is not met, and 0 otherwise. The terminal identifies accounting data that, while statistically abnormal, meets normal business requirements by traversing all business data conditions under normal business tags. The business rule set also includes the business data conditions, penalty weights, and importance coefficients for each business requirement tag, and these tags include both normal and abnormal business tags. The business rule set can take the form of: ,in, It is a pre-defined set of business rules. It is the first Business data conditions for each business requirement tag It is the execution amount. It is the type of account to be executed. It is the penalty weight. It is the importance coefficient. It is the first Individual business requirement tags The label indicating this business requirement is a normal business label, used to indicate that accounting data meeting the data conditions for this business is normal data, such as "normal expense reimbursement"; when At this time, the label indicating that the business demand is an abnormal business label is used to indicate that the accounting data that meets the data conditions of the business is abnormal data, such as "over-budget expenditure".

[0111] S402, based on the preset anomaly matching label threshold, select accounting data whose matching label values ​​meet the anomaly matching label threshold from the first anomaly dataset to obtain the business anomaly dataset.

[0112] Specifically, the terminal selects accounting data whose matching label values ​​meet a preset abnormal matching label threshold from the first abnormal dataset to form a business abnormal dataset. The preset abnormal matching label threshold is when the matching label value of the accounting data is 0, indicating that the accounting data does not meet the business data conditions of any normal business label.

[0113] S403: For each accounting data in the business anomaly dataset, match the accounting data with the business data conditions corresponding to each anomaly business label to obtain the anomaly business matching result for each accounting data; and select anomaly business labels based on the anomaly business matching result to obtain the anomaly business label set for each accounting data.

[0114] Specifically, the terminal performs a matching check on each accounting data in the business anomaly dataset against the business data conditions corresponding to all anomaly business tags, collects all anomaly business tags that meet the business data conditions, and the penalty weight and importance coefficient of the anomaly business tags, forming an anomaly business tag set for that accounting data.

[0115] S404, for each accounting data in the business anomaly dataset, based on the abnormal business label set of the accounting data, quantifies the difference between the accounting data and the business data conditions of each abnormal business label, and obtains the blacklist score of each accounting data.

[0116] Specifically, for each accounting data point in the business anomaly dataset, the penalty weight and importance coefficient of each anomaly label in the anomaly label set are calculated using the following formula: The data is then weighted and combined to obtain a blacklist score for the accounting data. It is the first in the business anomaly dataset. Blacklist scoring of accounting data It is accounting data Any abnormal business tag in the abnormal business tag set, It is accounting data Abnormal business tag set, It is an abnormal business tag. The penalty weight, It is an abnormal business tag. Importance coefficient.

[0117] S405: For each accounting data in the business anomaly dataset, based on the blacklist score and the first anomaly score, obtain the second anomaly score for each accounting data.

[0118] Specifically, for each accounting data point in the business anomaly dataset, the terminal uses the following formula: The second anomaly score for each accounting data point was calculated. It is the first in the business anomaly dataset. The second outlier score of the accounting data. It is the first Blacklist scoring of accounting data It is the first The first anomaly score for each accounting data point This is the blacklist scoring weight, used to characterize the impact of the blacklist score on the second abnormal score. This is the weight of the first anomaly score, used to characterize the influence of the first anomaly score on the second anomaly score; blacklist score weight. and the weight of the first abnormal score The sum of is 1, which can be set according to the actual work.

[0119] S406, based on the business scenario label of accounting data and the preset second abnormal threshold data, determine the second abnormal threshold of accounting data, and identify accounting data with a second abnormal score greater than the second abnormal threshold as abnormal accounting data.

[0120] Specifically, for each accounting data in the business anomaly dataset, the terminal can determine the second anomaly threshold for that accounting data from the preset second anomaly threshold data based on its business scenario label. The terminal then compares the second anomaly score of that accounting data with the second anomaly threshold. If the second anomaly score of the accounting data is greater than the second anomaly threshold, the terminal identifies that accounting data as an anomaly. The preset second anomaly threshold data is pre-calculated by analyzing the distribution characteristics of the second anomaly scores for each business scenario label in historical data.

