A financial risk communication strategy generation method performed by an electronic device

Abstract

The application discloses a financial risk communication strategy generation method executed by an electronic device, preprocessing financial feature data and external parameter sample data, constructing six types of financial evidence and multiple groups of prospect-sensitive parameters; sequentially calculating observation item level and portrait level diagnostic indicators, generating differentiated basic main communication strategies, secondary communication markers and traceable explanation information through quantile gating and a preset priority; meanwhile, three types of mutually isolated data storage areas are set, three-stage data isolation of strategy generation, expert calibration and post-deployment audit is realized, and future information is avoided in the strategy generation process. The application can convert risk assessment results into implementable communication strategies, and ensure that offline evaluation and actual deployment effects are consistent.

Description

Technical Field

[0001] This invention relates to a method for generating financial risk communication strategies executed by electronic devices. Background Technology

[0002] Financial education, robo-advisors, and risk management systems typically collect financial data such as a user's income, assets, liabilities, liquidity, financial knowledge, and risk appetite, and use this data to assess the user's risk tolerance. Existing systems often express assessment results in the form of risk scores, risk levels, or rankings. These results are primarily used to determine the level of risk a user can or is willing to take.

[0003] Existing technologies typically standardize financial data and employ weighted scoring, fuzzy decision-making, prospect theory models, or machine learning models to perform risk assessments. Some methods incorporate prospect theory parameters to describe the differences in how different agents respond to gains, losses, and uncertainty. However, these methods often use the calculation results for risk classification or ranking without further translating financial characteristics and prospect theory parameters into specific financial risk communication strategies.

[0004] In practical applications, users with similar risk scores may have different financial characteristics. For example, one user may lack the information needed to make a risk assessment, while another user may have higher financial vulnerability or greater potential for loss. The same risk description may not simultaneously meet the communication needs of both users. If the system relies solely on risk scores or risk levels, it will be difficult to determine whether supplementary information, adjustments to the interpretation, prompts regarding financial tolerance limits, or emphasis on financial resilience are needed.

[0005] Furthermore, some methods utilize historically observed risk indicators or expert annotations to train decision-making models. If the system uses observed risk indicators, audit results, or expert annotations on the test set that are only available after deployment during the strategy generation phase, the model may acquire information unavailable at the time of current decision-making, leading to offline evaluation results that do not accurately reflect the actual deployment effect. Therefore, existing technologies require a method and system that generates financial risk communication strategies based on financial characteristics and prospect theory parameters, and separates strategy generation, expert calibration, and post-deployment auditing. Summary of the Invention

[0006] The present invention provides a method for generating financial risk communication strategies executed by electronic devices in order to solve the problems existing in the prior art.

[0007] The technical solutions adopted in this invention are as follows:

[0008] A method for generating a financial risk communication strategy executed by an electronic device includes the following steps:

[0009] Acquire and preprocess the financial characteristic data of the object to be processed, as well as the external parameter sample data; the external parameter sample data includes prospect theory parameters, and also includes comprehensive parameters or values ​​of the generalized anxiety scale, state anxiety scale, and trait anxiety scale used to calculate the comprehensive parameters;

[0010] Based on the preprocessed financial feature data, financial evidence is constructed, including financial capacity evidence, uncertainty and burden of loss evidence, low-resolution indicators, change and shock indicators, financial vulnerability indicators, and tail risk proxy indicators; a benchmark financial evidence score is calculated based on the difference between the financial capacity evidence and the uncertainty and burden of loss evidence.

[0011] Based on external parameter sample data, risk tolerance multiplier, loss sensitivity multiplier, low-resolution sensitivity multiplier, change sensitivity multiplier, financial vulnerability sensitivity multiplier, and worst-case sensitivity parameter are constructed.

[0012] By combining various types of financial evidence with corresponding sensitive parameters, we can obtain the prospect value, support value, and information demand value of each financial observation item corresponding to each external parameter sample.

