Data anomaly analysis method, computer equipment and computer storage medium

Abstract

The embodiment of the invention discloses a data exception analysis method, computer equipment and a computer storage medium. The computer equipment determines abnormal fluctuation index results conforming to abnormal fluctuation characteristics in the fluctuation index results according to the data characteristics of the fluctuation index results of the business data, and determines influence factors contributing to the abnormal fluctuation index results according to the business data; and generating abnormal fluctuation explanation of the business data according to the fluctuation explanation knowledge, the abnormal fluctuation index result and the related data of the influence factors. Excessive dependence on artificial experience is avoided, time and energy of business personnel such as financial personnel are greatly saved, and labor cost is reduced. The method is simple in operation, is not influenced by personal experience, careful degree and working state, can guarantee the accuracy and consistency of analysis results, can be operated efficiently and stably even in the time period of huge data volume or busy business, effectively avoids analysis omission or judgment errors caused by personnel fatigue or negligence, and remarkably improves the analysis efficiency.

Description

Technical Field

[0001] This application relates to the field of data analysis, specifically to a data anomaly analysis method, computer equipment, and computer storage medium. Background Technology

[0002] Before closing the general ledger, it is necessary to analyze the accounts with abnormal fluctuations in transaction amounts and balances to find the reasons for the abnormal fluctuations in account amounts.

[0003] Currently, most financial management products on the market rely on manual methods to identify abnormal fluctuations, find their causes, and then manually interpret the discrepancies. However, this method is entirely dependent on human experience, requires a significant amount of time from financial personnel, and has high labor costs. Furthermore, the quality of manual analysis is highly dependent on individual experience, attentiveness, and work status. During periods of large data volume or busy times at the end of the month or year, it is easy for personnel fatigue or negligence to lead to omissions or errors in analysis, making it impossible to guarantee analysis efficiency. Summary of the Invention

[0004] This application provides a data anomaly analysis method, computer equipment, and computer storage medium to solve the problems of low efficiency, high cost, and difficulty in guaranteeing accuracy in manual analysis of abnormal data fluctuations in related technical solutions.

[0005] A first aspect of this application provides a data anomaly analysis method, the method comprising:

[0006] Obtain the fluctuation index results of the business data, which are used to measure the fluctuation of the business data;

[0007] Based on the data characteristics of the volatility index results, identify the abnormal volatility index results that conform to the abnormal volatility characteristics.

[0008] Based on the business data, determine the influencing factors that contribute to the results of the abnormal fluctuation index;

[0009] Obtain fluctuation interpretation knowledge, and generate an abnormal fluctuation interpretation of the business data based on the fluctuation interpretation knowledge, the abnormal fluctuation index results, and the relevant data of the influencing factors.

[0010] A second aspect of this application provides a computer device, the computer device comprising:

[0011] An acquisition unit is used to acquire fluctuation index results of business data, wherein the fluctuation index results are used to measure the fluctuation of the business data.

[0012] The first determining unit is used to determine, based on the data characteristics of the fluctuation index results, the abnormal fluctuation index results that conform to the abnormal fluctuation characteristics.

[0013] The second determining unit is used to determine, based on the business data, the influencing factors that contribute to the result of the abnormal fluctuation index.

[0014] The generation unit is used to acquire fluctuation interpretation knowledge and generate an abnormal fluctuation interpretation of the business data based on the fluctuation interpretation knowledge, the abnormal fluctuation index results, and the relevant data of the influencing factors.

[0015] A third aspect of this application 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 the method of the first aspect described above.

[0016] A fourth aspect of this application provides a computer storage medium storing instructions that, when executed on a computer, cause the computer to perform the method described in the first aspect.

[0017] A fifth aspect of this application provides a computer program product that, when run on a computer device, causes the computer device to perform the method described in the first aspect.

[0018] As can be seen from the above technical solutions, the embodiments of this application have the following advantages:

[0019] Compared to traditional manual analysis methods, this approach eliminates excessive reliance on human experience, significantly saving time and energy for finance and other business personnel, and reducing labor costs. Furthermore, because the entire analysis process is automated using computer equipment, it is unaffected by individual experience, attentiveness, or work status, ensuring the accuracy and consistency of the results. Even with large datasets or during peak business periods, it operates efficiently and stably, effectively avoiding omissions or errors due to human fatigue or negligence, thus significantly improving analytical efficiency. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the network framework in an embodiment of this application;

[0021] Figure 2 This is a flowchart illustrating the data anomaly analysis method in an embodiment of this application;

[0022] Figure 3 This is an exemplary schematic diagram illustrating the architecture and process of the data anomaly analysis method in the application scenario of financial data fluctuation anomaly analysis in the embodiments of this application;

[0023] Figure 4 This is an exemplary content diagram illustrating the knowledge of fluctuation interpretation in the embodiments of this application;

[0024] Figure 5 This is a schematic diagram of the structure of a computer device in an embodiment of this application;

[0025] Figure 6 This is another schematic diagram of the structure of the computer device in the embodiments of this application. Detailed Implementation

[0026] This application provides a data anomaly analysis method, computer equipment, and computer storage medium to solve the problems of low efficiency, high cost, and difficulty in guaranteeing accuracy in manual analysis of abnormal data fluctuations in related technical solutions.

[0027] Please see Figure 1 The network framework in this embodiment includes:

[0028] The business server 100 and the terminal cluster; the terminal cluster may include: terminal devices 200a, terminal devices 200b, terminal devices 200c, ..., terminal devices 200n and other terminal devices.

[0029] The aforementioned business server 100 can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud databases, cloud services, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The terminal devices (including terminal devices 200a, 200b, 200c, ..., 200n) can be smartphones, tablets, laptops, desktop computers, PDAs, mobile internet devices (MIDs), wearable devices (such as smartwatches, smart bracelets, etc.), smart computers, smart in-vehicle systems, and other intelligent terminals.

[0030] The service server 100 can establish communication connections with each terminal device in the terminal cluster, and the terminal devices in the terminal cluster can also establish communication connections with each other. In other words, the service server 100 can establish communication connections with each terminal device among terminal devices 200a, 200b, 200c, ..., 200n. For example, terminal device 200a can establish a communication connection with the service server 100. Terminal devices 200a and 200b can establish a communication connection, and terminal devices 200a and 200c can also establish a communication connection. The communication connection method is not limited; it can be established directly or indirectly through wired communication or wireless communication, etc., depending on the actual application scenario. This application does not impose any restrictions on this.

[0031] It should be understood that, such as Figure 1 Each terminal device in the terminal cluster shown can have an application client installed. When the application client runs on each terminal device, it can interact with the business server 100, allowing the business server 100 to receive business data from each terminal device (such as financial management data uploaded by users through the terminal device). This application client can be a financial management application, enterprise affairs management application, browser application, social application, instant messaging application, live streaming application, game application, short video application, video application, music application, shopping application, novel application, payment application, or any other application client with the ability to display text, images, audio, and video data. The specific application client can be determined based on the actual application scenario requirements and is not limited here. This application client can be a standalone client or an embedded sub-client integrated into a client (such as a financial management client, enterprise affairs management client, etc.), depending on the actual application scenario and is not limited here.