[0121] S407. Based on the second anomaly score of the abnormal accounting data, the business scenario label, and the abnormal business label set, the second anomaly dataset is obtained.

[0122] Specifically, the terminal integrates the second abnormality score, business scenario label, and abnormal business label set of each abnormal accounting data to obtain the second abnormality dataset.

[0123] This embodiment provides an intelligent accounting data processing method that obtains a second abnormal dataset through normal business label filtering, abnormal label matching, blacklist scoring calculation, score fusion, scenario-adaptive threshold determination, and result integration. This achieves deep business semantic verification of preliminary abnormal data, successfully upgrading purely statistically driven anomaly detection to intelligent verification that deeply integrates statistical analysis and business rules. While maintaining a high detection recall rate, it significantly improves accuracy and business relevance. The output results not only indicate "whether it is abnormal" but also clearly explain "why it is abnormal," thereby improving the accuracy and reliability of accounting data anomaly detection.

[0124] In one embodiment, the second anomaly threshold data is obtained by the following method:

[0125] S501, obtain the preset historical second abnormal score data; the historical second abnormal score data includes the second abnormal score of historical accounting data for each business scenario label.

[0126] Specifically, the terminal acquires preset historical second abnormal score data, wherein the historical second abnormal score data can be in the following format: ,in, This is the second most anomalous score in history. It can be a tag for any business scenario. The business scenario is tagged as The second set of outlier scores in historical accounting data, This refers to the total number of business scenario tags. The preset historical second anomaly score data can be obtained based on actual work records; this embodiment does not further limit the setting of this preset historical second anomaly score data.

[0127] S502, for the historical accounting data of each business scenario label, extract the statistical features of the second abnormal score to obtain the statistical feature value of the second abnormal score for each business scenario label; the statistical feature value of the second abnormal score includes the statistical mean of the second abnormal score and the statistical standard deviation of the second abnormal score.

[0128] Specifically, for each business scenario tag's historical accounting data, the terminal calculates the mean and standard deviation of the second abnormal scores under that business scenario tag based on the second abnormal score set of that business scenario tag, obtains the statistical mean and statistical standard deviation of the second abnormal scores, and concatenates the statistical mean and statistical standard deviation of the second abnormal scores to obtain the statistical feature value of the second abnormal score of that business scenario tag.

[0129] S503, for the statistical feature value of the second anomaly score of each business scenario label, the second anomaly threshold of the business scenario label is calculated using the following formula:

[0130]

[0131] in, It can be a tag for any business scenario. Business scenario tags The second abnormal threshold, Business scenario tags The statistical mean of the second abnormal score, It is an adjustment factor. Business scenario tags The statistical standard deviation of the second outlier score.

[0132] Specifically, for each business scenario tag, the terminal calculates the second anomaly threshold for that business scenario tag using a formula based on the second anomaly score statistical feature value. Wherein, the business scenario tag... Second abnormal score statistical mean and business scenario tags Second outlier score statistical standard deviation This can be obtained from S502. Adjustment coefficient. The threshold is used to control the strictness of the threshold. The default setting is 2, which means that the second abnormal threshold is set at the mean plus two standard deviations. It can also be set according to the actual work. This embodiment does not further limit the setting of this adjustment coefficient.

[0133] S504, based on the second anomaly threshold of each business scenario label, obtain the second anomaly threshold data.

[0134] Specifically, the terminal integrates the second anomaly thresholds of each business scenario label to obtain the second anomaly threshold data.

[0135] This embodiment provides an intelligent accounting data processing method that collects sufficient historical anomaly score samples and calculates the statistical characteristics of the score distribution for each business scenario label. It also utilizes statistical principles and adjustment coefficients to generate a second anomaly threshold for each business scenario label, which is then integrated into usable second anomaly threshold data. This elevates the most critical judgment threshold in anomaly detection from a static, empirically set value to a dynamically optimized value driven by historical data. This allows the anomaly judgment criteria to automatically adapt to the data characteristics of different business scenarios, thereby improving the accuracy and reliability of accounting data anomaly detection across various business scenarios.

[0136] In one embodiment, based on business scenario rules and accounting data sequences, a second abnormal dataset is comprehensively analyzed to obtain an anomaly analysis report, including:

[0137] S601 quantifies the abnormal distribution of each business scenario label based on business scenario rules, accounting data sequences, and a second abnormal dataset, thereby obtaining the risk index of each business scenario label.