[0013] Based on the preset external parameter sample weights, the prospect value, support value, and information demand value are weighted and averaged to obtain the prospect diagnostic value, support diagnostic value, and information demand diagnostic value corresponding to each financial observation item.

[0014] Based on the comprehensive parameters, the external parameter samples are divided into high-value and low-value groups. For each financial observation item, the mean prospect value and mean support value are calculated for the high-value and low-value groups respectively, and the differences in prospect value and support value between groups are obtained. The standard deviations of prospect value and support value are calculated for all external parameter samples. Based on the differences in prospect value, the differences in support value, the standard deviations of prospect value and support value, the mechanism load diagnostic value corresponding to each financial observation item is calculated.

[0015] Financial evidence vectors are constructed based on financial capability evidence, uncertainty and loss burden evidence, low-resolution indicators, change and shock indicators, and financial vulnerability indicators. These vectors are then grouped according to a preset precision to obtain multiple financial feature profiles. The average of the corresponding indicators for all financial observations within the same financial feature profile is taken to obtain profile-level financial evidence, profile-level diagnostic values, and profile-level baseline financial evidence scores. Using profile-level support diagnostic values ​​and profile-level baseline financial evidence scores as independent variables, information demand residuals are calculated using a ridge regression model with cross-fitting. Communication burden diagnostic values ​​are calculated based on profile-level information demand diagnostic values, profile-level mechanism load diagnostic values, and profile-level financial stress values. The profile-level financial stress is determined by the maximum values ​​of profile-level low-resolution indicators, profile-level change and shock indicators, profile-level financial vulnerability indicators, and profile-level tail risk proxy indicators.

[0016] At the financial characteristic profiling level, the following scores are calculated: information supplementation score, anxiety sensitivity explanation score, protective financial resilience communication score, risk boundary explanation score, and standard risk tolerance communication score. Based on the preset quantile gating conditions and strategy priority, the basic main communication strategy corresponding to each financial characteristic profile is determined, the secondary communication flags corresponding to the basic main communication strategy are set to effective, and the corresponding strategy rationale and main driving indicators are generated.

[0017] The entire process of generating the basic master communication strategy does not use observed risk indicators, post-deployment audit results, or expert annotations on the test set.

[0018] Furthermore, the financial characteristic data includes features used to characterize financial capability, liquidity, financial experience, financial knowledge, financial stability, debt stress, financial vulnerability, low financial resilience, financial pressure, and information complexity.

[0019] Furthermore, the evidence of financial capability is obtained by weighting and summing the normalized values ​​of financial capability, liquidity, financial experience, financial knowledge, and financial stability and limiting it to the interval [0,1].

[0020] The evidence of uncertainty and loss burden is obtained by weighting and summing the normalized values ​​of debt stress, financial vulnerability, low financial resilience, financial stress and information complexity and limiting it to the interval [0,1].

[0021] The low-resolution index is obtained by weighting and summing the normalized value of information complexity, the normalized value of the dispersion of positive and negative financial features within the same observation item, and the proportion of missing original data, and restricting it to the interval [0,1].

[0022] The change and shock indicators are obtained by weighted summation of the normalized values ​​of financial stability, financial vulnerability, and financial stress, and restricted to the interval [0,1].

[0023] The financial vulnerability index is obtained by weighting and summing the normalized values ​​of financial vulnerability, low financial resilience, and debt stress, and limiting it to the interval [0,1].

[0024] The tail risk proxy indicator is obtained by weighting the maximum and mean values ​​of uncertainty and loss burden evidence, low-definition indicators, change and shock indicators, and financial vulnerability indicators, and restricting them to the interval [0,1].

[0025] Furthermore, the prospect theory parameters include risk preference parameters, loss sensitivity parameters, and selection accuracy parameters;

[0026] The risk tolerance multiplier is obtained by performing a negative exponential transformation on the standardized value of the risk preference parameter after a negative logarithmic transformation, followed by mean normalization.

[0027] The loss-sensitive multiplier is obtained by exponential transformation of the standardized logarithmic value of the loss-sensitive parameter and by performing mean normalization.

