An attribution analysis method and system applied to complex indicators
By disassembling the complex index model, analyzing the redundancy and correlation relationship of index components, and quantifying the coupling effect, the problems of poor adaptability and low accuracy of attribution analysis in the existing technology are solved, and higher analysis reliability and Retrospective.
Patent Information
- Application Number
- CN202411346425.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-26
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2044-09-26
AI Technical Summary
In the prior art, the attribution analysis of complex indicators is poor in adaptability and low accuracy, and the complex role relationship between causes is not effectively analyzed.
By obtaining complex indicator models, disassembly into multiple indicator components, analyzing the redundancy and correlation between components, determining the type and interaction relationship of indicator components, quantifying the coupling effect, and visually displaying it to build business links and Mapping relationships between metric components.
It improves the adaptability and accuracy of attribution analysis of complex indicators, ensures the independence and analysis reliability of indicator components, and realizes the retrospective attribution display of business links.
Smart Images

Figure CN118885912B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of commercial data processing, and in particular to an attribution analysis method and system applied to complex indicators. Background Art
[0002] In the complex and ever-changing commercial and financial fields, attribution analysis solutions use advanced background technologies to accurately identify key factors that affect business results by deeply tracking and analyzing multi-dimensional and multi-level data indicators. These technologies not only cover the detailed decomposition of atomic indicators and composite indicators, but also combine multiple algorithms such as time series analysis to ensure that the root causes of problems can be quickly located in the complex business decision-making process, assisting management in achieving data-driven intelligent decision-making.
[0003] In the prior art, only the causes of complex indicators are often given, without analyzing the complex interactions between the causes, resulting in poor adaptability and low accuracy of attribution analysis.
[0004] Therefore, how to improve the adaptability and accuracy of attribution analysis of complex indicators is an issue that needs to be addressed urgently. Summary of the invention
[0005] The purpose of the present invention is to solve the problems of poor adaptability and low accuracy of attribution analysis of complex indicators in the prior art, and to propose an attribution analysis method applied to complex indicators, including:
[0006] Obtain complex indicator models and business objectives in the financial or commercial fields, obtain relevant business process flows through mapping the complex indicators to business objectives, decompose the complex indicator models according to the dimensional levels, obtain multiple indicator components, and compare the relevant business process flows with multiple indicator components to supplement the indicator components and build a mapping relationship between the relevant business process flows and indicator components;
[0007] Analyze the redundancy between indicator components, and select and retain some indicator components by screening the indicator components according to the redundancy. Then, classify the types of indicator components according to the relationship between the complex indicator and each indicator component, associate the indicator components according to the type of indicator components, and determine the association relationship between the indicator components.
[0008] Determine the interaction relationship between indicator components based on the association relationship between indicator components, and determine whether there is a coupling effect in the interaction relationship between indicator components. If there is a coupling effect, analyze and quantify the coupling effect;
[0009] Based on the mapping relationship between relevant business link processes and indicator components, indicator components, interaction relationships, and coupling effects are visualized to achieve attribution display of business links.
[0010] In some embodiments of the present application, the complex indicator model is disassembled according to the dimensional level to obtain multiple indicator components, including:
[0011] Determine the function of the complex indicator according to the complex indicator model, decompose the function of the complex indicator into multiple variables, confirm the dimensional level to which each variable belongs, find several other variables close to the variable at the same dimensional level, and calculate the stability of each variable and several other variables close to it;
[0012] The judgment period of each variable is assigned through the attributes of the variable, and the coefficient of variation of the variable under each judgment period is integrated to determine the stability of each variable. The value of the variable and its stability are used as indicator components.
[0013] In some embodiments of the present application, the redundancy between the analysis indicator components includes:
[0014] According to the values of the indicator component variables, a variable curve graph of the indicator component variables changing over time is constructed, the text keywords of each indicator component variable are determined, and the similarity of the text keywords of the variables between the indicator components is compared, so as to determine the candidate indicator components with similar text relationships;
[0015] Compare the variable curve graphs between the candidate indicator components with similar textual relationships to obtain the variable curve similarity, and calculate the redundancy between the candidate indicator components with similar textual relationships according to the variable curve similarity of the candidate indicator components with similar textual relationships, the similarity of text keywords and the stability of variables; ;in, is the redundancy between candidate indicator components with similar textual relations, is the redundant weight of the variable curve similarity of the candidate indicator components with similar textual relations, is the variable curve similarity of the candidate indicator components with similar textual relations, is the redundant weight of the similarity of the text keywords of the candidate indicator components with similar textual relations, is the similarity of the text keywords of the candidate indicator components with similar textual relations, is the redundant weight of the stability of the variables of the candidate indicator components with similar textual relations, is the conversion factor, , The stability of variables for candidate indicator components with similar textual relationships.
[0016] In some embodiments of the present application, the indicator components are screened and selected by redundancy to retain some indicator components, including:
[0017] If the redundancy exceeds the redundancy threshold, the candidate indicator component with the lower variable stability among the candidate indicator components with similar textual relations is eliminated, and the candidate indicator component with the higher variable stability is retained;
[0018] Otherwise, all candidate indicator components in the candidate indicator components having similar textual relations are retained.