[0138] Specifically, the terminal determines the business scenario label for each accounting data in the accounting data sequence based on business scenario rules, obtains a subset of accounting data for each business scenario label, and combines the business scenario label and the second anomaly score in the second anomaly dataset using the following formula: The risk index for each business scenario label is calculated by combining the preset business scenario label weights. The business scenario is tagged as The risk index, The label of any business scenario in the second abnormal dataset is Accounting data, yes The second abnormal score, Business scenario tags The weight, Business scenario tags The number of accounting data in the accounting data subset. For example, the preset business scenario label weights can be set according to the importance of the business scenario in actual work; this embodiment does not further limit the setting of these preset business scenario label weights.

[0139] S602, based on the abnormal business label set in the second abnormal data set, quantify the distribution of each abnormal business label in the abnormal business label set, and obtain the risk score of each abnormal business label.

[0140] Specifically, the terminal calculates the risk score for each abnormal service label by statistically analyzing the frequency of each abnormal service label appearing in the dataset and its associated second abnormality score level.

[0141] S603, input the abnormal business label set of each accounting data in the second abnormal dataset into the preset response action prediction model to obtain the processing action data of each accounting data.

[0142] Specifically, the terminal inputs the abnormal business tag set of each abnormal accounting data in the second abnormal dataset into the preset response action prediction model, and automatically generates the corresponding processing action data. The response action prediction model is a machine learning model trained based on historical processing experience. It recommends the most suitable processing solution for new abnormal accounting data by analyzing the mapping relationship between abnormal business tag combinations and the final processing actions taken.

[0143] S604 generates an anomaly analysis report based on the business scenario labels, processing action data, second anomaly scores, risk indices of business scenario labels, and risk scores of anomaly business labels for each accounting data in the second anomaly dataset.

[0144] Specifically, the terminal uses a pre-set structured template database to convert the business scenario tags, processing action data, second anomaly scores, risk indices of business scenario tags, and risk scores of anomaly business tags for each accounting data item in the second anomaly dataset into a structured anomaly analysis report. The pre-set structured template database can be configured according to the actual report template requirements; this embodiment does not further limit the configuration of this pre-set structured template database. For example, the pre-set structured template database can define the order and format of the execution summary, detailed analysis, processing suggestions, and data appendix in the anomaly analysis report. For instance, the execution summary highlights the most critical risk findings; the detailed analysis displays the risk quantification results for each scenario and anomaly type; the processing suggestions provide specific action plans based on predicted processing action data; and the data appendix provides detailed anomaly record information.

[0145] This embodiment provides an intelligent accounting data processing method that identifies high-risk areas by integrating abnormal information from business scenario tags. It also analyzes specific risk sources from the perspective of abnormal business tags and recommends processing solutions for each abnormal record based on an intelligent model. All analysis results are integrated into an abnormal analysis report. This successfully elevates scattered anomaly detection results into a systematic risk management insight, improving the accuracy and reliability of accounting data anomaly detection in different business scenarios. Simultaneously, it provides data-driven decision-making support for business process optimization and internal control system improvement.

[0146] In one embodiment, the response action prediction model is obtained through the following method:

[0147] S701, obtain preset historical response action data; historical response action data includes abnormal business tags of accounting data in historical processing and corresponding processing action data.

[0148] Specifically, the terminal acquires preset historical response action data; this data includes abnormal business tags for accounting data during historical processing and corresponding processing action data. The processing action data represents the historical handling measures for the abnormal business tags. The preset historical response action data can be obtained from historical records of abnormal accounting data processing in actual work; this embodiment does not further limit the setting of this preset historical response action data.

[0149] S702, using the abnormal business tags of accounting data in the historical processing process as feature values ​​and the corresponding processing action data as feature attributes, obtains historical response action mapping data, and divides the historical response action mapping data into training sample set and test sample set according to a preset ratio.

