Method, device, computer equipment and storage medium for attributing abnormal movement of indicators

Abstract

The present application discloses a method, apparatus, computer device, and storage medium for analyzing the attribution of abnormal indicator changes. The implementation of the method includes: obtaining problem information and decomposing it into intent, business indicators, and indicator dimensions; determining whether indicator attribution is required for the user's problem; if indicator attribution is required, determining whether there is an abnormality in the business indicators; if there is an abnormality in the business indicators, determining an attribution diagnosis model according to the intent and indicator type, so as to predict the attribution result of the business indicator through the attribution diagnosis model; determining the target external factors and target internal factors affecting the attribution result, and obtaining the countermeasures for the target internal factors or target external factors. By asking questions in a dialogue, introducing a large language model and an attribution analysis model, the reason for the abnormal fluctuation of the indicator can be quickly located, and minute-level attribution analysis can be achieved through a dialogue method, which has a wide range of application scenarios and potential.

Description

Technical Field

[0001] This application relates to the technical field of data processing, and in particular, to a method, device, computer device, and storage medium for analyzing the attribution of abnormal changes in indicators. Background Art

[0002] Currently, when performing data query and business analysis, it mainly relies on dashboards and business analysis platforms. A large number of data analysts are required to calculate business indicators and integrate them into the platform, and then present them to users through dashboards. When performing simple queries, users need to manually analyze the dashboard tables to obtain data, and the attribution diagnosis of indicator fluctuations also relies on manual dimension-by-dimension analysis of contribution degrees.

[0003] However, this mode has problems: First, the attribution efficiency is low. Manual operations are far less efficient than models under complex calculation rules, consuming a large amount of time and labor. Second, attribution is prone to errors. When manually processing a large amount of data, errors are likely to occur due to limited attention and concentration, and manual analysis has great limitations. Third, the analysis model is single and can only perform multi-dimensional attribution on current indicators. Fourth, the analysis scenarios are incomplete. It can only meet single-indicator attribution analysis. For composite indicators, it is necessary to trace upstream indicators for analysis, and the current technical solutions for attribution cannot meet this requirement. Fifth, specific reasons and countermeasures affecting abnormal changes in indicators are not given based on actual internal and external factors. There are certain problems in terms of attribution efficiency, real-time performance, comprehensiveness, reducing labor costs, and improving decision-making efficiency. Summary of the Invention

[0004] Based on this, in order to solve the above technical problems, it is necessary to provide a method, device, computer device, and storage medium for analyzing the attribution of abnormal changes in indicators to solve at least one of the problems existing in the above prior art.

[0005] In the first aspect, an embodiment of this application provides a method for analyzing the attribution of abnormal changes in indicators, including:

[0006] Obtain problem information, and decompose the problem information into an intention, a business indicator, and an indicator dimension;

[0007] Determine whether indicator attribution is required for the user problem;

[0008] If indicator attribution is required, determine whether the business indicator is abnormal;

[0009] If the business indicator is abnormal, determine an attribution diagnosis model according to the intention and the indicator dimension, so as to predict an attribution result of the business indicator through the attribution diagnosis model;

[0010] Determine target external factors and target internal factors affecting the attribution result, and obtain countermeasures for the target internal factors or the target external factors.

[0011] In one embodiment, determining whether there is an abnormality in the service metric includes:

[0012] Determining whether there is an abnormality in the service metric at the current time; and / or

[0013] Determining whether the numerical difference of the service data corresponding to the service metric in different comparison periods is greater than a preset difference threshold.

[0014] In one embodiment, determining whether there is an abnormality in the service metric at the current time includes:

[0015] Determining the mean value of the service data corresponding to all service metrics within each preset time range;

[0016] Determining whether the difference between the mean value and the mean value of the service data corresponding to all service metrics within an adjacent preset time range is greater than a preset difference;

[0017] If so, there is an abnormality in the service metric;

[0018] If not, determining the change trend of the service data corresponding to each index dimension;

[0019] When the difference between the peak value and the trough value of the change trend is greater than a preset threshold, there is an abnormality in the service metric.

[0020] In one embodiment, the attribution diagnosis model is a single-dimensional diagnosis model. Predicting the attribution result of the service metric through the attribution diagnosis model includes:

[0021] Calculating the contribution value of the service data corresponding to each index dimension respectively;

[0022] Sorting the contribution values in ascending or descending order to obtain the index dimension with the highest contribution value as the attribution result of the service metric.

[0023] In one embodiment, the attribution diagnosis model is a composite index blood relationship diagnosis model. Predicting the attribution result of the service metric through the attribution diagnosis model includes:

[0024] When the service metric is a composite index, querying the upstream indexes of the service metric to obtain multiple parent indexes of the service metric;

[0025] Calculating the influence value of each parent index;

[0026] Sorting the influence values in ascending or descending order to obtain the parent index with the highest influence value as the attribution result of the service metric.