[0019] In some embodiments of the present application, the types of indicator components are divided according to the relationship between the complex indicator and each indicator component, including:
[0020] The types of indicator components include direct indicator components, semi-direct indicator components, and indirect indicator components;
[0021] Construct a fishbone diagram of complex indicators based on the relevant business process and indicator components, thereby determining the fish head and fish bones in the fishbone diagram. The fish head is the complex indicator, and the fish bones are the indicator components. Convert the fishbone diagram into an indicator component impact path diagram. The indicator components are the nodes in the impact path diagram, and the complex indicators are the end points of the path.
[0022] Calculate the correlation between two adjacent nodes with a connected relationship, and define the path length between the two nodes by the correlation, divide the index component of the node directly connected to the path end point into a direct index component, and calculate the path length from the node to the path end point, divide the index component of the node whose path length is within a preset length range into a semi-direct index component, and determine the route from the node to the path end point;
[0023] The indicator components of other nodes are divided into indirect indicator components, and the route from the node to the end point of the path is determined.
[0024] In some embodiments of the present application, indicator components are associated according to their types to determine the association relationship between indicator components, including:
[0025] The corresponding indicator components are associated through direct indicator components, the corresponding indicator components are associated through the path routes involved in semi-direct indicator components, and the corresponding indicator components are associated through the path routes involved in indirect indicator components.
[0026] In some embodiments of the present application, determining whether there is a coupling effect in the interaction relationship between indicator components includes:
[0027] Standardize several variables involved in the interaction relationship between indicator components, calculate the covariance matrix between several variables, solve the covariance matrix to obtain eigenvalues and eigenvectors, and thereby obtain multiple new variables. Weights are assigned to the new variables through eigenvalues, and weighted summation is performed on the new variables to obtain the initial comprehensive representative indicator.
[0028] The initial comprehensive representative index is adjusted by the stability of several variables involved in the interaction relationship between the indicator components to obtain a comprehensive representative index; ;in, As a comprehensive representative indicator, is the initial comprehensive representative indicator, is the number of indicator components that have interaction relationships, For the interaction relationship The stability of the indicator components, is the maximum stability value of the indicator components with interaction relationship, Volatility factor, is the volatility factor obtained by mapping the stability maximum of the indicator components that interact with each other. is a preset constant; the nonlinear dependence degree between the comprehensive representative indicators and the complex indicators is calculated, and the presence of a coupling effect is determined by the nonlinear dependence degree.
[0029] In some embodiments of the present application, the method further includes:
[0030] If there is no coupling effect, the indicator components and their interaction relationships are visualized based on the mapping relationship between the relevant business process and the indicator components.
[0031] Correspondingly, the present application also provides an attribution analysis system applied to complex indicators, including:
[0032] The first module is used to obtain complex indicator models and business objectives in the financial or commercial fields, obtain relevant business process links through mapping the business objectives of complex indicators, decompose the complex indicator model according to the dimensional level, obtain multiple indicator components, and compare the relevant business process links with multiple indicator components to supplement the indicator components and build a mapping relationship between the relevant business process links and indicator components;
[0033] The second module is used to analyze the redundancy between indicator components, and select and retain some indicator components by screening the indicator components according to the redundancy, and then divide the types of indicator components according to the relationship between the complex indicator and each indicator component, and associate the indicator components according to the type of indicator components to determine the association relationship between the indicator components;
[0034] The third module is used to determine the interaction relationship between the indicator components according to the association relationship between the indicator components, and determine whether there is a coupling effect in the interaction relationship between the indicator components. If there is a coupling effect, the coupling effect is analyzed and quantified;
[0035] The fourth module is used to visualize the indicator components, interaction relationships, and coupling effects based on the mapping relationship between the relevant business link processes and the indicator components, thereby realizing the attribution display of the business links.
[0036] Compared with the prior art, the present invention has the following beneficial effects:
[0037] 1. By comparing the relevant business process and multiple indicator components, the indicator components are supplemented and a mapping relationship between the two is established to facilitate subsequent comparison.
[0038] 2. The indicator components are screened by redundancy to retain some indicator components and the redundant indicator components with similar meanings are removed, which ensures the independence of the indicator components and improves the reliability of the indicator component analysis.
[0039] 3. The interaction relationship between indicator components is determined based on the correlation relationship between indicator components, and whether there is a coupling effect in the interaction relationship between indicator components is determined. The correlation between indicator components is accurately given, and the interaction relationship between indicator components is obtained. It is also analyzed whether the interaction relationship will produce a coupling effect, which improves the adaptability and accuracy of attribution analysis of complex indicators and ensures the traceability of attribution analysis. BRIEF DESCRIPTION OF THE DRAWINGS
[0040] Figure 1 A schematic diagram of a flow chart of an attribution analysis method applied to complex indicators proposed by the present invention;
[0041] Figure 2 This is a schematic diagram of the structure of an attribution analysis system applied to complex indicators proposed by the present invention. DETAILED DESCRIPTION
[0042] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.