[0032] The following will combine Figure 1 The network framework described in this application describes the data anomaly analysis method in the embodiments of the present application:

[0033] Please see Figure 2 One embodiment of the data anomaly analysis method in this application includes:

[0034] 201. Obtain the fluctuation index results of the business data, wherein the fluctuation index results are used to measure the fluctuation of the business data;

[0035] The method of this embodiment can be applied to a computer device, which may be... Figure 1 The network framework shown includes a business server 100 or various terminal devices. Computer devices can obtain fluctuation index results for business data, which can be data from financial accounts. For example, in one application scenario of this embodiment, before closing the general ledger, it is necessary to analyze accounts with abnormal fluctuations in transaction amounts and balances. In this case, fluctuation index results for the account data can be obtained to perform anomaly analysis of data fluctuations based on these fluctuation index results.

[0036] In this context, "account" refers to the basic classification items in financial accounting, used to record and categorize various economic transactions of an enterprise, such as asset accounts, liability accounts, owner's equity accounts, cost accounts, and profit and loss accounts.

[0037] After obtaining the fluctuation index results of the business data, the results can intuitively reflect the fluctuation of the business data in different periods or compared with the average level of a specific range, providing basic data support for subsequent anomaly analysis.

[0038] Of course, besides scenarios involving financial data, this embodiment can also be applied to other scenarios. For example, in e-commerce operations, when analyzing fluctuations in product sales data, fluctuation index results can be obtained to determine whether fluctuations in product sales over different time periods or compared to the average sales level of similar products are abnormal. In industrial production scenarios, fluctuation index results can also be obtained for the operating parameter data of production equipment, such as the fluctuations of parameters like equipment temperature and pressure at different production stages or compared to the average value within the normal production range, thus providing a basis for subsequently determining whether abnormal fluctuations exist. By obtaining fluctuation index results for business data, regardless of the business scenario, a foundation can be laid for accurately analyzing data anomalies.

[0039] 202. Based on the data characteristics of the fluctuation index results, determine the abnormal fluctuation index results that meet the abnormal fluctuation characteristics among the fluctuation index results;

[0040] After obtaining the fluctuation index results of the business data, abnormal fluctuation index results can be filtered out, that is, abnormal fluctuation index results can reflect the abnormal fluctuations of the data.

[0041] Specifically, various methods can be used to determine abnormal fluctuation indicators. For example, a threshold range can be set; when a fluctuation indicator exceeds this range, it is considered an abnormal fluctuation indicator result. This threshold range can be set based on historical data, industry experience, or business needs.

[0042] Additionally, pre-trained language models can be used to identify abnormal fluctuation index results. These models are equipped with anomaly detection knowledge; by inputting the fluctuation index results and anomaly detection prompts, the model can automatically identify the abnormal parts of the fluctuation index results based on its built-in anomaly detection knowledge and prompts, and then output these abnormal fluctuation index results. This method leverages the advantages of artificial intelligence technology, enabling more efficient and accurate anomaly detection tasks.

[0043] In addition, statistical analysis methods can be used to identify abnormal fluctuation index results. For example, probability distribution analysis can be performed on the fluctuation index results, assuming that the data follows a certain distribution (such as a normal distribution), and then a confidence interval can be set based on the characteristics of the distribution. Fluctuation index results falling outside the confidence interval are considered abnormal. Alternatively, cluster analysis can be performed on the fluctuation index results, grouping similar fluctuation index results into one category, while those fluctuation index results that are significantly different from other categories may be abnormal.

[0044] Regardless of the method used, the principle for identifying abnormal fluctuation indicators is based on setting corresponding judgment criteria according to the characteristics and patterns of business data fluctuations, and then filtering out fluctuation indicator results that meet the abnormal characteristics. These abnormal fluctuation indicator results often contain potential problems or anomalies in the business data and are the focus of subsequent in-depth analysis.

[0045] 203. Based on the business data, determine the influencing factors that contribute to the results of the abnormal fluctuation index;

[0046] After determining the abnormal fluctuation indicators, the next step is to investigate the causes of the anomalies. This step is crucial for understanding the abnormal fluctuations in business data; only by finding the root cause can effective measures be taken to address and resolve the issue. In this step, the influencing factors contributing to the abnormal fluctuation indicators can be identified based on the business data. This involves comprehensively collecting multiple detailed records from the business data. These records cover all aspects of business activities and contain key information that may affect abnormal data fluctuations. For example, in a financial scenario, detailed records might include information such as the time, amount, and transaction counterparty for each transaction; in an e-commerce operation scenario, detailed records might involve the sales time, quantity, and region of the goods.

[0047] Therefore, the detailed records in the business data can be analyzed to identify the detailed records that cause abnormal data fluctuations, thereby identifying the influencing factors that contribute to the abnormal fluctuation index results.

[0048] 204. Obtain fluctuation interpretation knowledge, and generate an abnormal fluctuation interpretation of the business data based on the fluctuation interpretation knowledge, the abnormal fluctuation index results, and the relevant data of the influencing factors;

[0049] This knowledge of fluctuation interpretation refers to a set of professional knowledge used to explain the causes of abnormal fluctuations in business data. It covers abnormal fluctuations that may occur in various business scenarios and their corresponding causes. This knowledge can come from historical data summaries, industry experience accumulation, expert research results, etc. For example, in the financial field, fluctuation interpretation knowledge can include the impact of different accounting policy changes on account data, the correlation between specific economic events (such as corporate operations, market fluctuations, policy adjustments) and abnormal fluctuations in financial data, etc. In e-commerce operations, it can include knowledge of the patterns of abnormal fluctuations in promotional activities and sales data, the impact of seasonal factors on product sales, etc.

[0050] The relevant data on influencing factors refers to the detailed data closely related to these factors that are collected after the factors contributing to the abnormal fluctuation indicators have been identified. This data provides a specific quantification or description of the influencing factors, reflecting their role in the abnormal fluctuations of business data more directly and clearly.

[0051] Taking a financial scenario as an example, if it is determined that the abnormal fluctuation of a certain account is caused by a large transaction with a specific trading partner, then the relevant data on the influencing factors will include detailed information about that trading partner, such as the name of the trading partner, the transaction time, the transaction amount, the nature of the transaction (whether it is a sale or a purchase, etc.), and a description of the transaction background. This data can help to deeply analyze the specific degree and manner in which the trading partner's transaction behavior affects the abnormal fluctuation of the account data.