[0150] Specifically, the terminal uses the abnormal business tags of accounting data in the historical processing process as feature values ​​and the corresponding processing action data as feature attributes to form historical response action mapping data in the form of feature value-feature attribute pairs. The historical response action mapping data is then divided into a training sample set and a test sample set according to a preset ratio. For example, the preset ratio is set by default to a training sample set: test sample set ratio of 8:2, and this embodiment does not further limit the setting of this preset ratio.

[0151] S703, randomly select feature values ​​and combinations to construct decision tree models respectively, and construct a preliminary response action prediction model based on multiple decision tree models.

[0152] Specifically, the terminal constructs a preliminary action prediction model using an ensemble learning algorithm. This step employs the random forest algorithm as the basic modeling framework. Random forest is an ensemble learning algorithm based on decision trees, which improves the accuracy and stability of the model by constructing multiple decision trees and combining their prediction results. The specific implementation process includes first generating multiple sub-training sets through random sampling with replacement from the training sample set, then constructing a decision tree for each sub-training set, randomly selecting some features for node splitting during the decision tree construction process, and finally combining multiple decision trees into a random forest model to obtain the preliminary action prediction model.

[0153] S704. The initial response action prediction model is trained using the training sample set, and the trained response action prediction model is tested using the test sample set until the response action prediction model meets the preset accuracy requirements, thus obtaining the response action prediction model.

[0154] Specifically, the terminal uses a training sample set to train the parameters of an initial model, optimizing the model's internal parameters by minimizing the difference between the predicted and actual actions. Then, a test sample set is used to evaluate the accuracy of the trained model until the action prediction model meets a preset accuracy requirement, thus obtaining the action prediction model. Illustratively, the preset accuracy requirement is 90% accuracy, but this can be adjusted based on actual work requirements.

[0155] This embodiment provides an intelligent accounting data processing method that, through four key steps—historical data acquisition, feature mapping construction, integrated model training, and iterative optimization and verification—effectively transforms historical processing experience into intelligent predictive capabilities, resulting in an action prediction model. It successfully elevates anomaly handling decisions, which rely on human experience, to data-driven intelligent recommendations, significantly improving anomaly handling efficiency while reducing dependence on specific experts.

[0156] The aforementioned intelligent accounting data processing method involves acquiring the target company's accounting data sequence. This sequence includes at least the execution timestamp, execution amount, and execution account type. Based on preset business scenario rules and preset benchmark distribution data, anomalies in the accounting data sequence are analyzed to obtain a first abnormal dataset. This first abnormal dataset includes at least accounting data, business scenario tags for the accounting data, and a first anomaly score. Based on a preset set of business rules, the first abnormal dataset is filtered to obtain a second abnormal dataset. The business rule set is used to indicate and determine the business requirement tags for the accounting data. The second abnormal dataset includes at least abnormal accounting data, business scenario tags for the abnormal accounting data, and a second anomaly score. Based on the business scenario rules and the accounting data sequence, the second abnormal dataset is comprehensively analyzed to obtain an anomaly analysis report. This anomaly analysis report characterizes the degree of anomaly in the accounting data sequence. This successfully transforms the traditional accounting auditing model, which relies on fixed thresholds and human experience for delayed detection, into an intelligent detection paradigm that dynamically perceives business context, integrates statistical learning and symbolic rules, and possesses continuous learning capabilities. This improves the accuracy and practicality of accounting data anomaly detection in complex and ever-changing business environments.

[0157] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0158] Based on the same inventive concept, this application also provides an intelligent accounting data processing device for implementing the intelligent accounting data processing method described above. The solution provided by this device is similar to the implementation scheme described in the above method; therefore, the specific limitations in one or more intelligent accounting data processing device embodiments provided below can be found in the limitations of the intelligent accounting data processing method described above, and will not be repeated here.

[0159] In one exemplary embodiment, such as Figure 2 As shown, an intelligent accounting data processing device 200 is provided, comprising:

[0160] The data acquisition module 201 is used to acquire the accounting data sequence of the target company; the accounting data sequence includes at least the execution timestamp, execution amount, and execution account type of the accounting data.