[0043] Reference Figure 1 , an attribution analysis method applied to complex indicators, comprising the following steps:
[0044] Step S101, obtain the complex indicator model in the financial or commercial field and the business objectives targeted by the complex indicators, obtain the relevant business link process by mapping the complex indicators to the business objectives, decompose the complex indicator model according to the dimensional level, obtain multiple indicator components, compare the relevant business link process with the multiple indicator components, so as to supplement the indicator components and construct a mapping relationship between the relevant business link process and the indicator components.
[0045] In this embodiment, the complex indicator model (i.e., the functional relationship of complex indicators) refers to a multi-dimensional, multi-level data indicator system used to reflect business performance, operating conditions, etc. in the financial or commercial fields. These models usually contain multiple variables and parameters, which can comprehensively reflect multiple aspects of a specific business goal or performance. For example, the comprehensive profitability index is a complex indicator that measures the comprehensive profitability of a financial institution over a certain period of time. It comprehensively considers the performance of multiple aspects such as operating income, cost control, asset quality, and capital adequacy ratio. The business process is as follows:
[0046] 1. Product design and development (market share growth rate, product gross profit margin, etc. after new products are launched):
[0047] Design competitive financial products that meet market needs.
[0048] Optimize product pricing strategies to ensure profit margins.
[0049] 2. Marketing and sales (marketing cost input-output ratio, new customer acquisition cost, customer repurchase rate, etc.):
[0050] Develop effective marketing plans to increase brand awareness and product penetration.
[0051] Expand sales channels and increase sales.
[0052] 3. Risk management (non-performing loan ratio, capital adequacy ratio, risk-weighted asset ratio, etc.):
[0053] Establish a sound risk management system to effectively control credit risk, market risk and operational risk.
[0054] Conduct risk assessments regularly and adjust risk management strategies in a timely manner.
[0055] 4. Cost control (ratio of operating costs to total revenue, per capita output value, etc.):
[0056] Optimize operational processes and reduce operating costs.
[0057] Strengthen budget management to ensure that all expenditures are reasonable and efficient.
[0058] 5. Customer service and relationship management (customer satisfaction survey scores, customer complaint handling timeliness, etc.):
[0059] Provide high-quality customer service and enhance customer stickiness.
[0060] Through the customer relationship management system, we can deeply understand customer needs and provide customized services.
[0061] In this embodiment, the original intention of designing complex indicators is to better reflect and track the achievement of business goals. Therefore, there is a close connection between these indicators and business goals. Through in-depth analysis of complex indicators, we can more clearly understand how they correspond to business goals and how they help achieve business goals. The complex indicator model is disassembled according to the dimensional level, and the dimensional level includes time, space, user characteristics, product attributes, channel sources, etc. The relevant business link processes and multiple indicator components are proofread to supplement the indicator components. All related indicators can be added here to ensure the integrity of the indicators. The mapping relationship between the relevant business link processes and the indicator components is that different indicator components correspond to different business links, reflecting a certain part of the business.
[0062] In some embodiments of the present application, the complex indicator model is disassembled according to the dimensional level to obtain multiple indicator components, including:
[0063] Determine the function of the complex indicator according to the complex indicator model, decompose the function of the complex indicator into multiple variables, confirm the dimensional level to which each variable belongs, find several other variables close to the variable at the same dimensional level, and calculate the stability of each variable and several other variables close to it;
[0064] The judgment period of each variable is assigned through the attributes of the variable, and the coefficient of variation of the variable under each judgment period is integrated to determine the stability of each variable. The value of the variable and its stability are used as indicator components.
[0065] In this embodiment, the complex indicator obtained by the complex indicator model is usually composed of multiple sub-indicators or variables through a certain mathematical relationship. It is used to measure the achievement of a specific business goal. It is necessary to identify the mathematical relationship between the sub-indicators that constitute the complex indicator, that is, the function expression of the complex indicator. This function may be a linear combination, a polynomial, a logarithm, an exponential function, or a composite form of these functions. For example, the complex indicator Y may be composed of multiple variables such as X1, X2, ..., Xn through a certain functional relationship f, expressed as Y=f(X1, X2, ..., Xn). Decompose the function into multiple variables. For example, if the function is a polynomial, it is necessary to identify the coefficients of each term and the corresponding variables.
[0066] In this embodiment, at the same dimensional level, several other variables that are close to the variable (may represent similar meanings or related meanings, and can be used by using correlation analysis or cluster analysis to find variables that are closely related to each other at the same dimension) are found to expand the index. Different variables may fluctuate differently. According to the business meaning of the variable, the determination period is allocated, the coefficient of variation under each period is calculated, and the coefficient of variation under multiple periods is integrated (weighted sum or weighted average) to obtain the overall coefficient of variation. The overall coefficient of variation corresponds to the stability of a variable. The indicator component includes two parts: the variable and its stability attributes.