[0052] Therefore, after acquiring the knowledge of fluctuation interpretation, the computer equipment combines the results of abnormal fluctuation indicators with relevant data on influencing factors, and uses this knowledge for intelligent analysis and reasoning. It matches cases, templates, or rules similar to the current abnormal fluctuation situation from the fluctuation interpretation knowledge base, analyzes the fluctuation characteristics reflected by the abnormal fluctuation indicator results and the role played by influencing factors, and then generates accurate and detailed explanations of abnormal fluctuations in business data.

[0053] For example, if the abnormal fluctuation indicator shows a significant increase in the amount of a certain financial item within a certain period, and the influencing factor analysis reveals that it is caused by a few large transactions, then by combining the rules regarding the impact of large transactions on item data in the fluctuation interpretation knowledge, the generated fluctuation interpretation will clearly indicate that the abnormal increase in the amount of that item is due to these specific large transactions. In this way, business personnel can be provided with a clear and easy-to-understand explanation of the reasons for data fluctuations, helping them quickly understand data anomalies and make corresponding decisions.

[0054] In this embodiment, based on the data characteristics of the fluctuation index results of business data, abnormal fluctuation index results that meet the characteristics of abnormal fluctuations are identified. Influencing factors contributing to the abnormal fluctuation index results are determined based on the business data. An explanation of the abnormal fluctuations in the business data is generated based on fluctuation interpretation knowledge, the abnormal fluctuation index results, and the relevant data of the influencing factors. Compared to traditional manual analysis methods, this approach eliminates excessive reliance on human experience, significantly saving time and effort for financial personnel and other business staff, and reducing labor costs.

[0055] Meanwhile, because the entire analysis process is based on automated processing by computer equipment, it is not affected by personal experience, carelessness, or work status, which can ensure the accuracy and consistency of the analysis results. Even during periods of large data volume or busy business hours, it can run efficiently and stably, effectively avoiding analysis omissions or judgment errors caused by human fatigue or negligence. This significantly improves analysis efficiency and provides strong data support for various business scenarios such as corporate financial management, e-commerce operations, and industrial production. It helps companies to promptly identify potential problems in business data, make scientific and reasonable decisions, and improve the level of corporate operation and management.

[0056] based on Figure 2 In some optional implementations of the embodiments shown, when generating explanations for abnormal fluctuations in business data, the weights corresponding to the abnormal fluctuation index results, influencing factors, and fluctuation explanation knowledge can be determined. These weights represent the relative importance of the three information sources in the process of generating explanations for abnormal fluctuations. That is, the higher the weight of the information source, the more emphasis is placed on generating explanations for abnormal fluctuations based on that type of information source, the greater the depth of analysis based on that type of information source, and the higher the proportion of explanations for abnormal fluctuations analyzed based on that type of information source.

[0057] Furthermore, based on the respective weights of the abnormal fluctuation index results, influencing factors, and fluctuation interpretation knowledge, anomaly fluctuation explanations for business data can be generated from the relevant data of the abnormal fluctuation index results and influencing factors, as well as the fluctuation interpretation knowledge. By determining these weights, computer equipment can more accurately grasp the key roles of different information sources when performing intelligent analysis and reasoning to generate anomaly fluctuation explanations, and reasonably adjust the focus of analysis according to the weight allocation.

[0058] For example, if the weight of the abnormal fluctuation indicator result is high, the computer equipment will focus on the characteristics of the fluctuation indicator itself, deeply analyze its degree of abnormality and trend, and provide a detailed explanation of the specific numerical changes of the fluctuation indicator and the degree of deviation from the normal range in the generated abnormal fluctuation explanation. When the weight of the influencing factors is large, the equipment will increase the mining and analysis of the relevant data of the influencing factors, explaining in detail how each influencing factor interacts and causes abnormal data fluctuations. For example, in a financial scenario, it will deeply analyze the specific impact of factors such as the time, amount, and transaction object of each transaction on the abnormality of the account data. If the weight of the fluctuation explanation knowledge is prominent, the computer equipment will make full use of the existing set of professional knowledge, referencing historical data summaries, industry experience accumulation, and expert research results, and compare and analyze the current abnormal fluctuation situation with similar cases, thereby generating an abnormal fluctuation explanation that is more in line with the actual situation and more professional and instructive.

[0059] This weighted approach to generating explanations for abnormal fluctuations makes the analysis results more scientific and comprehensive, better meeting the needs for data anomaly analysis in different business scenarios, providing enterprises with more valuable information support, and helping them optimize business processes and decision-making through business data anomalies.

[0060] Optionally, a feedback mechanism can be introduced during the process of generating explanations for abnormal fluctuations in business data. That is, after reviewing the generated explanations, business personnel can provide feedback on the accuracy and completeness of the explanations based on the actual situation. The computer equipment then dynamically adjusts and optimizes the weights of the abnormal fluctuation index results, influencing factors, and fluctuation explanation knowledge based on this feedback.

[0061] Specifically, when determining the weights of the three information sources mentioned above, the computer device can also obtain the user's evaluation information on the previous abnormal fluctuation explanation, and adjust the weights of the abnormal fluctuation index results, influencing factors and fluctuation explanation knowledge on which the previous abnormal fluctuation explanation was based based on the evaluation information. The adjusted weights are used as the weights on which the current fluctuation explanation is based.

[0062] One method for adjusting weights based on user feedback is as follows: If user feedback indicates dissatisfaction with the previous explanation of abnormal fluctuations, then a weight adjustment factor is determined based on this feedback, and the weights corresponding to the abnormal fluctuation index results, influencing factors, and fluctuation explanation knowledge used in the previous explanation are adjusted accordingly. Conversely, if user feedback indicates satisfaction with the previous explanation, then the weights corresponding to the abnormal fluctuation index results, influencing factors, and fluctuation explanation knowledge used in the previous explanation are adjusted based on the adjustment factor employed in the previous explanation.

[0063] For example, if business personnel report that the explanation for a certain abnormal fluctuation did not adequately consider a certain influencing factor, resulting in an incomplete explanation, then in subsequent analyses of similar scenarios, the computer equipment will appropriately increase the weight of that influencing factor to improve the quality of the explanation.

[0064] For example, in financial data analysis scenarios, a weight matrix can be used to dynamically adjust the emphasis of the three different information sources when generating explanations for abnormal fluctuations. This weighting is dynamically adjusted based on user feedback, and subsequent fluctuation analyses of the same subject will use the adjusted weights, resulting in more accurate explanations for abnormal fluctuations.

[0065] The weight matrix can be initialized by financial experts, for example, by referencing abnormal volatility indicators (50%), influencing factors (30%), and knowledge of volatility explanations (20%). The next step is to set rules for generating explanations of abnormal volatility in the prompts and use a pre-trained language model. Before generating explanations, the model should use a custom variable factor based on its input to calculate the actual weights, such as: Actual Weight = Initial Value × Variable Factor.

[0066] Therefore, the reference weights for each input component of the model's current response (i.e., the results of abnormal fluctuation indicators, relevant data on influencing factors, and knowledge of fluctuation interpretation) are the actual weights calculated above.