[0028] The low-resolution sensitivity multiplier, change sensitivity multiplier, and financial vulnerability sensitivity multiplier are all obtained by exponential transformation of the standardized values ​​of the comprehensive parameters and mean normalization.

[0029] The worst-case sensitivity parameter is obtained by transforming the standardized value of the comprehensive parameter, the standardized value of the risk preference parameter after negative logarithmic transformation, the standardized value of the loss sensitivity parameter after logarithmic transformation, and the standardized value of the selection precision parameter after logarithmic transformation, and then limiting it to a predetermined range through a logical function transformation.

[0030] When the external parameter sample data includes values ​​from the General Anxiety Scale, the State Anxiety Scale, and the Trait Anxiety Scale, the electronic device first performs standardization processing on the values ​​of the three scales respectively, then calculates the standardized mean, and finally performs standardization processing on the mean again to obtain the comprehensive parameter.

[0031] Furthermore, the benchmark financial evidence score is based on... calculate;

[0032] No. The external parameter sample corresponds to the _th _th The prospective value of each financial observation item is based on calculate;

[0033] Support values ​​according to calculate;

[0034] Information demand value according to Calculate; where, This refers to the index number of the external parameter sample. For the index number of the financial observation item; Based on the benchmark financial evidence score, For the first Evidence of financial capability for each financial observation item For the first Evidence regarding the uncertainty and loss burden of individual financial observations. For the first Low-resolution indicators for several financial observation items. For the first Changes and impact indicators of individual financial observation items For the first Financial vulnerability indicators for each financial observation item, For the first Tail risk proxy indicators for each financial observation item; For the first Risk tolerance multiplier corresponding to each external parameter sample For the first Loss-sensitive multipliers corresponding to each external parameter sample For the first The low-resolution sensitive multiplier corresponding to each external parameter sample. For the first The change-sensitive multiplier corresponding to each external parameter sample For the first The financial vulnerability sensitivity multiplier corresponding to each external parameter sample For the first The worst-case sensitive parameter corresponding to each external parameter sample; For the first The external parameter sample corresponds to the th . The prospective value of a financial observation item For the first The external parameter sample corresponds to the th . Support values ​​for each financial observation item For the first The external parameter sample corresponds to the th . Information requirement value for each financial observation item; , , , The weights of low-resolution indicators, change and shock indicators, financial vulnerability indicators, and tail risk proxy indicators in the calculation of support values ​​are as follows: Preset coefficients for the constant term in the formula for calculating information demand. , , , The preset coefficients for low-resolution indicators, change and impact indicators, financial vulnerability indicators, and tail risk proxy indicators in the calculation of information demand value are as follows: The electronic device performs a weighted average of the foreground value, support value, and information demand value based on the weights of the external parameter samples, to obtain the corresponding foreground diagnostic value, support diagnostic value, and information demand diagnostic value.

[0035] Furthermore, the mechanism load diagnostic value is obtained by normalizing and weighting the sum of the absolute value of the difference in prospective value between groups, the standard deviation of prospective value, the absolute value of the difference in support value between groups, and the standard deviation of support value; the communication burden diagnostic value is obtained by weighting and summing the normalized values ​​of the profile-level information demand diagnostic value, the normalized values ​​of the profile-level mechanism load diagnostic value, and the normalized values ​​of the profile-level financial stress value.

[0036] Furthermore, the information supplementation score is obtained by weighted summation of the normalized values ​​of the positive-zero portion of the profile-level information demand diagnostic value, low-resolution index, change and impact index, and information demand residual.

[0037] The anxiety sensitivity explanatory score is obtained by weighted summation of the absolute value of the difference in support values ​​between groups, the standard deviation of support values, the absolute value of the difference in prospect values ​​between groups, and the normalized value of the diagnostic value of mechanism load.

[0038] The protective financial resilience communication score is obtained by weighted summation of the complement of the normalized values ​​of the financial vulnerability index, the tail risk agency index, and the benchmark financial evidence score.