[0067] For example, when the complex indicator is the bank credit card business profit indicator, the functional relationship of the bank credit card business profit indicator (the independent variables of the credit card business interest income, annual fee income, handling fee income, bad debt loss and operating cost) is as follows: Among them, B is the profit index of bank credit card business. is a constant term, which usually represents the basic net profit without any independent variables (i.e., without interest income, annual fee income, commission income, bad debt loss and operating costs). The coefficients representing the interest income, annual fee income, handling fee income, bad debt loss and operating cost of the credit card business, respectively. They represent the interest income, annual fee income, handling fee income, bad debt loss and operating costs of the credit card business respectively. is the tax rate, which is used to calculate the tax payable on pre-tax profit (i.e., profit without considering the impact of tax). Pre-tax profit is obtained by deducting bad debt loss and operating costs from interest income, annual fee income, and commission income. The tax rate multiplied by pre-tax profit (the portion after deducting bad debt loss and operating costs) gives the tax payable, which is then deducted from pre-tax profit to obtain the final net profit. The determination period is allocated according to the business meaning of the variable. Specifically, for example, interest income mainly comes from interest expenses incurred by cardholders who fail to repay on time. Since interest income is closely related to the repayment behavior of cardholders, and the repayment behavior of cardholders may be affected by many factors, such as income level, consumption habits, etc., the changes of these factors are usually not very frequent. Therefore, a relatively long period can be selected to observe the stability of interest income, such as a monthly or quarterly cycle. Commission income includes cash withdrawal fees, installment fees, etc., which are usually related to the specific transaction behavior of cardholders. The fluctuation of commission income may be more frequent because it is directly related to the transaction frequency and amount of cardholders. In order to capture its changing trend more accurately, a shorter judgment period can be selected, such as monthly or weekly.
[0068] It should be noted that when the above-mentioned complex indicators are profit indicators of bank credit card business, their variables can be further divided, or other complex indicators include more variables, which are not displayed one by one due to space constraints.
[0069] Step S102, analyze the redundancy between indicator components, and select and retain some indicator components by screening the indicator components according to the redundancy, and then divide the types of indicator components according to the relationship between the complex indicator and each indicator component, associate the indicator components according to the type of the indicator components, and determine the association relationship between the indicator components.
[0070] In this embodiment, redundancy between indicator components means that different indicators may describe the same or very similar situations, which may lead to duplication and lower accuracy of attribution analysis. Redundancy is defined here and the indicator components are screened.
[0071] In some embodiments of the present application, the redundancy between the analysis indicator components includes:
[0072] According to the values of the indicator component variables, a variable curve graph of the indicator component variables changing over time is constructed, the text keywords of each indicator component variable are determined, and the similarity of the text keywords of the variables between the indicator components is compared, so as to determine the candidate indicator components with similar text relationships;
[0073] Compare the variable curve graphs between the candidate indicator components with similar textual relationships to obtain the variable curve similarity, and calculate the redundancy between the candidate indicator components with similar textual relationships according to the variable curve similarity of the candidate indicator components with similar textual relationships, the similarity of text keywords and the stability of variables; ;in, is the redundancy between candidate indicator components with similar textual relations, is the redundant weight of the variable curve similarity of the candidate indicator components with similar textual relations, is the variable curve similarity of the candidate indicator components with similar textual relations, is the redundant weight of the similarity of the text keywords of the candidate indicator components with similar textual relations, is the similarity of the text keywords of the candidate indicator components with similar textual relations, is the redundant weight of the stability of the variables of the candidate indicator components with similar textual relations, is the conversion factor, , The stability of variables for candidate indicator components with similar textual relationships.
[0074] In this embodiment, the similarity of text keywords of variables between indicator components is compared, and the similarity of text keywords (the similarity of variable meanings) is defined by the similarity or proximity of keywords and keywords. Similar indicator components may also have similar stability, so the deviation of stability is considered here to determine the similarity of stability.
[0075] In this embodiment, NLP tools can be used to segment the description text of the indicator component variables, and TF-IDF (term frequency-inverse document frequency) can be used to calculate the TF-IDF value of each word, where TF is the term frequency and IDF is the inverse document frequency (here, "document" can be regarded as a collection of description texts of all indicator component variables). The TF-IDF (term frequency-inverse document frequency) weighted method can be used to calculate the similarity between keyword sets.
[0076] It should be noted that, generally speaking, there are two candidate indicator components with similar textual relations, so the stability in the redundancy formula is the difference between two parameters. If there are more than two candidate indicator components with similar textual relations, the redundancy formula can be modified to the difference between multiple parameters accordingly.
[0077] In some embodiments of the present application, the indicator components are screened and selected by redundancy to retain some indicator components, including:
[0078] If the redundancy exceeds the redundancy threshold, the candidate indicator component with the lower variable stability among the candidate indicator components with similar textual relations is eliminated, and the candidate indicator component with the higher variable stability is retained;
[0079] Otherwise, all candidate indicator components in the candidate indicator components having similar textual relations are retained.