[0067] Therefore, whether the actual weights in this calculation are reliable or should be used as the initial values ​​for the next calculation needs to be determined based on user feedback. If a user likes the answer, it indicates that the variable factor is reliable, and a new weight matrix can still be generated based on that variable factor. If a user dislikes the generated explanation, it indicates that the variable factor is unreliable, and the variable factor should not be used in this calculation. The variable factor can be redefined, and the weights can be adjusted based on the redefined variable factor.

[0068] Therefore, through flexible weight configuration, the generation of anomaly fluctuation explanations has achieved a leap from "single explanation" to "adaptive explanation." That is, by combining user feedback and dynamic weight adjustments, the system can generate highly customized anomaly fluctuation explanations for different business scenarios and data characteristics. This adaptive explanation mechanism not only improves the accuracy of the analysis results but also, through continuous optimization of weight allocation, allows the model to gradually learn the analytical preferences and decision-making needs of business personnel, greatly enhancing the practicality of the anomaly fluctuation explanation conclusions and the user experience.

[0069] based on Figure 2 In some alternative implementations of the illustrated embodiment, when determining the influencing factors contributing to the abnormal fluctuation index results based on business data, multiple detailed record data points from the business data can be obtained. Furthermore, for each detailed record data point, its contribution to the abnormal fluctuation index results is calculated.

[0070] The method for calculating this contribution can be determined based on specific business scenarios and needs. For example, statistical methods such as correlation analysis and regression analysis can be used. By calculating the contribution, the impact of each detailed record of data on abnormal fluctuations can be quantified, thereby more accurately identifying which factors are the key factors causing abnormal fluctuations.

[0071] After calculating the contribution of all detailed record data, the business factors corresponding to the detailed record data whose contribution meets preset conditions are identified as influencing factors contributing to the abnormal fluctuation index results. These preset conditions can be flexibly set according to actual circumstances; for example, a contribution threshold can be set. When the contribution of a detailed record data exceeds this threshold, its corresponding business factor is considered an influencing factor. In this way, factors that truly affect abnormal fluctuations can be filtered from massive amounts of business data, providing effective support for subsequent analysis and decision-making.

[0072] based on Figure 2In some alternative implementations of the illustrated embodiment, the fluctuation index results of the business data can be obtained by acquiring the business data and comparative data associated with it. This comparative data corresponds to the same index as the business data but to different periods, and / or the comparative data corresponds to the same index as the business data and represents the average level of that index within a target range. For example, the target range could be the same index data from different branches within the same business segment, or the average index data of companies of similar size in the same industry. By comparing and analyzing the business data with these comparative data, the fluctuation of the business data under different periods or different reference standards can be more clearly observed.

[0073] After acquiring the business data and comparison data, the volatility index of the business data relative to the comparison data can be calculated according to the calculation method of the volatility index, thus obtaining the volatility index result of the business data. A volatility index is an indicator that reflects the volatility of data. For example, some volatility indices reflect the amount of data volatility, in which case the difference between the business data and the comparison data can be calculated. This difference can be expressed as an absolute numerical difference, such as the difference between specific values; or as a relative proportional difference, such as a percentage change. Other volatility indices reflect the volatility rate of data, in which case the proportion of the difference between the business data and the comparison data relative to the comparison data can be calculated to measure the volatility rate of the data.

[0074] This method of obtaining business data fluctuation indicators can make full use of historical data, industry data, and other references to more accurately identify fluctuations in business data, providing a reliable prerequisite for data anomaly analysis.

[0075] based on Figure 2 In some alternative implementations of the illustrated embodiment, when determining abnormal fluctuation index results, a pre-trained language model can be used to identify abnormal fluctuation index results within the fluctuation index results. Specifically, a pre-trained language model configured with anomaly recognition knowledge can be obtained. The fluctuation index results and anomaly recognition prompts (prompts) can be input into the pre-trained language model. Then, guided by the anomaly recognition prompts, the pre-trained language model, based on the abnormal fluctuation characteristics taught by the anomaly recognition knowledge and the data characteristics of the fluctuation index results, identifies abnormal fluctuation index results that conform to the abnormal fluctuation characteristics and outputs the abnormal fluctuation index results.

[0076] The pre-trained language model learns from a large amount of data, mastering the patterns and regularities of abnormal fluctuations. When inputting fluctuation index results and anomaly identification prompts, the model analyzes the data characteristics of the fluctuation index results based on its learned anomaly identification knowledge. It compares each data point in the fluctuation index results with known abnormal fluctuation characteristics to determine if there are any matches. Once it finds that the data characteristics of certain fluctuation index results match abnormal fluctuation characteristics, it identifies them as abnormal fluctuation index results and outputs them. This approach leverages the powerful data processing and pattern recognition capabilities of the pre-trained language model, enabling it to quickly and accurately identify anomalies from a large number of fluctuation index results, providing crucial evidence for subsequent data anomaly analysis and significantly improving analysis efficiency and accuracy.

[0077] Anomaly identification knowledge can consist of user feedback triggered by the explanations of abnormal fluctuations output by internal enterprise knowledge and models. This user feedback can be used to specify the range of anomalies, allowing the range to remain dynamically changing. The enterprise's internal knowledge base can maintain information such as enterprise announcements, economic patterns, and seasonal cycles. For example, an enterprise announcement may mention a major strategic adjustment in the near future, which may affect the normal fluctuation range of certain business data; economic patterns show that certain industry indicators will fluctuate regularly under specific economic cycles; seasonal cycle factors will cause some business data to show similar change patterns at specific times of the year.

[0078] This internal corporate knowledge helps the model determine whether the fluctuation indicators of business data are abnormal. For example, if a company announcement clearly states that a certain business segment will be conducting a large-scale promotional activity, and the sales data of that business segment shows a significant increase during the promotion period, based on this background knowledge, the model will not simply identify it as an abnormal fluctuation, but will take into account the promotional activity as a factor.

[0079] Similarly, based on economic principles, if the current economy is in a period of expansion, certain industry indicators, such as enterprise order volume and production scale, will show an upward trend. When analyzing the fluctuations in these business data indicators, the model will consider them as normal fluctuations consistent with economic laws, unless the fluctuation range exceeds the reasonable range for that economic cycle. Regarding seasonal factors, taking the retail industry as an example, sales typically see a significant increase during annual holidays. When analyzing business data fluctuations during this period, the model will refer to historical data from the same period and seasonal cycle patterns to accurately determine whether the current fluctuation falls within the normal range of seasonal fluctuations.

[0080] The anomaly range specified by user feedback further enhances the flexibility and accuracy of the model's judgment. If, during actual business operations, users find that the abnormal fluctuations identified by the model do not match the actual business situation, they can adjust the anomaly range through the feedback mechanism. The model then determines the abnormal fluctuation index results that conform to the user-provided range. For example, if a user reports that fluctuations in a certain type of business data within a specific time period due to special business activities are within the normal range, the model, upon receiving this feedback, will incorporate this information into its anomaly identification knowledge system. Subsequent analysis of similar business data will then take this special case into account to avoid misjudgment.