[0039] The risk boundary interpretation score is obtained by weighted summation of the baseline financial evidence score and the normalized value of the maximum value of the financial vulnerability index and the tail risk proxy index.

[0040] The standard risk tolerance communication score is obtained by weighted summation of the baseline financial evidence score, the complement of the normalized value of the information demand diagnosis value, the complement of the normalized value of the financial vulnerability index, and the complement of the normalized value of the tail risk proxy index.

[0041] The preset quantile gate condition is determined based on whether the differences between each communication score and its related financial evidence, diagnostic value or parameter group reach the corresponding preset quantile threshold.

[0042] The strategy priority order is as follows: information supplementation takes precedence over anxiety-sensitive interpretation, anxiety-sensitive interpretation takes precedence over risk boundary interpretation, risk boundary interpretation takes precedence over protective financial resilience communication, and protective financial resilience communication takes precedence over standard risk tolerance communication.

[0043] Furthermore, after determining the basic main communication strategy, the method also includes:

[0044] A primary communication strategy classification model and a secondary communication indicator classification model are trained using an expert calibration set. Expert annotations in the calibration set are obtained through blind review, where experts do not review observed risk indicators, post-deployment audit results, future results, or the generated basic primary communication strategy during the annotation process. Financial characteristics, diagnostic values, and communication scores generated during the generation of the basic primary communication strategy are input into the trained primary communication strategy classification model and secondary communication indicator classification model to obtain the calibrated primary communication strategy, calibrated secondary communication indicators, strategy confidence level, and manual review indicator. The calibrated primary communication strategy serves as the final financial risk communication strategy.

[0045] Specifically, the training of the primary communication strategy classification model and the secondary communication indicator classification model uses only the financial features, diagnostic values, and communication scores generated during the generation of the basic primary communication strategy, without using observed risk indicators, post-deployment audit results, future results, and test set expert annotations.

[0046] Furthermore, after generating the final financial risk communication strategy, the method further includes: acquiring observed risk indicators; evaluating the benchmark financial evidence score based on the observed risk indicators; calculating the mean and variance of the observed risk indicators within the financial feature profile; calculating the calibration prediction value and calibration residual using a cross-fit monotonic calibration method; and generating profile heterogeneity audit indicators and residual audit indicators based on whether the variance of the observed risk indicators and the calibration residual reach a preset quantile threshold, respectively.

[0047] The observed risk indicators, calibration residuals, and audit flags are used only to evaluate the existing financial risk communication strategies and do not participate in the generation of the current financial risk communication strategy.

[0048] Furthermore, before acquiring data, the electronic device sets up a first data area, a second data area, and a third data area in its memory that are mutually isolated from each other; the first data area stores financial feature data and external parameter sample data that serve as inputs for generating the current strategy, the second data area stores expert annotations, and the third data area stores observed risk indicators and post-deployment audit results; when generating the basic master communication strategy, the electronic device only reads the first data area and does not read the second and third data areas.

[0049] The present invention has the following beneficial effects:

[0050] (1) This invention converts financial characteristics and prospect theoretical parameters into financial risk communication strategies, enabling the system to determine the corresponding communication methods based on differences in financial characteristics, thus solving the problem that existing technologies cannot convert risk scores or ranking results into specific financial risk communication strategies.

[0051] (2) The present invention calculates information demand, parameter group differences, mechanism load and communication burden respectively, so that the system can distinguish financial feature profiles with similar risk scores but different communication needs, and solves the problem that users with the same risk score need different communication strategies due to differences in financial features.

[0052] (3) The present invention simultaneously outputs the main communication strategy, secondary communication indicators, strategy rationale and main driving indicators, so that the financial risk communication strategy can be traced back to the corresponding financial characteristics and calculation results, which solves the problem that the output results of the prior art lack interpretability and traceability.