[0080] In this embodiment, when the redundancy is high, when screening and retaining multiple candidate indicator components, the indicator component corresponding to the one with the highest stability is retained to improve the stability of the indicator component.
[0081] In some embodiments of the present application, the types of indicator components are divided according to the relationship between the complex indicator and each indicator component, including:
[0082] The types of indicator components include direct indicator components, semi-direct indicator components, and indirect indicator components;
[0083] Construct a fishbone diagram of complex indicators based on the relevant business process and indicator components, thereby determining the fish head and fish bones in the fishbone diagram. The fish head is the complex indicator, and the fish bones are the indicator components. Convert the fishbone diagram into an indicator component impact path diagram. The indicator components are the nodes in the impact path diagram, and the complex indicators are the end points of the path.
[0084] Calculate the correlation between two adjacent nodes with a connected relationship, and define the path length between the two nodes by the correlation, divide the index component of the node directly connected to the path end point into a direct index component, and calculate the path length from the node to the path end point, divide the index component of the node whose path length is within a preset length range into a semi-direct index component, and determine the route from the node to the path end point;
[0085] The indicator components of other nodes are divided into indirect indicator components, and the route from the node to the end point of the path is determined.
[0086] In this embodiment, the path length between two nodes is defined by the correlation. The higher the correlation, the shorter the path length. According to the relevant business link process and indicator components, the fishbone diagram of the complex indicator is constructed to analyze the impact path of the indicator components of the complex indicator. The direct indicator component is a variable that directly affects the complex indicator, the semi-direct indicator component is a variable that affects the complex indicator by affecting other direct indicators, and has a greater impact on the complex indicator (shorter path), and the indirect indicator component is a variable that affects the complex indicator by affecting other direct indicators, and has a smaller impact on the complex indicator. The fishbone diagram or Ishikawa diagram is a tool for identifying potential causes of specific events or results. When determining the impact path, a cause-and-effect diagram can be drawn to show the cause-and-effect chain between the target indicator and each factor. By decomposing layer by layer, you can clearly see which are direct factors, which are indirect factors, and the paths of mutual influence between them.
[0087] It should be noted that the correlation between two nodes is calculated using the Pearson correlation coefficient. The Pearson correlation coefficient is an indicator that measures the degree of correlation between two variables. The Pearson correlation coefficient is calculated using the covariance and the standard deviation of the variables, and its range is [-1, 1]. When the correlation between two variables increases, the correlation coefficient tends to 1 or -1; when there is no relationship between the two variables, the correlation coefficient tends to 0.
[0088] Step S103, determining the interaction relationship between the indicator components according to the association relationship between the indicator components, and determining whether there is a coupling effect in the interaction relationship between the indicator components, and if there is a coupling effect, analyzing and quantifying the coupling effect.
[0089] In this embodiment, the interaction relationship and the coupling effect are related but not exactly the same. The interaction relationship more broadly describes the mutual influence between factors, while the coupling effect specifically refers to the part of this mutual influence that causes nonlinear or amplification / reduction effects on the target variable, which is more complicated. The interaction relationship between the indicator components is determined according to the association relationship between the indicator components. The direct indicator components, semi-direct indicator components and indirect indicator components each have their own associations (associations of multiple indicator components). Using data analysis tools such as Excel, SPSS, SAS, etc., the data of the indicator components can be statistically analyzed, such as correlation analysis, regression analysis, etc., to reveal the interaction relationship between them.
[0090] In some embodiments of the present application, indicator components are associated according to their types to determine the association relationship between indicator components, including:
[0091] The corresponding indicator components are associated through direct indicator components, the corresponding indicator components are associated through the path routes involved in semi-direct indicator components, and the corresponding indicator components are associated through the path routes involved in indirect indicator components.
[0092] In this embodiment, the direct indicator components correspond to direct indicators, that is, variables that directly affect complex indicators. The semi-direct indicator components correspond to direct indicators and indirect indicators, which have a greater impact on complex indicators. The indirect indicator components correspond to direct indicators and indirect indicators, which have a smaller impact on complex indicators.
[0093] It can be understood that the correlation between indicator components here refers to which variables are correlated, and this corresponding relationship is the correlation between indicator components.
[0094] In this embodiment, determining the interaction relationship between indicator components based on the association relationship between indicator components means clarifying the specific functional relationship between variables based on the corresponding relationship between variables, which can be achieved through regression analysis, using linear regression, polynomial regression, logistic regression or other advanced regression models to fit the relationship between indicator variables. These models can help understand how one indicator variable changes with changes in other indicator variables.
[0095] It is understandable that the determination of interaction relationships can also be achieved through path analysis. If the relationship between indicator variables is more complex and involves multiple intermediate variables, path analysis can be used to determine the direct and indirect effects between these variables. Path analysis allows researchers to test a preset causal model and estimate the strength and significance of each path.