[0081] After organizing this internal enterprise knowledge, it is input into a pre-trained language model. Simultaneously, user feedback generated by the model's previous explanations of abnormal fluctuations is incorporated to continuously update and improve the anomaly identification knowledge. In this way, the pre-trained language model can more accurately identify abnormal fluctuation indicators from the fluctuation results based on the latest and most comprehensive anomaly identification knowledge. This provides an accurate basis for subsequent data anomaly analysis, ensuring that the entire data anomaly analysis method is more aligned with the enterprise's actual business situation and dynamic changing needs.

[0082] The following will use the scenario of analyzing abnormal fluctuations in financial data as an example to further explain in detail the method of this application's embodiments. Figure 3 As shown, in this scenario, the entire process can be broken down into four distinct layers: the data layer, the anomaly detection layer, the attribution layer, and the explanation generation layer. A process engine connects these layers, achieving fully automated processing from raw data to the final interpretation conclusion. The functions and operations of each layer will be described below:

[0083] First, multiple algorithms are used for parallel pre-computation at the data layer to provide rich feature dimensions for subsequent fluctuation and anomaly analysis, while improving security. Pre-computation is performed before the data enters the model to ensure that real financial data does not flow directly into the large language model, thus avoiding the inability to guarantee data security by directly providing subject data to the large language model.

[0084] First, determine the subject name to be analyzed, the current period, and the analysis method (same-year or month-on-month).

[0085] Then, extract the current period and the data from the previous five periods of the current year from the database, and extract the data from the same period of each of the previous two years, for a total of three years and eighteen periods. Subject data preprocessing methods can include filling null values ​​with 0 or performing data anonymization.

[0086] Finally, ten algorithms are pre-computed and output in structured JSON format (JavaScript Object Notation). The ten algorithms are as follows:

[0087] (1) Basic volatility

[0088] 1. Month-on-month fluctuation amount

[0089] Current period value - Previous period value

[0090] 2. Cycle-on-cycle volatility

[0091] (Current period value - Previous period value) / Previous period value

[0092] 3. Year-on-year fluctuation amount

[0093] Current period value - same period last year value

[0094] 4. Year-on-year volatility

[0095] (Current period value - Interval value of the same period last year) / Interval value of the same period last year

[0096] (2) Trend Analysis

[0097] 5. Month-on-month increment series

[0098] Calculate the month-on-month fluctuation amount over multiple consecutive periods to form a time series.

[0099] Example: [ (Issue 9-Issue 8), (Issue 8-Issue 7), (Issue 7-Issue 6), (Issue 6-Issue 5), (Issue 5-Issue 4) ]

[0100] 6. Month-on-month growth rate series

[0101] Calculate the year-on-year volatility over multiple consecutive periods to form a time series.

[0102] Example: [ (Issue 9-Issue 8) / Issue 8, (Issue 8-Issue 7) / Issue 7, ... , (Issue 5-Issue 4) / Issue 4]

[0103] 7. Year-on-year increment series

[0104] Calculate the year-on-year fluctuation amount over several consecutive years to form a time series.

[0105] Example: [(September 2025 issue - September 2024 issue), (September 2024 issue - September 2023 issue)]

[0106] 8. Year-on-year growth rate series

[0107] Calculate the year-on-year volatility over several consecutive years to form a time series.

[0108] Example: [(September 2025 issue - September 2024 issue) / September 2024 issue, (September 2024 issue - September 2023 issue) / September 2023 issue]

[0109] (3) Potential energy analysis

[0110] 9. Month-on-month changes in year-on-year increments

[0111] Calculate the month-on-month fluctuation amount for the year-on-year increment sequence (i.e., the result of the 7th algorithm). That is: year-on-year increment of this period - year-on-year increment of the previous period.

[0112] Essentially, it measures whether the "absolute momentum of year-on-year growth" is accelerating or decelerating.

[0113] Example (2025): [(Year-on-year increase in 9 periods - Year-on-year increase in 8 periods), (Year-on-year increase in 8 periods - Year-on-year increase in 7 periods),...]

[0114] 10. Month-on-month changes in year-on-year growth rate

[0115] Calculate the month-on-month fluctuation amount for the year-on-year growth rate series (i.e., the result of the 8th algorithm). That is: current period year-on-year growth rate - previous period year-on-year growth rate.

[0116] Essentially, it measures whether the "relative speed of year-on-year growth" is accelerating or slowing down, and is used to identify inflection points in growth.

[0117] Example (2025): [(Year-on-year growth rate of 9 periods - Year-on-year growth rate of 8 periods), (Year-on-year growth rate of 8 periods - Year-on-year growth rate of 7 periods), ...]

[0118] Therefore, through the above-mentioned algorithms, the calculation results of various volatility indicators of subject data (i.e., volatility indicator results) can be obtained, including the calculation results of various volatility indicators such as month-on-month volatility amount, month-on-month volatility rate, month-on-month increment sequence, month-on-month growth rate sequence, year-on-year volatility amount, year-on-year volatility rate, year-on-year increment sequence, year-on-year growth rate sequence, month-on-month change of year-on-year increment, and month-on-month change of year-on-year growth rate.

[0119] Second, through a carefully designed prompt word engineering at the anomaly identification layer, a hierarchical, diversionary, and dynamic knowledge base analysis mechanism is set up to control the input and output of the large language model, ensuring the professionalism and structure of the analysis;

[0120] The system retrieves pre-calculated data from the data layer (i.e., volatility index results), calls the large language model based on prompt words, and finally outputs a structured conclusion on the anomaly identification of the volatility index results. The structure of the anomaly identification prompt words is as follows:

[0121] (1) Hierarchical mechanism of prompt words

[0122] First, the model sets an initial screening for the calculation results of the "basic volatility" part in the data layer. If the value exceeds the set range, the subject needs to be given special attention; if it is within the set range, it is considered normal volatility.

[0123] The model then continues to search for abnormal fluctuations in the calculation results of "trend analysis" and "potential energy analysis", outputs abnormal indicators, and outputs the cause of each abnormality, including the indicator name, the calculated value of the indicator, and the abnormal contribution rate.

[0124] Finally, the model combines with an anomaly identification knowledge base to give a final conclusion, and marks the subjects with abnormal fluctuations based on the conclusion.

[0125] (2) The diversion mechanism of prompt words

[0126] First, we conduct a month-on-month abnormal fluctuation analysis, and then we conduct a year-on-year abnormal fluctuation analysis.

[0127] Month-on-month and year-on-year analyses require the use of the calculated metrics after the data split, including:

[0128] Month-on-month comparisons include: month-on-month fluctuation amount, month-on-month volatility, month-on-month increment sequence, and month-on-month growth rate sequence.