[0027] In one embodiment, the attribution diagnosis model is a dimensional cross-attribution diagnosis model. The attribution result of the service metric predicted by the attribution diagnosis model includes:

[0028] Combine the metric dimensions to obtain multiple groups of multi-dimensional metric dimensions;

[0029] Calculate the contribution value of each metric dimension under multiple groups of multi-dimensional metric dimensions;

[0030] Use the group of metric dimensions with the largest contribution value as the attribution result of the service metric.

[0031] In one embodiment, obtain problem information and decompose the problem information into intent, service metric, and metric dimension, including:

[0032] Set prompt words, which include dimension information and metric metadata;

[0033] Obtain the problem information and determine the metric to be attributed according to the problem information;

[0034] Query the attribution dimension metadata corresponding to the metric to be attributed in the database;

[0035] Based on the prompt words, problem information, and the attribution dimension metadata, perform semantic analysis through a large language model to obtain a DSL object;

[0036] Query the database according to the DSL object to obtain metric data.

[0037] In a second aspect, there is provided an apparatus for analyzing the attribution of metric anomalies, including:

[0038] A problem information decomposition unit, configured to obtain problem information and decompose the problem information into intent, service metric, and metric dimension;

[0039] A metric attribution determination unit, configured to determine whether metric attribution is required for the user problem;

[0040] A metric anomaly analysis unit, configured to determine whether the service metric is abnormal if metric attribution is required;

[0041] An attribution result prediction unit, configured to, if the service metric is abnormal, determine an attribution diagnosis model according to the intent and metric dimension, so as to predict the attribution result of the service metric through the attribution diagnosis model;

[0042] A countermeasure determination unit, configured to determine the target external factors and target internal factors affecting the attribution result, and obtain the countermeasures for the target internal factors or the target external factors.

[0043] In a third aspect, a computer device is provided, including a memory, a processor, and computer-readable instructions stored on the memory and running on the processor. When the processor executes the computer-readable instructions, the method for analyzing the attribution of abnormal changes in the above-mentioned metrics is implemented.

[0044] In a fourth aspect, a readable storage medium is provided, on which computer-readable instructions are stored. When the computer-readable instructions are executed by a processor, the method for analyzing the attribution of abnormal changes in the above-mentioned metrics is implemented.

[0045] For the above-mentioned method for analyzing the attribution of abnormal changes in metrics, device, computer device, and storage medium, the implementation of the method includes: obtaining problem information, decomposing the problem information into intent, business metrics, and metric dimensions; determining whether metric attribution is required for the user problem; if metric attribution is required, determining whether the business metric is abnormal; if the business metric is abnormal, determining an attribution diagnosis model according to the intent and metric dimensions, so as to predict the attribution result of the business metric through the attribution diagnosis model; determining the target external factors and target internal factors affecting the attribution result, and obtaining the countermeasures for the target internal factors or the target external factors. In the embodiments of the present application, through dialogue questions, introducing a large language model and an attribution analysis model, the reason for the abnormal fluctuation of the metric can be quickly located, and minute-level attribution analysis can be achieved through the dialogue method. Moreover, by querying the influence factor knowledge base, the internal and external factors affecting the metric fluctuation can be automatically identified, providing a specific basis for the adjustment of the operation strategy, and having a wide range of application scenarios and potential. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for the description of the embodiments of the present application will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0047] Figure 1 is a flowchart of a method for analyzing the attribution of abnormal changes in metrics in an embodiment of the present application;

[0048] Figure 2 is a structural diagram of an apparatus for analyzing the attribution of abnormal changes in metrics in an embodiment of the present application;

[0049] Figure 3 is a schematic diagram of a computer device in an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0050] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0051] In one embodiment, as Figure 1 shown, a method for analyzing the attribution of abnormal changes in indicators is provided, including the following steps:

[0052] In step S110, obtain problem information and decompose the problem information into intent, business indicators, and indicator dimensions;

[0053] In the embodiments of the present application, the problem input by the user can be decomposed through a large language model, such as GPT4o. Specifically, a prompt template can be set in advance. When the problem information is obtained, the attribution indicator corresponding to the problem information can be determined, and the attribution dimension metadata corresponding to the attribution indicator can be queried. Then, through this prompt, the attribution dimension metadata, and the problem information, a semantic analysis request is generated, and based on the accurate semantic analysis of this GPT4o, a DSL object is obtained. An attribution analysis request is generated according to the DSL object, and the indicator data is queried from the database.

[0054] In step S120, determine whether the user's problem requires indicator attribution;

[0055] In the embodiments of the present application, the understanding ability and rules of the large language model can be called. For example, keywords can be configured, such as fluctuations, anomalies, attribution analysis, abnormal, etc., to determine whether indicator attribution is required. Exemplarily, if the question is "Why has the xxx indicator dropped so much?", the large language model can identify the abnormal fluctuation of the indicator and whether attribution analysis is required.