[0161] The first anomaly screening module 202 is used to analyze the anomalies in the accounting data sequence based on preset business scenario rules and preset benchmark distribution data to obtain a first anomaly dataset; the first anomaly dataset includes at least accounting data, business scenario tags of the accounting data, and a first anomaly score;

[0162] The second anomaly filtering module 203 is used to filter the first anomaly dataset based on a preset set of business rules to obtain a second anomaly dataset; the set of business rules is used to indicate the business requirement labels for determining accounting data; the second anomaly dataset includes at least the abnormal accounting data, the business scenario labels of the abnormal accounting data, and the second anomaly score.

[0163] The report generation module 204 is used to comprehensively analyze the second abnormal dataset based on business scenario rules and accounting data sequences to obtain an anomaly analysis report; the anomaly analysis report is used to characterize the degree of anomaly in the accounting data sequence.

[0164] Furthermore, the device also includes a baseline distribution data determination module, used for:

[0165] Obtain the preset historical accounting dataset; the historical accounting dataset includes accounting data from the historical accounting data processing process and corresponding business scenario tags;

[0166] Based on historical accounting datasets, the distribution relationship between execution amounts and corresponding business scenario labels is quantified to obtain amount distribution parameters; the amount distribution parameters are used to characterize the probability distribution characteristics of execution amounts under each business scenario label.

[0167] Based on historical accounting datasets, time-series features are extracted from execution timestamps and corresponding business scenario labels to obtain time distribution parameters. These time distribution parameters are used to characterize the time distribution features of execution timestamps under each business scenario label.

[0168] Based on historical accounting datasets, the relationship between execution account types and corresponding business scenario labels is quantified to obtain account association rules; account association rules are used to characterize the association features between execution account types under each business scenario label;

[0169] Based on the amount distribution parameters, time distribution parameters, and account association rules, the baseline distribution data is obtained.

[0170] Furthermore, the first anomaly filtering module is also used for:

[0171] Based on accounting data sequences and business scenario rules, business scenario labels are obtained for each accounting data; business scenario rules are used to characterize the mapping relationship between business scenario labels and execution account types.

[0172] Based on the business scenario labels and the amount distribution parameters in the benchmark distribution data, the abnormality of the execution amount of each accounting data in the accounting database column is quantified to obtain the amount abnormality feature score of each accounting data.

[0173] Based on the time pattern distribution parameters in the business scenario labels and benchmark distribution data, the expected execution timestamp of each accounting data is determined, and the difference between the expected execution timestamp and the execution timestamp is quantified to obtain the time anomaly characteristic score of each accounting data.

[0174] Based on the account association rules in business scenario tags and benchmark distribution data, the abnormal situation of the execution account type of each accounting data in the accounting database column is quantified to obtain the account association abnormal feature score of each accounting data.

[0175] Based on the abnormal characteristic scores of monetary amount, time, and account association, the first abnormality score for each accounting data is calculated using the following formula:

[0176]

[0177] in, It is accounting data The first abnormality score, It is the weight of abnormal amounts. It is accounting data The abnormal feature score of the amount, It is the time anomaly weight. It is accounting data Temporal anomaly feature score, It is the subject association weight. It is accounting data Subject-related abnormal feature scores;

[0178] The accounting data with a first anomaly score greater than a preset first anomaly threshold are selected from the accounting database column to obtain the first anomaly dataset.

[0179] Furthermore, the set of business rules also includes the business data conditions, penalty weights, and importance coefficients for each business requirement tag; business requirement tags include normal business tags and abnormal business tags;

[0180] The second anomaly filtering module is also used for:

[0181] For each accounting data point in the first abnormal dataset, the following formula is used to match the accounting data with the business data conditions of each normal business label to obtain the matching label value of the accounting data:

[0182]

[0183] in, It is the first Matching tag values ​​for accounting data It is any normal business tag in the set of business rules. Business data conditions, It is a matching indicator function. It is the first One accounting data;

[0184] Based on the preset anomaly matching label threshold, accounting data whose matching label values ​​meet the anomaly matching label threshold are selected from the first anomaly dataset to obtain the business anomaly dataset;

[0185] For each accounting data in the business anomaly dataset, the accounting data is matched with the business data conditions corresponding to each anomaly business tag to obtain the anomaly business matching result for each accounting data; and based on the anomaly business matching result, anomaly business tags are selected to obtain the anomaly business tag set for each accounting data.