[0129] Year-on-year data can be selected from: year-on-year fluctuation amount, year-on-year volatility, year-on-year increment series, year-on-year growth rate series, month-on-month change of year-on-year increment, and month-on-month change of year-on-year growth rate.

[0130] For example, a possible prompt could be, "You are an expert in accounting subject fluctuation analysis. I will provide you with some calculated indicator data for the current accounting subject in the current period in sys_prompt. Help me determine if there are any anomalies in the amount for the current period, record the basis for your judgment, and finally output according to my requirements."

[0131] Furthermore, an anomaly detection knowledge base can be input to inform the model of its anomaly detection strategy. For example, the strategy could be: "First, select the volatility indicator and combine it with the 'anomaly judgment criteria' to perform a preliminary screening. If it is within the normal range, it is judged as normal month-on-month or year-on-year. If it is within the warning range, further analysis is performed. For warning cases, you can use some professional methods (such as rule-based methods, statistics, machine learning, etc.) to analyze all calculated indicators for month-on-month or year-on-year comparisons to determine the cause of the anomaly (80%). For warning cases, you can search for information from multiple sources, such as the volatility pattern of the subject, industry experience, seasonal influences, and other factors, to determine the cause of the anomaly (20%)."

[0132] Therefore, in the anomaly identification layer, a carefully designed prompt word diversion mechanism can ensure that the model can perform month-on-month and year-on-year anomaly fluctuation analysis in an orderly manner during the analysis process.

[0133] Of course, not all of the ten algorithms mentioned above are necessary. A pluggable algorithm library can be used, allowing users or the system to dynamically select a combination of algorithms based on the characteristics of the account (such as asset or profit and loss). Alternatively, a feature selection model can be used to automatically filter the most effective features for the current account from the results of the ten algorithms and then input them into the anomaly detection model.

[0134] Third, a "drill down into detailed data of abnormal items" mechanism is introduced in the attribution layer to directly target the root cause of the anomaly and sort them according to their contribution, making the discovery of the root cause of the problem more accurate.

[0135] After the anomaly identification layer identifies the abnormal accounts, the attribution layer undertakes the crucial task of moving from "identifying the problem" to "locating the root cause." This layer uses an automated and quantitative process to accurately pinpoint the specific business events causing account fluctuations. Its core process is as follows:

[0136] (1) Drill down the data for abnormal items

[0137] The system automatically performs a "drill-down" operation, penetrating from the general ledger level down to the fine-grained account ledger data. This step is fundamental for in-depth attribution, linking macro-level account fluctuations with each specific micro-level business record (such as voucher number, transaction date, customer / supplier, and amount).

[0138] This operation enables automatic tracing from financial data to actual business data, breaking down the data barriers between finance and business in traditional analysis.

[0139] (2) Calculation of the contribution of abnormal items in the subject details

[0140] The system calculates the "contribution" of each detailed data point drilled to the overall abnormal fluctuations.

[0141] The contribution calculation logic is based on the aforementioned anomaly analysis method. For example:

[0142] If it is a month-on-month anomaly, then the contribution of the detailed record = (the current period's amount of the detailed record - the previous period's amount of the detailed record) / the total month-on-month fluctuation amount of the account;

[0143] If it is an anomaly compared to the same period last year, then the contribution of the detailed record = (the amount of the detailed record in the current period - the amount of the detailed record in the same period last year) / the total year-on-year fluctuation amount of the item.

[0144] This operation quantifies the abstract fluctuations in accounts into specific responsibilities for each detailed business transaction, so that the conclusion of anomaly analysis is no longer a vague "increased expenses", but a precise "expenses for project XX led to an increase in expenses of XX yuan".

[0145] (3) Ranking and TopN Filtering of Detailed Anomalies in Subject Details

[0146] The system automatically sorts the contribution of all detailed records with calculated contribution in descending order and extracts the Top N detailed data according to preset rules (such as the top 5 items, or items with a cumulative contribution of more than 80%).

[0147] This step follows the "Pareto principle" and can automatically focus on the most critical influencing factors, efficiently eliminating a large number of non-critical details. This allows analysts to concentrate on the core issues and greatly improves the efficiency and accuracy of attribution analysis.

[0148] (4) Structured output

[0149] The results of the attribution analysis, including the Top N detailed items that caused the abnormal fluctuations, their specific contributions, and amounts, will be output in a standardized structure (JSON format).

[0150] The structured output provides a data interface that can be directly and seamlessly called by the downstream "generative interpretation layer", ensuring the automation and integration of the entire process, enabling subsequent AI models to generate accurate natural language interpretations based on this standardized data.

[0151] Optionally, in addition to drilling down into the detailed financial ledgers, one can also drill down into related business system data, such as contracts, orders, and logistics documents, in parallel or sequentially. The system can calculate changes in business data (such as a sudden increase in order amounts) and correlate them with anomalies in financial accounts. This approach provides a deeper, more integrated business and financial attribution insight, facilitating the explanation of financial results from the perspective of business drivers.

[0152] Fourth, in the interpretation generation layer, on the one hand, sufficient data sources are ensured, and on the other hand, the weight of input information is dynamically adjusted to generate the most perfect interpretation that fits the actual business.

[0153] (1) Input of multi-source, high-reliability data

[0154] The system does not generate explanations out of thin air, but rather performs comprehensive reasoning based on three verified, high-quality data sources:

[0155] The structured output of the anomaly identification layer provides authoritative conclusions on "what kind of fluctuation is abnormal" and its severity, such as abnormal year-on-year fluctuations in sales expenses.

[0156] The attribution layer structured output provides the specific root cause of "why the fluctuations are abnormal," such as abnormal changes in a certain detailed dimension of a subject causing abnormal fluctuations in the subject.

[0157] The expert-explained knowledge base provides domain knowledge and script templates for "how to explain such abnormal fluctuations," such as... Figure 4 The following is an example of the explanatory knowledge for fluctuations. For instance, when the account "Cash and Deposits" increases year-on-year and the influencing factor is "Company Operations", the corresponding explanatory knowledge is "Increased sales revenue leads to increased cash inflow from operating activities; or successful recovery of previous accounts receivable improves liquidity".

[0158] Therefore, this multi-source input design ensures that the generated explanations are "accurate (from data), deep (from attribution), and professional (from knowledge base)," avoiding the "illusion" or lack of professionalism that may occur in large language models.

[0159] (2) Model calling and structured filling

[0160] After obtaining the aforementioned multiple information sources, appropriate weights can be assigned to data from different sources based on the aforementioned dynamic weight setting mechanism, and information fusion and reasoning can be performed based on the assigned weights to generate the final natural language interpretation conclusion.

[0161] Model invocation: The weighted and fused multi-source information, along with the generation instructions, is sent to the large language model.

[0162] Structured output: The model is required to output structured explanations and conclusions.

[0163] Fill in the difference explanation: The system will finally fill the structured explanation fields into the preset report template or interactive interface to form the final natural language difference explanation.