[0096] In some embodiments of the present application, determining whether there is a coupling effect in the interaction relationship between indicator components includes:
[0097] Standardize several variables involved in the interaction relationship between indicator components, calculate the covariance matrix between several variables, solve the covariance matrix to obtain eigenvalues and eigenvectors, and thereby obtain multiple new variables. Weights are assigned to the new variables through eigenvalues, and weighted summation is performed on the new variables to obtain the initial comprehensive representative indicator.
[0098] The initial comprehensive representative index is adjusted by the stability of several variables involved in the interaction relationship between the indicator components to obtain a comprehensive representative index; ;in, As a comprehensive representative indicator, is the initial comprehensive representative indicator, is the number of indicator components that have interaction relationships, For the interaction relationship The stability of the indicator components, is the maximum stability value of the indicator components with interaction relationship, Volatility factor, is the volatility factor obtained by mapping the stability maximum of the indicator components that interact with each other. is a preset constant; the nonlinear dependence degree between the comprehensive representative indicators and the complex indicators is calculated, and the presence of a coupling effect is determined by the nonlinear dependence degree.
[0099] In this embodiment, in order to determine whether a coupling effect is generated or exists in the interaction relationship, several variables are integrated to obtain a comprehensive representative index, and principal component analysis (PCA) is used for integration.
[0100] Specifically, since the dimensions and distribution of each variable may be different, the data needs to be standardized first. Since the dimensions (i.e., measurement units) and distribution (i.e., the range and form of the data) of each variable may be different, direct analysis may lead to inaccurate or biased results. Therefore, the data needs to be standardized first so that each variable has the same dimensions and distribution.
[0101] More specifically, after standardization, the covariance matrix of the data is calculated to reflect the correlation between the variables. The covariance matrix is a matrix that describes the correlation between the variables in the data set, which reflects the strength and direction of the relationship between the variables.
[0102] More specifically, by solving the eigenvalues and eigenvectors of the covariance matrix, the main principal components (new variables) are extracted. The principal components are new, independent variables extracted from the original data, which can retain the information of the original data to the maximum extent. By solving the eigenvalues and eigenvectors of the covariance matrix, the main direction of change of the data can be found. These directions are the principal components, which are sorted according to the size of the eigenvalues. The larger the eigenvalue, the more information the corresponding principal component contains.
[0103] More specifically, the value of the comprehensive index is calculated based on the extracted principal components and the corresponding eigenvalues. Based on the extracted principal components and the corresponding eigenvalues, the value of one or more comprehensive indexes can be calculated. These indexes are usually obtained by weighting the principal components according to their eigenvalues, and the weight is the eigenvalue itself or some transformed form of the eigenvalue.
[0104] In this embodiment, Represents the stability class average of multiple variables, and adjusts the initial comprehensive representative index according to the stability class average of the variables. The stability class averages representing multiple variables are used to correct the initial comprehensive representative index.
[0105] In this embodiment, the presence of a coupling effect is determined by analyzing the degree of nonlinear dependence between the comprehensive representative index and the complex index. If two or more variables are interdependent when affecting the complex index, and when the interaction between the variables causes the impact on the complex index to present a nonlinear relationship, this is often a sign of a coupling effect.
[0106] In this embodiment, the nonlinear dependence degree is the nonlinear mutual dependence degree, which is a method based on state space reconstruction and nearest neighbor distance, and is used to determine the direction and magnitude of the causal relationship. For two independent systems or factors X and Y, which represent comprehensive representative indicators and complex indicators respectively, the state spaces of the two systems are established according to the state space reconstruction theory.
[0107] For the sample points xn, xrn,1, ..., xrn,k in the state space X, it represents the k nearest neighbor points of xn in the state space X, and the average Euclidean distance between xn and the k nearest neighbor points is calculated;
[0108] ; For the sample point yn in the state space Y, ysn,1,...,ysn,k represents the k neighboring points of yn in the state space Y, map it to the state space X, and calculate the average Euclidean distance between xn and the k neighboring points xsn,1,...,xsn,k;
[0109] ;
[0110] In order to simplify the calculation, the average distance between xn and all N sample points can be used; ; The degree of nonlinear interdependence is the state space method. The causal relationship between systems is determined based on the mapping relationship of the state space and is defined as: ; According to the definition, ,when When it approaches 0, systems X and Y are independent of each other. When it is significantly greater than 0, there is a causal relationship from system X to Y, and the closer it is to 1, the stronger the causality.
[0111] It should be noted that other indicators that can characterize causal relationships are also possible, and this application only provides a specific method.
[0112] In this embodiment, the coupling effect can be analyzed and quantified by a preset coupling mathematical model, a regression analysis model, a causal chain model, etc.
[0113] In some embodiments of the present application, the method further includes:
[0114] If there is no coupling effect, the indicator components and their interaction relationships are visualized based on the mapping relationship between the relevant business process and the indicator components.
[0115] Step S104, based on the mapping relationship between the relevant business link processes and the indicator components, the indicator components, the interaction relationships, and the coupling effects are visualized to achieve attribution display of the business links.