[0186] For each accounting data in the business anomaly dataset, based on the abnormal business tag set of the accounting data, the difference between the accounting data and the business data conditions of each abnormal business tag is quantified to obtain the blacklist score of each accounting data.

[0187] For each accounting data in the business anomaly dataset, a second anomaly score is obtained based on the blacklist score and the first anomaly score.

[0188] Based on the business scenario tags of accounting data and the preset second abnormal threshold data, the second abnormal threshold of accounting data is determined, and accounting data with a second abnormal score greater than the second abnormal threshold is identified as abnormal accounting data.

[0189] The second abnormal dataset is obtained based on the second abnormal score of abnormal accounting data, business scenario labels, and abnormal business label set.

[0190] Furthermore, the device also includes a second abnormal threshold data calculation module, used for:

[0191] Obtain preset historical second anomaly score data; the historical second anomaly score data includes the second anomaly scores of historical accounting data for each business scenario tag;

[0192] For each business scenario tag, the statistical features of the second anomaly score are extracted from the historical accounting data to obtain the statistical feature value of the second anomaly score for each business scenario tag; the statistical feature value of the second anomaly score includes the statistical mean of the second anomaly score and the statistical standard deviation of the second anomaly score.

[0193] For each business scenario tag, the second anomaly score statistical feature value is used to calculate the second anomaly threshold of the business scenario tag using the following formula:

[0194]

[0195] in, It can be a tag for any business scenario. Business scenario tags The second abnormal threshold, Business scenario tags The statistical mean of the second abnormal score, It is an adjustment factor. Business scenario tags The statistical standard deviation of the second outlier score;

[0196] The second anomaly threshold data is obtained based on the second anomaly threshold of each business scenario label.

[0197] Furthermore, the report generation module is also used for:

[0198] Based on business scenario rules, accounting data sequences, and a second abnormal dataset, the abnormal distribution of each business scenario label is quantified to obtain the risk index of each business scenario label.

[0199] Based on the abnormal business label set in the second abnormal data set, the distribution of each abnormal business label in the abnormal business label set is quantified to obtain the risk score of each abnormal business label.

[0200] Input the abnormal business label set of each accounting data in the second abnormal dataset into the preset response action prediction model to obtain the processing action data of each accounting data.

[0201] An anomaly analysis report is generated based on the business scenario labels, processing action data, second anomaly scores, risk indices of business scenario labels, and risk scores of anomalous business labels for each accounting data in the second anomaly dataset.

[0202] Furthermore, the device also includes a model building module for:

[0203] Acquire preset historical response action data; historical response action data includes abnormal business tags of accounting data in historical processing and corresponding processing action data;

[0204] Using the abnormal business tags of accounting data in the historical processing process as feature values ​​and the corresponding processing action data as feature attributes, historical response action mapping data is obtained, and the historical response action mapping data is divided into training sample set and test sample set according to a preset ratio.

[0205] Decision tree models are constructed by randomly selecting feature values ​​and combinations, and a preliminary response action prediction model is built based on multiple decision tree models.

[0206] The initial response action prediction model is trained using a training sample set, and the trained response action prediction model is tested using a test sample set until the response action prediction model meets the preset accuracy requirements, thus obtaining the response action prediction model.

[0207] In one embodiment, such as Figure 3 A computer device is provided, comprising:

[0208] At least one processor 301, and a memory 302 communicatively connected to at least one of the processors 301: the memory stores application code executable by at least one of the processors, the application code being executed by at least one of the processors to enable at least one of the processors to perform the intelligent accounting data processing method as described above.

[0209] Computer equipment may also include: sensor 303.

[0210] The processor 301, memory 302 and sensor 303 can be connected via a bus or other means, with the bus being an example in the figure.