[0164] In addition to the aforementioned model-based anomaly identification of volatility indicator results, a configurable, multi-threshold rule engine can be built without relying on AI models. This engine receives all ten pre-calculated indicators and outputs structured volatility anomaly conclusions through a set of "IF-THEN" rules (e.g., IF month-on-month volatility > 50% AND year-on-year volatility < 10% THEN marked as 'short-term sudden anomaly').

[0165] For example, a set of rules can be set up such that if the month-on-month fluctuation exceeds a certain value while the year-on-year volatility is below a certain level, it is judged as "abnormal fluctuation in a specific period"; if both the month-on-month and year-on-year volatility are in a high range, it is marked as "significant anomaly in two dimensions," and so on. This rule engine approach is characterized by its flexibility and high customizability. Enterprises can flexibly configure different thresholds and rule combinations according to their own business characteristics and needs to adapt to various complex business scenarios. Moreover, the rule engine does not rely on a complex model training process, making it relatively simple and quick to implement, and it can be quickly deployed and produce results. At the same time, for some scenarios with relatively fixed business logic and relatively clear abnormal patterns, the rule engine can efficiently and accurately output the conclusion of abnormal fluctuation. This approach is simpler to implement, more interpretable, and suitable for scenarios where there are concerns about the credibility of AI models or where computing resources are limited.

[0166] Therefore, in this embodiment, the traditional approach to data anomaly analysis suffers from high barriers to entry and difficulty in replication. Specifically, the ability to perform data anomaly analysis is tied to a few experienced finance professionals, making it difficult for newcomers to quickly learn, resulting in inconsistent team capabilities and hindering standardization and scalability. This embodiment addresses these shortcomings by constructing a systematic, automated, and flexible data anomaly analysis method, effectively lowering the barrier to entry. This method not only enables individuals without a strong financial background to quickly begin data anomaly analysis but also ensures the accuracy and consistency of analysis results through standardized processes and structured outputs, thereby achieving standardization and scalability of data anomaly analysis. Both large and small enterprises can flexibly apply this method for data anomaly analysis based on their own business characteristics and needs, improving financial management and decision-making efficiency.

[0167] Traditional methods of data anomaly analysis suffer from limitations in depth and scope, often relying on simple year-on-year or month-on-month comparisons based on account balances and total transactions. This hinders rapid, multi-dimensional, drill-down analysis (e.g., tracing back to specific business types, product lines, departments, customers / suppliers, or even voucher levels), leaving the analysis superficial and failing to address the root causes. This application's implementation introduces a "drill-down anomaly account details" mechanism and multi-source data fusion analysis, achieving deep penetration from macro-level account fluctuations to micro-level business records. The system automatically traces back to underlying data such as voucher numbers and transaction dates, and calculates the contribution of detailed records to overall anomalies, transforming vague conclusions like "expense increase" into precise details like "XX project expenditures led to an increase of XX yuan in expenses." Combined with corroborating data from business systems (such as contracts and orders), the system further explains financial results from a business driver perspective, forming a comprehensive "financial-business" dual-dimensional attribution analysis. This multi-level, multi-dimensional approach significantly improves the depth and accuracy of anomaly analysis, helping companies quickly pinpoint the root causes of problems and develop targeted improvement measures.

[0168] The data anomaly analysis method in the embodiments of this application has been described above. The computer device in the embodiments of this application is described below. Please refer to [link / reference]. Figure 5 One embodiment of the computer device in this application includes:

[0169] An acquisition unit is used to acquire fluctuation index results of business data, wherein the fluctuation index results are used to measure the fluctuation of the business data.

[0170] The first determining unit is used to determine, based on the data characteristics of the fluctuation index results, the abnormal fluctuation index results that conform to the abnormal fluctuation characteristics.

[0171] The second determining unit is used to determine, based on the business data, the influencing factors that contribute to the result of the abnormal fluctuation index.

[0172] The generation unit is used to acquire fluctuation interpretation knowledge and generate an abnormal fluctuation interpretation of the business data based on the fluctuation interpretation knowledge, the abnormal fluctuation index results, and the relevant data of the influencing factors.

[0173] In a preferred embodiment of this invention, the generating unit is specifically used for:

[0174] Determine the weights corresponding to the abnormal fluctuation index results, the influencing factors, and the fluctuation explanation knowledge;

[0175] Based on the weights corresponding to the abnormal fluctuation index results, the influencing factors, and the fluctuation interpretation knowledge, an abnormal fluctuation interpretation of the business data is generated from the abnormal fluctuation index results, the relevant data of the influencing factors, and the fluctuation interpretation knowledge.

[0176] In a preferred embodiment of this invention, the generating unit is specifically used for:

[0177] Obtain user feedback on the explanation of the previous abnormal fluctuation;

[0178] The weights corresponding to the abnormal fluctuation index results, the influencing factors, and the fluctuation interpretation knowledge used in the previous abnormal fluctuation interpretation are adjusted based on the evaluation information, and the adjusted weights are used as the weights used to generate the fluctuation interpretation this time.

[0179] In a preferred embodiment of this invention, the generating unit is specifically used for:

[0180] If the evaluation information indicates that the user is not satisfied with the previous explanation of the abnormal fluctuation, then the weight change factor is determined according to the evaluation information, and the weights corresponding to the abnormal fluctuation index results, the influencing factors and the fluctuation explanation knowledge on which the previous explanation of the abnormal fluctuation was based are adjusted according to the change factor.

[0181] If the evaluation information indicates that the user is satisfied with the previous explanation of the abnormal fluctuation, then the weights of the abnormal fluctuation index results, the influencing factors, and the fluctuation explanation knowledge used in the previous explanation of the abnormal fluctuation are adjusted according to the variation factors used in the previous explanation of the abnormal fluctuation.

[0182] In a preferred embodiment of this invention, the second determining unit is specifically used for:

[0183] Obtain multiple detailed record data from the aforementioned business data;

[0184] For each detailed record data item, calculate the contribution of the detailed record data item to the result of the abnormal fluctuation index;

[0185] Among the multiple detailed record data, the business factors corresponding to the detailed record data whose contribution meets the preset conditions are determined as the influencing factors that contribute to the result of the abnormal fluctuation index.

[0186] In a preferred embodiment of this invention, the acquisition unit is specifically used for:

[0187] Obtain the business data and the comparison data associated with the business data; the comparison data and the business data correspond to the same indicator but to different periods, and / or the comparison data and the business data correspond to the same indicator and the comparison data is the average level of the indicator within the target range;

[0188] Based on the calculation method of the volatility index, the volatility index of the business data relative to the comparison data is calculated to obtain the volatility index result of the business data.

[0189] In a preferred embodiment of this invention, the first determining unit is specifically used for:

[0190] Obtain a pre-trained language model, wherein the pre-trained language model is configured with anomaly recognition knowledge;

[0191] The fluctuation index results and anomaly identification prompts are input into the pre-trained language model so that, prompted by the anomaly identification prompts, the pre-trained language model identifies the abnormal fluctuation index results that conform to the abnormal fluctuation characteristics taught by the anomaly identification knowledge and the data characteristics of the fluctuation index results, and outputs the abnormal fluctuation index results.