[0056] In step S130, if indicator attribution is required, determine whether the business indicator is abnormal;

[0057] In the embodiments of the present application, abnormal identification can be performed on the current date corresponding to the indicator. Abnormal identification can be carried out from two aspects. In the first aspect, the overall trend of the indicator can be observed to determine whether there is an anomaly. For example, the mean value of the business data corresponding to each indicator within a preset time range, such as the daily mean value, can be calculated, and then a line graph can be established based on this mean value. According to the overall trend of the line in the line graph, it can be determined whether there is a large fluctuation. In the second aspect, it can be determined whether there is an anomaly in the business indicator data corresponding to each dimension. That is, a line graph can be established for the indicator data of each dimension, and then it can be observed whether there is a large fluctuation in the trend of the line. If there is a large fluctuation, it indicates that the business indicator is abnormal.

[0058] In the embodiments of the present application, the specified periods of indicators can be compared and analyzed to determine whether there are anomalies, such as day-on-day, week-on-week, month-on-month, year-on-year, etc., or other specified times, holidays, etc. The corresponding values and difference values of the indicator data in two different periods are calculated. If the difference value is greater than the preset difference threshold, it indicates that there is an anomaly.

[0059] It should be noted that anomaly identification can be performed on the current date corresponding to the indicator. If there is no anomaly, the specified periods of the indicator can be compared and analyzed. Or, first, the specified periods of the indicator are compared and analyzed. If it is determined that the indicator has no anomaly, then anomaly identification is performed again on the current date corresponding to the indicator. Or, anomaly identification of the current date corresponding to the indicator and comparison and analysis of the specified periods of the indicator are performed simultaneously. Or, either one can be selected for anomaly judgment of business indicators.

[0060] In step S140, if the business indicator is abnormal, according to the intention and indicator dimension, an attribution diagnosis model is determined to predict the attribution result of the business indicator through the attribution diagnosis model.

[0061] Among them, the attribution diagnosis model can include a single-dimension diagnosis model, a composite-index blood relationship diagnosis model, and a dimension-crossing attribution diagnosis model. All indicators can be directly diagnosed through the single-dimension diagnosis model. If the indicator is a composite indicator, it can be accurately diagnosed through the composite-index blood relationship diagnosis model.

[0062] When the business indicator is abnormal, one or more attribution diagnosis models can be determined according to the intention of indicator attribution and the type of the indicator, and the attribution result of the business indicator can be predicted through the attribution diagnosis model. For example, when the business indicator is a composite indicator, the composite-index attribution blood relationship diagnosis model can be selected for predicting the attribution result.

[0063] In step S150, the target external factors and target internal factors affecting the attribution result are determined, and the countermeasures for the target internal factors or the target external factors are obtained.

[0064] In the embodiments of the present application, the internal factors refer to factors such as internal operation strategies and technical changes. The operation strategy specifically can include operations such as distributing red envelopes / driver subsidies (which will affect indicators such as the number of user orders and driver order acceptance), price cuts (driver response rate, order acceptance rate indicators), etc. The technical change refers to new product functions (affecting user and driver experience indicators), technical failure problems (affecting user and driver usage indicators, order placement and acceptance indicators, etc.).

[0065] Among them, the external factors refer to factors such as road conditions, weather, and traffic.

[0066] In the embodiments of the present application, the business database can be queried to obtain internal factors such as operation strategies, activities, and technology release records related to the metric dimension, so as to determine whether it is the attribution result affecting the business metrics. For example, a corresponding impact level can be configured for each internal factor, and then according to the impact level, a corresponding weight is assigned to each internal factor. The higher the impact level, the greater the weight and the higher the impact coefficient. Similarly, corresponding impact levels can be configured for external factors and weights can be assigned. Based on this, the influence of internal and external factors on the attribution result can be determined. It should be noted that an attribution routing engine can be set up to associate the metrics with internal and external factors. When the metrics fluctuate, the weights of internal and external factors can be judged, so as to judge the internal and external factors that ultimately affect the attribution result.

[0067] In the embodiments of the present application, external factors such as weather, such as rain, snow, etc., will affect the supply and demand metrics between drivers and consumers. Road conditions, such as congestion and traffic restrictions, will affect the actual average order completion time metric. For policies, such as holding activities, it affects the driver distribution metric, etc. Therefore, different countermeasures can be taken for different external factors. For example, for weather, the order price can usually be increased to attract drivers to take orders. For road conditions, the transportation capacity can be dispatched to let drivers in other areas take orders. For policies, order-taking subsidies can be provided to dispatch vehicles to areas with fewer vehicles.

[0068] In the embodiments of the present application, internal factors such as operation strategies and product changes. Operation strategies, such as issuing red envelopes and subsidies, will affect the consumer order placement metric and the driver order-taking metric. And price cuts will affect the driver response rate and order-taking rate metrics. Product changes, such as new product features, will affect the consumer and driver experience metrics. Technical failures will affect the usage metrics of consumers and drivers. Therefore, for the above internal factors, product functions can be managed, or technical failures can be repaired, and emergency scaling can be carried out, etc.