[0211] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0212] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0213] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. An intelligent accounting data processing method, characterized by, The method comprises: obtaining an accounting data sequence of a target company; the accounting data sequence at least comprises an execution timestamp, an execution amount and an execution subject type of accounting data; based on a preset business scenario rule and a preset reference distribution data, analyzing the abnormal situation of the accounting data sequence to obtain a first abnormal data set; the first abnormal data set at least comprises the accounting data, a business scenario label of the accounting data and a first abnormal score; based on a preset business rule set, the first abnormal data set is filtered to obtain a second abnormal data set; the business rule set is used to indicate the determination of the business demand label of the accounting data; the second abnormal data set at least comprises abnormal accounting data, a business scenario label of the abnormal accounting data and a second abnormal score; based on the business scenario rule and the accounting data sequence, the second abnormal data set is comprehensively analyzed to obtain an abnormal analysis report; the abnormal analysis report is used to represent the abnormal degree of the accounting data in the accounting data sequence.

2. The method of claim 1, wherein, The reference distribution data is obtained by the following method: obtaining a preset historical accounting data set; the historical accounting data set comprises the accounting data of the historical accounting data processing process and the corresponding business scenario label; based on the historical accounting data set, the distribution relationship between the execution amount and the corresponding business scenario label is quantified to obtain an amount distribution parameter; the amount distribution parameter is used to represent the probability distribution characteristics of the execution amount under each business scenario label; based on the historical accounting data set, the execution timestamp and the corresponding business scenario label are extracted for time sequence characteristics to obtain a time distribution parameter; the time distribution parameter is used to represent the time distribution characteristics of the execution timestamp under each business scenario label; based on the historical accounting data set, the association relationship between the execution subject type and the corresponding business scenario label is quantified to obtain a subject association rule; the subject association rule is used to represent the association characteristics between the execution subject types under each business scenario label; based on the amount distribution parameter, the time distribution parameter and the subject association rule, the reference distribution data is obtained.

3. The method of claim 2, wherein, Based on the preset business scenario rule and the preset reference distribution data, the abnormal situation of the accounting data sequence is analyzed to obtain a first abnormal data set, comprising: based on the accounting data sequence and the business scenario rule, the business scenario label of each accounting data is obtained; the business scenario rule is used to represent the mapping relationship between the business scenario label and the execution subject type; based on the business scenario label and the amount distribution parameter in the reference distribution data, the abnormal situation of the execution amount of each accounting data in the accounting database column is quantified to obtain the amount abnormal feature score of each accounting data; determine an expected execution timestamp of each of the accounting data based on the business scenario label and the time regularity distribution parameter in the benchmark distribution data, and quantify a difference between the expected execution timestamp and the execution timestamp to obtain a time anomaly feature score of each of the accounting data; quantify an abnormal case of the execution subject type of each of the accounting data in the accounting database column based on the subject association rule in the business scenario label and the benchmark distribution data to obtain a subject association anomaly feature score of each of the accounting data; based on the amount anomaly feature score, the time anomaly feature score, and the subject association anomaly feature score, calculate a first anomaly score of each of the accounting data using the following formula: wherein, is a first anomaly score of the accounting data , is an amount anomaly weight, is an amount anomaly feature score of the accounting data , is a time anomaly weight, is a time anomaly feature score of the accounting data , is a subject association weight, is a subject association anomaly feature score of the accounting data ; select the accounting data with the first anomaly score greater than a preset first anomaly threshold from the accounting database column to obtain a first anomaly data set.

4. The method of claim 1, wherein, The business rule set further includes a business data condition, a penalty weight, and an importance coefficient of each of the business requirement labels; the business requirement labels include normal business labels and abnormal business labels; The filtering of the first anomaly data set based on the preset business rule set to obtain a second anomaly data set includes: For each of the accounting data in the first anomaly data set, match the accounting data with the business data condition of each of the normal business labels using the following formula to obtain a matching label value of the accounting data: wherein, is a matching tag value of the accounting data, is a business data condition of any normal business tag in the business rule set, is a matching indication function, is the accounting data; select the accounting data with a matching label value meeting a preset anomaly matching label threshold from the first anomaly data set to obtain a business anomaly data set; For each of the accounting data in the business anomaly data set, match the accounting data with the business data condition corresponding to each of the abnormal business labels to obtain an abnormal business matching result of each of the accounting data; and select the abnormal business label based on the abnormal business matching result to obtain an abnormal business label set of each of the accounting data; For each of the accounting data in the business anomaly data set, quantify a difference between the accounting data and the business data condition of each of the abnormal business labels based on the abnormal business label set of the accounting data to obtain a blacklist score of each of the accounting data; For each of the accounting data in the business anomaly data set, obtain a second anomaly score of each of the accounting data based on the blacklist score and the first anomaly score; determine a second anomaly threshold of the accounting data based on the business scenario label of the accounting data and a preset second anomaly threshold data, and determine the accounting data with a second anomaly score greater than the second anomaly threshold as the abnormal accounting data; obtain the second anomaly data set based on the second anomaly score, the business scenario label, and the abnormal business label set of the abnormal accounting data.