[0116] In this embodiment, indicators, interaction relationships, and coupling effects are visualized in the relevant business process to help decision makers analyze the reasons for the formation of complex indicators more clearly.
[0117] Compared with the prior art, the present invention has the following beneficial effects:
[0118] 1. By comparing the relevant business process and multiple indicator components, the indicator components are supplemented and a mapping relationship between the two is established to facilitate subsequent comparison.
[0119] 2. The indicator components are screened by redundancy to retain some indicator components and the redundant indicator components with similar meanings are removed, which ensures the independence of the indicator components and improves the reliability of the indicator component analysis.
[0120] 3. The interaction relationship between indicator components is determined based on the correlation relationship between indicator components, and whether there is a coupling effect in the interaction relationship between indicator components is determined. The correlation between indicator components is accurately given, and the interaction relationship between indicator components is obtained. It is also analyzed whether the interaction relationship will produce a coupling effect, which improves the adaptability and accuracy of attribution analysis of complex indicators and ensures the traceability of attribution analysis.
[0121] Through the description of the above implementation methods, those skilled in the art can clearly understand that the present invention can be implemented by hardware, or by software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present invention can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.), including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each implementation scenario of the present invention.
[0122] Correspondingly, the present application also provides an attribution analysis system applied to complex indicators, such as Figure 2 As shown, including,
[0123] The first module is used to obtain complex indicator models and business objectives in the financial or commercial fields, obtain relevant business process links through mapping the business objectives of complex indicators, decompose the complex indicator model according to the dimensional level, obtain multiple indicator components, and compare the relevant business process links with multiple indicator components to supplement the indicator components and build a mapping relationship between the relevant business process links and indicator components;
[0124] The second module is used to analyze the redundancy between indicator components, and select and retain some indicator components by screening the indicator components according to the redundancy, and then divide the types of indicator components according to the relationship between the complex indicator and each indicator component, and associate the indicator components according to the type of indicator components to determine the association relationship between the indicator components;
[0125] The third module is used to determine the interaction relationship between the indicator components according to the association relationship between the indicator components, and determine whether there is a coupling effect in the interaction relationship between the indicator components. If there is a coupling effect, the coupling effect is analyzed and quantified;
[0126] The fourth module is used to visualize the indicator components, interaction relationships, and coupling effects based on the mapping relationship between the relevant business link processes and the indicator components, thereby realizing the attribution display of the business links.
[0127] Those skilled in the art will appreciate that the accompanying drawings are merely schematic diagrams of a preferred implementation scenario, and the modules or processes in the accompanying drawings are not necessarily required for implementing the present invention.
[0128] Those skilled in the art will appreciate that the modules in the system in the implementation scenario can be distributed in the system of the implementation scenario according to the implementation scenario description, or can be changed accordingly and located in one or more systems different from the implementation scenario. The modules in the above implementation scenario can be combined into one module, or can be further split into multiple submodules.
[0129] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.
Claims
1. An attribution analysis method applied to complex indicators, characterized in that: include, Obtain complex indicator models and business objectives in the financial or commercial fields, obtain relevant business process flows through mapping the complex indicators to business objectives, decompose the complex indicator models according to the dimensional levels, obtain multiple indicator components, and compare the relevant business process flows with multiple indicator components to supplement the indicator components and build a mapping relationship between the relevant business process flows and indicator components; Analyze the redundancy between indicator components, and select and retain some indicator components by screening the indicator components according to the redundancy. Then, classify the types of indicator components according to the relationship between the complex indicator and each indicator component, associate the indicator components according to the type of indicator components, and determine the association relationship between the indicator components. Determine the interaction relationship between indicator components based on the association relationship between indicator components, and determine whether there is a coupling effect in the interaction relationship between indicator components. If there is a coupling effect, analyze and quantify the coupling effect; Based on the mapping relationship between the relevant business process and indicator components, the indicator components, interaction relationships, and coupling effects are visualized to achieve attribution display of business links; in, The complex indicator model is disassembled according to the dimension level to obtain multiple indicator components, including: Determine the function of the complex indicator according to the complex indicator model, decompose the function of the complex indicator into multiple variables, confirm the dimensional level to which each variable belongs, find several other variables close to the variable at the same dimensional level, and calculate the stability of each variable and several other variables close to it; Assign the judgment period of each variable by the attributes of the variable, integrate the coefficient of variation of the variable under each judgment period to determine the stability of each variable, and use the value of the variable and its stability as the indicator component; And determine whether there are coupling effects in the interaction relationships between indicator components, including, Standardize several variables involved in the interaction relationship between indicator components, calculate the covariance matrix between several variables, solve the covariance matrix to obtain eigenvalues and eigenvectors, and thereby obtain multiple new variables. Weights are assigned to the new variables through eigenvalues, and weighted summation is performed on the new variables to obtain the initial comprehensive representative indicator. The initial comprehensive representative index is adjusted by the stability of several variables involved in the interaction relationship between the indicator components to obtain a comprehensive representative index; in, is the comprehensive representative index, L is the initial comprehensive representative index, n is the number of indicator components with interactive relationships, and W i is the stability of the i-th indicator component with interaction relationship, max(W i ) is the maximum stability value of the indicator components with interaction relationship, τ is the volatility factor, max(W i )→τ is the volatility factor obtained by mapping the maximum stability of the indicator components with interactive relationships, and k is a preset constant; The nonlinear dependence degree between comprehensive representative indicators and complex indicators is calculated, and whether there is a coupling effect is determined by the nonlinear dependence degree.