[0069] In an embodiment of the present application, a method for analyzing the attribution of abnormal changes in indicators is provided, including: obtaining problem information, decomposing the problem information into intent, business indicators, and indicator dimensions; determining whether indicator attribution is required for the user problem; if indicator attribution is required, determining whether the business indicator is abnormal; if the business indicator is abnormal, determining an attribution diagnosis model according to the intent and the indicator dimension, so as to predict the attribution result of the business indicator through the attribution diagnosis model; determining the target external factors and target internal factors that affect the attribution result, and obtaining the countermeasures for the target internal factors or the target external factors. In an embodiment of the present application, through dialogue questions, introducing a large language model and an attribution analysis model, the reason for abnormal fluctuations in indicators can be quickly located, and minute-level attribution analysis can be achieved. Minute-level attribution analysis is achieved through a dialogue method. Moreover, by querying the influencing factor knowledge base, the internal and external factors that affect indicator fluctuations are automatically identified, providing a specific basis for the adjustment of operation strategies, and having a wide range of application scenarios and potential.

[0070] In an embodiment of the present application, the determining whether the business indicator is abnormal includes:

[0071] determining whether the business indicator is abnormal at the current time; and / or

[0072] determining whether the numerical difference of the business data corresponding to the business indicator in different comparison periods is greater than a preset difference threshold.

[0073] Optionally, to determine whether the business indicator is abnormal at the current time, abnormal identification can be carried out from two aspects. On the first aspect, the overall trend can be observed through the indicator to see if there is any abnormality. For example, the mean value of the business data corresponding to each indicator within a preset time range, such as the daily mean value, can be calculated, and then a line graph can be established based on this mean value, and whether there is a large fluctuation in the overall trend of the line in the line graph can be observed. On the second aspect, it can be determined whether the business indicator data corresponding to each dimension is abnormal, that is, a line graph can be established for the indicator data of each dimension, and then whether there is a large fluctuation in the trend of the line can be observed. If there is a large fluctuation, it indicates that the business indicator is abnormal.

[0074] Determine whether the numerical difference of the business data corresponding to the business indicator in different comparison periods is greater than a preset difference threshold, such as daily comparison, week-on-week comparison, month-on-month comparison, year-on-year comparison, etc., or other specified times, holidays, etc., and calculate the corresponding numerical values and difference values of the indicator data in two different periods. If the difference value is greater than the preset difference threshold, it indicates an anomaly. It should be noted that according to the problem information, the identified indicators and dates, the corresponding numerical values can be queried in the doris database table. For example: To calculate the week-on-week comparison of the order completion volume indicator, the indicator values of the order completion volume today and 7 days ago can be queried in the doris database, and then, the difference value = (current value - benchmark value) / benchmark value * 100%. Where the current value refers to the indicator value in the current period, and the benchmark value refers to the indicator value in the same period last year.

[0075] In an embodiment of the present application, the determination of whether there is an anomaly in the business indicator at the current time includes:

[0076] Determine the mean value of the business data corresponding to all business indicators within each preset time range;

[0077] Determine whether the difference between the mean value and the mean value of the business data corresponding to all business indicators within the adjacent preset time range is greater than a preset difference;

[0078] If so, the business indicator has an anomaly;

[0079] If not, determine the change trend of the business data corresponding to each indicator dimension;

[0080] When the difference between the peak value and the trough value of the change trend is greater than a preset threshold, the business indicator has an anomaly.

[0081] Optionally, the mean value of the business data corresponding to all business indicators within each preset time range can be calculated, such as the mean value of the business data within each day, that is, the daily mean value, and then a line graph can be established. According to the overall trend of the line in the line graph, determine whether there is a large fluctuation. It can be determined whether the difference between different daily mean values is large. If so, it can indicate that there is an anomaly in the business data. If not, the numerical values of the business data corresponding to each indicator dimension can be further calculated, and a line graph can be established. Based on the change trend of the line, determine whether there is a large fluctuation, that is, when the difference between the peak value and the trough value of the line is greater than a preset threshold. If so, it can indicate that there is an anomaly in the business indicator.

[0082] In an embodiment of the present application, the attribution diagnosis model is a single-dimensional diagnosis model. The attribution result of the business indicator predicted by the attribution diagnosis model includes:

[0083] Calculate the contribution value of the business data corresponding to each indicator dimension respectively;

[0084] Sort the contribution values in ascending or descending order to obtain the index dimension with the highest contribution value as the attribution result of the service metric.