[0053] (4) This invention separates strategy generation, expert calibration and post-deployment auditing, avoiding the use of observation risk indicators, audit results or test set expert annotations that can only be obtained after deployment. This solves the problem in the prior art where future information is mixed into the strategy generation process, resulting in inconsistencies between offline evaluation and actual deployment effect. Attached Figure Description

[0054] Figure 1 This is a flowchart illustrating the overall process of generating financial risk communication strategies according to the present invention.

[0055] Figure 2 This is a flowchart illustrating the process of constructing financial evidence and prospect theory parameters for this invention.

[0056] Figure 3 This is a schematic diagram of the process for calculating diagnostic values, information demand residuals, mechanism load diagnostic values, and communication burden diagnostic values ​​in this invention.

[0057] Figure 4 A flowchart illustrating the process of generating basic primary communication strategies, secondary communication markers, and supplementary explanatory information for this invention;

[0058] Figure 5 This diagram illustrates the data flow and isolation relationships for the generation of the basic strategy, expert calibration, and post-deployment verification and auditing of this invention.

[0059] Figure 6 This is a schematic diagram of the module structure of the financial risk communication strategy generation system of the present invention. Detailed Implementation

[0060] The invention will now be further described with reference to the accompanying drawings.

[0061] like Figures 1 to 5 This embodiment provides a method for generating financial risk communication strategies executed by an electronic device. The electronic device includes a processor and a memory. The memory stores program instructions that can be executed by the processor. When the processor executes the program instructions, it completes the following processing flow.

[0062] In this embodiment, the following symbols are used: m represents a financial observation item, and i represents an external parameter sample; This means restricting x to the interval [a, b]. express ; This indicates that a minimum-maximum normalization is performed on a finite number of values ​​of x within the current batch to be processed, when x does not have a finite number of values ​​or the difference between the maximum and minimum values ​​is no greater than 1. hour, Take a uniform value of 0.5; This indicates mean normalization, which involves replacing non-finite values ​​in the input data with 1, dividing by the sample mean, restricting the result to the range [0.25, 4.00], and then dividing by the sample mean of the restricted result again to keep the sample mean of each multiplier at 1. A positive number set to prevent undefined logarithms or abnormal normalization.

[0063] This embodiment divides the overall processing into three mutually data-isolated stages: strategy generation, expert calibration, and post-deployment audit. The strategy generation stage uses only the financial characteristics and external parameter data available at the current decision-making moment, without introducing expert annotations, observed risk indicators, and audit results from subsequent stages, to ensure consistency between offline evaluation and actual deployment effects. The electronic device has three mutually data-isolated data areas in its memory: a first data area, a second data area, and a third data area. The first data area stores the financial characteristic data and external parameter sample data used as input for the current strategy generation; the second data area stores expert annotations; and the third data area stores observed risk indicators and post-deployment audit results. During the strategy generation stage, the electronic device only reads the first data area and not the second or third data area.

[0064] The processor first retrieves the financial characteristic data and external parameter sample data of the object to be processed from the first data area. The financial characteristic data includes at least the features used to characterize financial capability, liquidity, financial experience, financial knowledge, financial stability, debt stress, financial vulnerability, low financial resilience, financial pressure, and information complexity. It may also include the proportion of missing data, observation item identifiers, observation weights, and data source identifiers.

[0065] The external parameter sample data includes prospect theory parameters, specifically risk preference parameters. Loss-sensitive parameters and select precision parameters The external parameter sample data also includes comprehensive parameters. Alternatively, it may include the General Anxiety Scale-7 (GAD-7) score, state anxiety score, and trait anxiety score used to calculate the comprehensive parameters.

[0066] The processor performs numerical transformation, anomaly coding identification, missing value marking, value direction unification, and interval normalization on financial feature data. For positive features, the larger the normalization value, the more sufficient the corresponding financial conditions are. For negative features, the larger the normalization value, the stronger the corresponding pressure is. For missing values ​​involved in numerical calculation, the median of the effective samples of the same feature is used to fill them. When the effective samples are insufficient to calculate the median, a preset neutral value is used to fill them. The filled values ​​are limited to the interval [0,1].