2. The attribution analysis method applied to complex indicators according to claim 1, characterized in that: Analyze the redundancy between indicator components, including, According to the values of the indicator component variables, a variable curve graph of the indicator component variables changing over time is constructed, the text keywords of each indicator component variable are determined, and the similarity of the text keywords of the variables between the indicator components is compared, so as to determine the candidate indicator components with similar text relationships; Compare the variable curve graphs between the candidate indicator components with similar textual relationships to obtain the variable curve similarity, and calculate the redundancy between the candidate indicator components with similar textual relationships according to the variable curve similarity of the candidate indicator components with similar textual relationships, the similarity of text keywords and the stability of variables; P=α1Q1+α2Q2+α3β|Q3-Q4| Among them, P is the redundancy between the candidate indicator components with similar textual relations, α1 is the redundant weight of the variable curve similarity of the candidate indicator components with similar textual relations, Q1 is the variable curve similarity of the candidate indicator components with similar textual relations, α2 is the redundant weight of the similarity of the text keywords of the candidate indicator components with similar textual relations, Q2 is the similarity of the text keywords of the candidate indicator components with similar textual relations, α3 is the redundant weight of the stability of the variables of the candidate indicator components with similar textual relations, β is the conversion coefficient, and Q3 and Q4 are the stability of the variables of the candidate indicator components with similar textual relations.
3. The attribution analysis method applied to complex indicators according to claim 2 is characterized in that: And select and retain some indicator components by redundancy, including: If the redundancy exceeds the redundancy threshold, the candidate indicator component with the lower variable stability among the candidate indicator components with similar textual relations is eliminated, and the candidate indicator component with the higher variable stability is retained; Otherwise, all candidate indicator components in the candidate indicator components having similar textual relations are retained.
4. The attribution analysis method applied to complex indicators according to claim 1, characterized in that: Then, the types of indicator components are divided according to the relationship between complex indicators and each indicator component, including: The types of indicator components include direct indicator components, semi-direct indicator components, and indirect indicator components; Construct a fishbone diagram of complex indicators based on the relevant business process and indicator components, thereby determining the fish head and fish bones in the fishbone diagram. The fish head is the complex indicator, and the fish bones are the indicator components. Convert the fishbone diagram into an indicator component impact path diagram. The indicator components are the nodes in the impact path diagram, and the complex indicators are the end points of the path. Calculate the correlation between two adjacent nodes with a connected relationship, and define the path length between the two nodes by the correlation, divide the index component of the node directly connected to the path end point into a direct index component, and calculate the path length from the node to the path end point, divide the index component of the node whose path length is within a preset length range into a semi-direct index component, and determine the route from the node to the path end point; The indicator components of other nodes are divided into indirect indicator components, and the route from the node to the end point of the path is determined.
5. The attribution analysis method applied to complex indicators according to claim 4 is characterized in that: According to the type of indicator components, the indicator components are associated and the association relationship between the indicator components is determined, including: The corresponding indicator components are associated through direct indicator components, the corresponding indicator components are associated through the path routes involved in semi-direct indicator components, and the corresponding indicator components are associated through the path routes involved in indirect indicator components.
6. The attribution analysis method applied to complex indicators according to claim 1, characterized in that: The method further comprises, If there is no coupling effect, the indicator components and their interaction relationships are visualized based on the mapping relationship between the relevant business process and the indicator components.
7. An attribution analysis system applied to complex indicators, used to implement the attribution analysis method applied to complex indicators as described in any one of claims 1 to 6 above, characterized in that: include, The first module is used to obtain complex indicator models and business objectives in the financial or commercial fields, obtain relevant business process links through mapping the business objectives of complex indicators, decompose the complex indicator model according to the dimensional level, obtain multiple indicator components, and compare the relevant business process links with multiple indicator components to supplement the indicator components and build a mapping relationship between the relevant business process links and indicator components; The second module is used to analyze the redundancy between indicator components, and select and retain some indicator components by screening the indicator components according to the redundancy, and then divide the types of indicator components according to the relationship between the complex indicator and each indicator component, and associate the indicator components according to the type of indicator components to determine the association relationship between the indicator components; The third module is used to determine the interaction relationship between the indicator components according to the association relationship between the indicator components, and determine whether there is a coupling effect in the interaction relationship between the indicator components. If there is a coupling effect, the coupling effect is analyzed and quantified; The fourth module is used to visualize the indicator components, interaction relationships, and coupling effects based on the mapping relationship between the relevant business link processes and the indicator components, thereby realizing the attribution display of the business links.