[0085] Optionally, the service data can be grouped according to the index dimension, and then the contribution value of the service data under each index dimension can be calculated. The contribution value can be obtained through the paired weight fluctuation algorithm or the fluctuation contribution algorithm. According to the calculated contribution value of each index dimension, an ascending or descending order can be performed, and then the maximum contribution value can be used as the result of this attribution analysis.

[0086] Among them, the paired weight fluctuation algorithm is specifically as follows: a single weight for each dimension pair can be set. For example, the proportion of the paired single quantity of dimension m to the reference date / (1 - the proportion of the paired single quantity of dimension m to the reference date of the reference day), then the contribution value = (the change rate of dimension m - the overall change rate) * the paired single weight of the dimension.

[0087] Among them, the fluctuation contribution algorithm is specifically as follows: the fluctuation contribution algorithm for absolute value indicators: the change amount of dimension m divided by the total reference value. The fluctuation contribution algorithm for rate value indicators: (the rate value of dimension m in this period - the overall reference rate value) * (the absolute value of dimension m in the denominator of the rate value in this period / the overall absolute value of the denominator of the rate value in this period) - (the reference rate value of dimension m - the overall reference rate value) * (the absolute value of dimension m in the denominator of the reference rate value / the overall absolute value of the denominator of the reference rate value). The rate value of dimension m in this period refers to the ratio or proportion of dimension m within this week, and the rate value in this period refers to a certain ratio or proportion within the current cycle or time period.

[0088] In an embodiment of the present application, the attribution diagnosis model is a composite index blood relationship diagnosis model. By using the attribution diagnosis model, the attribution result of the service metric is predicted, including:

[0089] When the service metric is a composite index, query the upstream index of the service metric to obtain multiple parent indexes of the service metric;

[0090] Calculate the influence value of each parent index;

[0091] Sort the influence values in ascending or descending order to obtain the parent index with the highest influence value as the attribution result of the service metric.

[0092] Optionally, after obtaining the business metric, it can be determined whether it is a composite metric. If so, the lineage of the metric, i.e., its upstream metrics, can be queried in the metric system. For example, the average order value = total order amount / total number of orders. The parent metrics of the average order value can be the total order amount and the total number of orders. Then, the influence value can be calculated using the method of successive substitution. After that, the influence values of each parent metric can be sorted in ascending or descending order to obtain the parent metric with the highest influence value, which is used as the attribution result of the business metric.

[0093] The calculation process of the method of successive substitution is as follows:

[0094] Suppose an analysis metric M is obtained from three interrelated factors A, B, and C. The current metric and the comparison metric are:

[0095] Current metric M1 = A1 * B1 * C1

[0096] Comparison metric M0 = A0 * B0 * C0

[0097] When measuring the influence degree of each factor change metric on the metric, it needs to be sorted in a certain order (from high to low in importance, and the sorting order will affect the final result):

[0098] Comparison metric M0 = A0 * B0 * C0...(1)

[0099] First substitution A1 * B0 * C0...(2)

[0100] Second substitution A1 * B1 * C0...(3)

[0101] Third substitution A1 * B1 * C1...(4)

[0102] (2)-(1) → The influence of A change on M.

[0103] (3)-(2) → The influence of B change on M.

[0104] (4)-(3) → The influence of C change on M.

[0105] Total influence: △M = M1 - M0 = (4)-(3)+(3)-(2)+(2)-(1).

[0106] For example, if the sales amount = number of exposures * conversion rate * average order value decreased year-on-year, the influence of each metric can be determined using the above method of successive substitution. The details are shown in the following table:

[0107] Table 1: Calculation result table of the method of successive substitution

[0108]

[0109] Based on Table 1 above, we can see that the final number of exposures affected is 22,400, the conversion rate affected is -9,600, and the average order value affected is 14,400.

[0110] In one embodiment of the present application, the attribution diagnosis model is a dimensional cross-attribution diagnosis model. The attribution result of the business indicator is predicted by the attribution diagnosis model, including:

[0111] Combine the indicator dimensions to obtain multiple groups of multidimensional indicator dimensions;

[0112] Calculate the contribution value of each indicator dimension under multiple groups of multi-dimensional indicator dimensions;

[0113] The set of indicator dimensions with the largest contribution value is used as the attribution result of the business indicator.

[0114] Optionally, the business data can be grouped according to the indicator dimension, and then the contribution value under each indicator dimension can be calculated. For example, if there are 3 dimensions, after combining them in pairs, 6 multi-dimensional indicator dimension combinations can be obtained. Then the business data under each indicator dimension can be calculated for contribution value twice.

[0115] The contribution value can be calculated by a paired weight fluctuation algorithm, or it can be obtained by a fluctuation contribution algorithm. According to the calculated contribution value of each indicator dimension, it can be arranged in ascending or descending order, and then the group of indicator dimensions with the largest contribution value can be used as the result of this attribution analysis.