5. The method of claim 4, wherein, The second anomaly threshold data is obtained by the following method: obtaining preset historical second anomaly score data; the historical second anomaly score data includes the second anomaly score of historical accounting data of each business scenario label; extracting statistical features of the second anomaly score for the historical accounting data of each business scenario label to obtain second anomaly score statistical feature values of each business scenario label; the second anomaly score statistical feature values include a second anomaly score statistical mean and a second anomaly score statistical standard deviation; for the second anomaly score statistical feature values of each business scenario label, the second anomaly threshold of the business scenario label is calculated using the following formula: wherein, is any business scenario label, is a business scenario label is a second abnormal threshold value of the business scenario label, is a business scenario label is a second abnormal score statistical mean of the business scenario label, is an adjustment coefficient, is a business scenario label is a second abnormal score statistical standard deviation of the business scenario label; obtaining the second anomaly threshold data based on the second anomaly threshold of each business scenario label.

6. The method of claim 4, wherein, comprehensive analysis of the second anomaly data set based on the business scenario rule and the accounting data sequence to obtain an anomaly analysis report, including: quantifying the anomaly distribution of each business scenario label based on the business scenario rule, the accounting data sequence, and the second anomaly data set to obtain a risk index of each business scenario label; quantifying the distribution of each abnormal business label in the abnormal business label set in the second anomaly data set to obtain a risk score of each abnormal business label; inputting the abnormal business label set of each accounting data in the second anomaly data set into a preset coping action prediction model to obtain processing action data of each accounting data; generating an anomaly analysis report based on the business scenario label, the processing action data, the second anomaly score of each accounting data in the second anomaly data set, and the risk index of the business scenario label and the risk score of the abnormal business label.

7. The method of claim 6, wherein, The coping action prediction model is obtained by the following method: obtaining preset historical coping action data; the historical coping action data includes the abnormal business label of the accounting data in the historical processing process and the corresponding processing action data; taking the abnormal business label of the accounting data in the historical processing process as a feature value and taking the corresponding processing action data as a feature attribute to obtain historical coping action mapping data, and dividing the historical coping action mapping data into a training sample set and a test sample set according to a preset proportion; randomly selecting the feature values and combinations to construct decision tree models, and constructing a preliminary coping action prediction model according to multiple decision tree models; training the preliminary coping action prediction model using the training sample set, and testing the trained coping action prediction model using the test sample set until the coping action prediction model meets the preset accuracy requirement to obtain the coping action prediction model.

8. An intelligent accounting data processing apparatus characterized by comprising: The device comprises: a data acquisition module for acquiring an accounting data sequence of a target company; the accounting data sequence includes at least an execution timestamp, an execution amount, and an execution subject type of accounting data; The first anomaly screening module is configured to analyze an abnormal situation of the accounting data sequence based on a preset business scenario rule and a preset reference distribution data, and obtain a first abnormal data set; the first abnormal data set at least includes the accounting data, a business scenario label of the accounting data, and a first abnormal score; The second anomaly screening module is configured to screen the first abnormal data set based on a preset business rule set, and obtain a second abnormal data set; the business rule set is used to indicate a business demand label of the accounting data; the second abnormal data set at least includes an abnormal accounting data, a business scenario label of the abnormal accounting data, and a second abnormal score; The report generation module is configured to comprehensively analyze the second abnormal data set based on the business scenario rule and the accounting data sequence, and obtain an abnormal analysis report; the abnormal analysis report is used to represent an abnormal degree of the accounting data in the accounting data sequence. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-8 when the computer program is executed by the processor. The processor executes the computer program to implement the steps of the method in any one of claims 1 to 7.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method in any one of claims 1 to 7.

Images