[0192] In this embodiment, the operations performed by each unit in the computer device are the same as described above. Figure 2 The embodiments shown and their various alternative implementations are similar to those described herein, and will not be repeated here.

[0193] The computer device in the embodiments of this application is described below. Please refer to [link / reference]. Figure 6 One embodiment of the computer device in this application includes:

[0194] The computer device 600 may include one or more central processing units (CPUs) 601 and a memory 605, in which one or more applications or data are stored.

[0195] The memory 605 can be volatile or persistent storage. The program stored in the memory 605 can include one or more modules, each module including a series of instruction operations on the computer device. Furthermore, the central processing unit 601 can be configured to communicate with the memory 605 and execute the series of instruction operations stored in the memory 605 on the computer device 600.

[0196] The computer device 600 may also include one or more power supplies 602, one or more wired or wireless network interfaces 603, one or more input / output interfaces 604, and / or one or more operating systems, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, etc.

[0197] The central processing unit 601 can perform the aforementioned... Figure 2 The operations performed by the computer device in the illustrated embodiments and their various alternative embodiments are not described in detail here.

[0198] This application also provides a computer storage medium, one embodiment of which includes: the computer storage medium storing instructions, which, when executed on a computer, cause the computer to perform the aforementioned... Figure 2 The operations performed by the computer device in the illustrated embodiments and their various alternative embodiments.

[0199] This application also provides a computer program product, one embodiment of which includes: when the computer program product is run on a computer device, it causes the computer device to perform the aforementioned... Figure 2 The operations performed by the computer device in the illustrated embodiments and their various alternative embodiments.

[0200] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0201] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between apparatuses or units through some interfaces, and may be electrical, mechanical, or other forms.

[0202] The units described as separate components may or may not be physically separate. 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 units can be selected to achieve the purpose of this embodiment according to actual needs.

[0203] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0204] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A data anomaly analysis method, characterized in that, The method includes: Obtain the fluctuation index results of the business data, which are used to measure the fluctuation of the business data; Based on the data characteristics of the volatility index results, identify the abnormal volatility index results that conform to the abnormal volatility characteristics. Based on the business data, determine the influencing factors that contribute to the results of the abnormal fluctuation index; Obtain fluctuation interpretation knowledge, and generate an abnormal fluctuation interpretation of the business data based on the fluctuation interpretation knowledge, the abnormal fluctuation index results, and the relevant data of the influencing factors.

2. The method according to claim 1, characterized in that, The step of generating an explanation of abnormal fluctuations in the business data based on the fluctuation interpretation knowledge, the abnormal fluctuation index results, and relevant data on the influencing factors includes: Determine the weights corresponding to the abnormal fluctuation index results, the influencing factors, and the fluctuation explanation knowledge; Based on the weights corresponding to the abnormal fluctuation index results, the influencing factors, and the fluctuation interpretation knowledge, an abnormal fluctuation interpretation of the business data is generated from the abnormal fluctuation index results, the relevant data of the influencing factors, and the fluctuation interpretation knowledge.

3. The method according to claim 2, characterized in that, The determination of the weights corresponding to the abnormal fluctuation index results, the influencing factors, and the fluctuation explanation knowledge includes: Obtain user feedback on the explanation of the previous abnormal fluctuation; The weights corresponding to the abnormal fluctuation index results, the influencing factors, and the fluctuation interpretation knowledge used in the previous abnormal fluctuation interpretation are adjusted based on the evaluation information, and the adjusted weights are used as the weights used to generate the fluctuation interpretation this time.

4. The method according to claim 3, characterized in that, The step of adjusting the weights of the abnormal fluctuation index results, the influencing factors, and the fluctuation explanation knowledge used in the previous abnormal fluctuation explanation based on the evaluation information includes: If the evaluation information indicates that the user is not satisfied with the previous explanation of the abnormal fluctuation, then the weight change factor is determined according to the evaluation information, and the weights corresponding to the abnormal fluctuation index results, the influencing factors and the fluctuation explanation knowledge on which the previous explanation of the abnormal fluctuation was based are adjusted according to the change factor. If the evaluation information indicates that the user is satisfied with the previous explanation of the abnormal fluctuation, then the weights of the abnormal fluctuation index results, the influencing factors, and the fluctuation explanation knowledge used in the previous explanation of the abnormal fluctuation are adjusted according to the variation factors used in the previous explanation of the abnormal fluctuation.

5. The method according to claim 1, characterized in that, The step of determining the influencing factors that contribute to the abnormal fluctuation index results based on the business data includes: Obtain multiple detailed record data from the aforementioned business data; For each detailed record data item, calculate the contribution of the detailed record data item to the result of the abnormal fluctuation index; Among the multiple detailed record data, the business factors corresponding to the detailed record data whose contribution meets the preset conditions are determined as the influencing factors that contribute to the result of the abnormal fluctuation index.

6. The method according to claim 1, characterized in that, The results of the fluctuation indicators for acquiring business data include: Obtain the business data and the comparison data associated with the business data; the comparison data and the business data correspond to the same indicator but to different periods, and / or the comparison data and the business data correspond to the same indicator and the comparison data is the average level of the indicator within the target range; Based on the calculation method of the volatility index, the volatility index of the business data relative to the comparison data is calculated to obtain the volatility index result of the business data.

7. The method according to claim 1, characterized in that, The step of determining abnormal fluctuation index results that meet the abnormal fluctuation characteristics based on the data characteristics of the fluctuation index results includes: Obtain a pre-trained language model, wherein the pre-trained language model is configured with anomaly recognition knowledge; The pre-trained language model is input with the fluctuation index results and anomaly identification prompts, so that the pre-trained language model, under the prompts of the anomaly identification prompts, identifies the abnormal fluctuation index results that meet the abnormal fluctuation characteristics taught by the anomaly identification knowledge and the data characteristics of the fluctuation index results, and outputs the abnormal fluctuation index results.

8. A computer device, characterized in that, The computer device includes: An acquisition unit is used to acquire fluctuation index results of business data, wherein the fluctuation index results are used to measure the fluctuation of the business data. The first determining unit is used to determine, based on the data characteristics of the fluctuation index results, the abnormal fluctuation index results that conform to the abnormal fluctuation characteristics. The second determining unit is used to determine, based on the business data, the influencing factors that contribute to the result of the abnormal fluctuation index. The generation unit is used to acquire fluctuation interpretation knowledge and generate an abnormal fluctuation interpretation of the business data based on the fluctuation interpretation knowledge, the abnormal fluctuation index results, and the relevant data of the influencing factors.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that, The computer storage medium stores instructions that, when executed on the computer, cause the computer to perform the method as described in any one of claims 1 to 7.

Images