[0116] In one embodiment of the present application, problem information is obtained and decomposed into intent, business indicators, and indicator dimensions, including:

[0117] Setting prompt words, wherein the prompt words include dimension information and indicator metadata;

[0118] Obtaining the problem information, and determining the indicator to be attributed based on the problem information;

[0119] Querying a database for attribution dimension metadata corresponding to the indicator to be attributed;

[0120] Based on the prompt word, question information and the attribution dimension metadata, semantic analysis is performed using a large language model to obtain a DSL object;

[0121] The database is queried according to the DSL object to obtain indicator data.

[0122] Optionally, a prompt template (prompt word) can be set in advance. The prompt contains information such as the dimensions and indicator metadata of the business data. When question information is obtained, the indicator to be attributed to the question information can be determined, and the attribution dimension metadata corresponding to the attribution dimension can be queried in the database. Then, based on the question information, prompt template, and attribution dimension metadata, a semantic analysis request is generated. The semantic analysis request is processed using the large language model GPT4o to obtain a DSL object. Based on the DSL object, the corresponding attribution analysis request is generated, and query SQL is generated to query the database to obtain the indicator data.

[0123] For example, a user asks "Why has the order volume for freight trucks in Region A decreased compared to last week?"

[0124] By utilizing the information extraction capabilities of the big model, we can obtain the parsing results of the big model in Json format. The big model identifies dimensions (city / business line / vehicle type, etc.), indicators, and intentions based on the prompt. The dimensions are: City: Region A, Business Line: Freight Car, Indicator to be attributable: Order volume, Current time: Time: 2024-11-18, Same period time: Time: 2024-11-11, Intention: Attribution analysis of the reasons for the decline.

[0125] In the embodiments of the present application, by means of dialogue questions, a large language model and an attribution diagnosis model are introduced, which can quickly locate the reasons for abnormal fluctuations of indicators, achieve minute-level attribution analysis, and the results are single-dimensional diagnosis, models such as composite indicator lineage diagnosis and multi-dimensional cross diagnosis can improve the accuracy and efficiency of attribution analysis. By querying the knowledge base for external and internal factors affecting indicator fluctuations and obtaining corresponding measures for processing, effective decision-making suggestions are provided. Due to its efficient and accurate characteristics, the present application can be adapted to various scenarios, such as: Enterprise operation monitoring: In the daily operation of an enterprise, the fluctuations of various key indicators (such as sales volume, number of users, conversion rate, etc.) are crucial for operation decisions. Through the above solution, abnormal fluctuations of these indicators can be quickly identified, and attribution analysis can be automatically given to help the operation team quickly adjust strategies and ensure the stable operation of the enterprise. Financial risk management: In the financial field, changes in market risks, credit risks, etc. often lead to fluctuations in various financial indicators. Through the above solution, financial institutions can monitor abnormal changes in these indicators in real time, quickly identify potential risks, and take corresponding risk management measures. E-commerce platform analysis: Indicators such as product sales volume and user behavior on an e-commerce platform are crucial for merchants. Through the above solution, merchants can quickly identify the reasons for problems such as decreased sales volume and user loss, so as to adjust marketing strategies and improve user experience and sales volume. Medical and health monitoring: In the medical and health field, fluctuations in various physiological indicators of patients (such as heart rate, blood pressure, blood sugar, etc.) are crucial for doctors' diagnosis and treatment. Although the above solution is mainly applied to the attribution analysis of non-medical data, its concept and technology can also be borrowed for medical and health monitoring. By real-time monitoring of patients' physiological indicators, abnormal fluctuations can be quickly identified, providing timely diagnosis and treatment suggestions for doctors. Supply chain management: In supply chain management, indicators such as inventory turnover rate and order processing time are crucial for the operation efficiency of an enterprise. Through the above solution, an enterprise can quickly identify bottlenecks and problems in the supply chain, so as to optimize the supply chain management process and improve operation efficiency. In these scenarios, through the implementation of the above solution, it can help professionals in related fields quickly understand the reasons behind data changes, so as to make more accurate and timely decisions.

[0126] It should be understood that the magnitudes of the sequence numbers of the steps in the above embodiments do not mean the order of execution is prior or subsequent. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0127] In one embodiment, an apparatus for analyzing the attribution of abnormal indicator changes is provided, and this apparatus for analyzing the attribution of abnormal indicator changes corresponds one-to-one with the method for analyzing the attribution of abnormal indicator changes in the above embodiments. As Figure 2As shown in the figure, the device for analyzing the attribution of abnormal indicators includes a problem information decomposition unit 10, an indicator attribution determination unit 20, an indicator anomaly analysis unit 30, an attribution result prediction unit 40, and a countermeasure determination unit 50. The detailed description of each functional module is as follows:

[0128] The problem information decomposition unit 10 is used to obtain problem information and decompose the problem information into intentions, business indicators, and indicator dimensions;

[0129] The indicator attribution determination unit 20 is used to determine whether indicator attribution needs to be performed on the user problem;

[0130] The indicator anomaly analysis unit 30 is used to determine whether the business indicator is abnormal if indicator attribution needs to be performed;

[0131] The attribution result prediction unit 40 is used to determine an attribution diagnosis model according to the intention and indicator dimension if the business indicator is abnormal, so as to predict the attribution result of the business indicator through the attribution diagnosis model;

[0132] The countermeasure determination unit 50 is used to determine the target external factors and target internal factors affecting the attribution result, and obtain the countermeasures for the target internal factors or the target external factors.

[0133] In an embodiment of the present application, the indicator anomaly analysis unit 30 is further used to:

[0134] Determine whether the business indicator is abnormal at the current time; and / or

[0135] Determine whether the numerical difference of the business data corresponding to the business indicator in different comparison periods is greater than a preset difference threshold.

[0136] In an embodiment of the present application, the indicator anomaly analysis unit 30 is further used to:

[0137] Determine the mean value of the business data corresponding to all business indicators within each preset time range;

[0138] Determine whether the difference between the mean value and the mean value of the business data corresponding to all business indicators within the adjacent preset time range is greater than a preset difference;

[0139] If so, the business indicator is abnormal;

[0140] If not, determine the change trend of the business data corresponding to each indicator dimension;

[0141] When the difference between the peak value and the trough value of the change trend is greater than a preset threshold, the business indicator is abnormal.

[0142] In an embodiment of the present application, the attribution diagnosis model is a single-dimensional diagnosis model, and the attribution result prediction unit 40 is further configured to:

[0143] The attribution diagnosis model is a single-dimensional diagnosis model, and predicting the attribution result of the service metric through the attribution diagnosis model includes:

[0144] Calculate the contribution value of the service data corresponding to each metric dimension respectively;

[0145] Sort the contribution values in ascending or descending order to obtain the metric dimension with the highest contribution value as the attribution result of the service metric.

[0146] In an embodiment of the present application, the attribution diagnosis model is a composite metric lineage diagnosis model, and the attribution result prediction unit 40 is further configured to:

[0147] When the service metric is a composite metric, query the upstream metrics of the service metric to obtain multiple parent metrics of the service metric;

[0148] Calculate the influence value of each parent metric;

[0149] Sort the influence values in ascending or descending order to obtain the parent metric with the highest influence value as the attribution result of the service metric.

[0150] In an embodiment of the present application, the attribution diagnosis model is a dimension cross-attribution diagnosis model, and the attribution result prediction unit 40 is further configured to:

[0151] Combine the metric dimensions to obtain multiple groups of multi-dimensional metric dimensions;

[0152] Calculate the contribution value of each metric dimension under multiple groups of multi-dimensional metric dimensions;

[0153] Take the group of metric dimensions with the largest contribution value as the attribution result of the service metric.

[0154] In an embodiment of the present application, the problem information decomposition unit 10 is further configured to:

[0155] Set prompt words, which include dimension information and metric metadata;

[0156] Obtain the problem information, and determine the metric to be attributed according to the problem information;

[0157] Query the attribution dimension metadata corresponding to the metric to be attributed in the database;

[0158] Based on the prompt words, problem information, and the attribution dimension metadata, perform semantic analysis through a large language model to obtain a DSL object;

[0159] Query the database according to the DSL object to obtain index data.

[0160] In the embodiments of the present application, by means of dialogue questions, a large language model and an attribution diagnosis model are introduced, which can quickly locate the reasons for abnormal fluctuations of indicators, achieve minute-level attribution analysis, and the results are single-dimensional diagnosis, compound index lineage diagnosis, multi-dimensional cross-diagnosis and other models, which can improve the accuracy and efficiency of attribution analysis. Query the external and internal factors affecting the index fluctuations through the knowledge base, and obtain corresponding measures for processing, providing effective decision-making suggestions. Due to its high efficiency and accuracy, the present application can be adapted to various scenarios.

[0161] For the specific limitations of the index anomaly attribution analysis device, reference can be made to the limitations of the index anomaly attribution analysis method in the above text, which will not be elaborated here. Each module in the above index anomaly attribution analysis device can be implemented in whole or in part by software, hardware and their combination. The above modules can be embedded in the processor of the computer device in hardware form or independent of it, or stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to the above modules.

[0162] In one embodiment, a computer device is provided. The computer device can be a terminal device, and its internal structure diagram can be as Figure 3 shown. The computer device includes a processor, a memory, and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a readable storage medium. The readable storage medium stores computer-readable instructions. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer-readable instructions are executed by the processor, an index anomaly attribution analysis method is implemented. The readable storage medium provided in this embodiment includes a non-volatile readable storage medium and a volatile readable storage medium.

[0163] In the embodiments of the present application, a computer device is provided, including a memory, a processor, and computer-readable instructions stored in the memory and executable on the processor. When the processor executes the computer-readable instructions, the steps of the index anomaly attribution analysis method as described above are implemented.

[0164] In the embodiments of the application, a readable storage medium is provided. The readable storage medium stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the steps of the index anomaly attribution analysis method as described above are implemented.

[0165] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through computer-readable instructions. The computer-readable instructions can be stored in a non-volatile readable storage medium or a volatile readable storage medium. When the computer-readable instructions are executed, they can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium used in the various embodiments provided in this application can include non-volatile and / or volatile memories. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or an external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0166] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the above-mentioned division of each functional unit and module is used as an example. In actual applications, the above functions can be allocated to different functional units and modules according to needs, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0167] The above embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A method for attribution analysis of indicator changes, characterized in that: The method comprises: Obtain problem information and decompose the problem information into intent, business indicators, and indicator dimensions; Determine whether user issues require metric attribution; If indicator attribution is required, determine whether there are any abnormalities in the business indicators; If the business indicator is abnormal, an attribution diagnosis model is determined based on the intention and indicator dimension, so as to predict the attribution result of the business indicator through the attribution diagnosis model. The attribution diagnosis model is a composite indicator lineage diagnosis model. The attribution result of the business indicator is predicted through the attribution diagnosis model, including: When the business indicator is a composite indicator, query the upstream indicator of the business indicator to obtain multiple parent indicators of the business indicator; Based on the parent indicator, determine the current indicator and the comparative indicator; Sort by importance of each indicator, and replace the corresponding indicators in the comparison indicators in sequence based on the sorted indicators; Determine the impact of each replacement on the business indicator to determine the impact value of each parent indicator; Sort the impact values in ascending or descending order to obtain the parent indicator with the highest impact value as the attribution result of the business indicator; Determine the target external factors and target internal factors that affect the attribution result, and obtain response measures for the target internal factors or the target external factors.

2. The indicator change attribution analysis method according to claim 1, characterized in that: Determining whether the business indicator is abnormal includes: Determine whether there is an anomaly in the business indicator at the current time; and / or Determine whether the difference in values of the business data corresponding to the business indicator in different comparison periods is greater than a preset difference threshold.

3. The indicator change attribution analysis method according to claim 2, characterized in that: Determining whether the business indicator at the current time is abnormal includes: Determine the mean of all business indicators corresponding to business data within each preset time range; Determine whether the difference between the mean and the mean of the business data corresponding to all business indicators within an adjacent preset time range is greater than a preset difference; If so, then the business indicator is abnormal; If not, determine the changing trend of the business data corresponding to each indicator dimension; When the difference between the peak value and the trough value of the change trend is greater than a preset threshold, the business indicator is abnormal.

4. The indicator change attribution analysis method according to claim 1, characterized in that: The obtaining of problem information and decomposing the problem information into intent, business indicators, and indicator dimensions include: Setting prompt words, wherein the prompt words include dimension information and indicator metadata; Obtaining the problem information, and determining the indicator to be attributed based on the problem information; Querying a database for attribution dimension metadata corresponding to the indicator to be attributed; Based on the prompt words, question information and attribution dimension metadata, semantic analysis is performed through a large language model to obtain a DSL object; The database is queried according to the DSL object to obtain indicator data.

5. An indicator change attribution analysis device, characterized in that: The device comprises: A problem information decomposition unit, configured to obtain problem information and decompose the problem information into intent, business indicators, and indicator dimensions; An indicator attribution determination unit, used to determine whether a user problem requires indicator attribution; An indicator anomaly analysis unit, used to determine whether the business indicator is abnormal if indicator attribution is required; an attribution result prediction unit, configured to determine an attribution diagnosis model based on the intention and indicator dimension if the business indicator is abnormal, so as to predict the attribution result of the business indicator through the attribution diagnosis model; The attribution diagnosis model is a composite indicator blood relationship diagnosis model, and the attribution result prediction unit is further used to: When the business indicator is a composite indicator, query the upstream indicator of the business indicator to obtain multiple parent indicators of the business indicator; Based on the parent indicator, determine the current indicator and the comparative indicator; Sort by importance of each indicator, and replace the corresponding indicators in the comparison indicators in sequence based on the sorted indicators; Determine the impact of each replacement on the business indicator to determine the impact value of each parent indicator; Sort the impact values in ascending or descending order to obtain the parent indicator with the highest impact value as the attribution result of the business indicator; The countermeasure determination unit is used to determine the target external factors and target internal factors that affect the attribution result, and obtain the countermeasures for the target internal factors or the target external factors.

6. A computer device comprising a memory, a processor, and computer-readable instructions stored in the memory and executed on the processor, wherein: When the processor executes the computer-readable instructions, the indicator change attribution analysis method according to any one of claims 1 to 4 is implemented.

7. A computer-readable storage medium having computer-readable instructions stored thereon, characterized in that: When the computer-readable instructions are executed by a processor, the indicator change attribution analysis method according to any one of claims 1 to 4 is implemented.

Images