[0067] The processor determines observations to be processed solely based on the validity of the features required to construct financial evidence, without using the existence of risk indicators as a screening criterion. For extrinsic parameter sample data, if a sample does not provide a composite parameter and all three scale scores are invalid, the processor sets the composite parameter of that sample to 0 and marks it as lacking extrinsic parameter information; if at least one sample has this mark, the extrinsic parameter information deficiency flag for the current batch is set to valid. Before data processing, the processor removes or isolates information that can directly identify an individual, retaining only the feature fields required for generating the strategy.

[0068] Based on the preprocessed financial feature data, the processor constructs six types of financial evidence. For the m-th financial observation, the evidence is financial capacity. Calculate according to the following formula:

[0069] ,

[0070] in, , , , , These represent the normalized values ​​for financial capability, liquidity, financial experience, financial knowledge, and financial stability, respectively.

[0071] Evidence of uncertainty and burden of loss Calculate according to the following formula:

[0072] in, , , , , These represent the normalized values ​​for debt stress, financial vulnerability, low financial resilience, financial stress, and information complexity, respectively.

[0073] The processor calculates the dispersion of positive and negative feature values ​​within the same observation and normalizes this dispersion. The original missing proportion of the observations is denoted as Low-resolution indicators Calculate according to the following formula:

[0074] ,

[0075] Changes and shock indicators Calculate according to the following formula:

[0076] ,

[0077] Financial vulnerability indicators Calculate according to the following formula:

[0078] ,

[0079] Tail Risk Proxy Indicator Calculate according to the following formula:

[0080] ,

[0081] This indicator is used to highlight extreme levels of various types of adverse financial evidence while preserving the overall level of adverse evidence. The processor simultaneously calculates the baseline financial evidence score for the m-th financial observation based on the difference between evidence of financial capability and evidence of uncertainty and burden of loss. It is used to characterize the basic risk-bearing conditions that form financial characteristics.

[0082] The processor constructs various foreground-sensitive parameters based on the extrinsic parameter sample data. For the i-th extrinsic parameter sample, it first performs a basic parameter transformation and calculates the following:

[0083] ,

[0084] ,

[0085] ,

[0086] The processor performs standardization on RA, LA, and CP according to the mean and standard deviation of the reference sample, respectively, to obtain... , and .

[0087] For comprehensive parameters When external parameter samples directly provide comprehensive parameters, they are standardized to obtain... When the external parameter sample provides a generalized anxiety scale score... State anxiety score Trait anxiety score First, the effective values ​​of the three scales are standardized separately. Then, the mean of the effective standardized scores of the same sample is taken. Finally, the mean is standardized again to obtain the result. When all three scales have no valid scores, Set to 0. The processor calculates each type of multiplier according to the aforementioned mean normalization rules:

[0088] ,

[0089] ,

[0090] ,

[0091] ,

[0092] ,

[0093] in, As the risk tolerance multiplier, For loss-sensitive multipliers, For low-resolution sensitive multipliers, For change-sensitive multipliers, This is a financial vulnerability sensitivity multiplier. It is also a worst-case sensitivity parameter. Calculate according to the following formula:

[0094] ,

[0095] The calculation results are limited to the interval [0.02, 0.65].

[0096] The processor combines various types of financial evidence with their corresponding prospective sensitivity parameters to calculate fine-grained median and diagnostic values. The prospective value of the i-th extrinsic parameter sample corresponding to the m-th financial observation is calculated using the following formula:

[0097] ,

[0098] The corresponding support value is calculated according to the following formula:

[0099] ,

[0100] The support value, based on the prospective value, deducts the communication support burden caused by low clarity, changes and shocks, financial vulnerability, and tail risks. A lower value indicates a greater need for supplementary protective conditions or explanations of adverse evidence when communicating financial risks. The corresponding information requirement value is calculated using the following formula:

[0101] ,

[0102] The information requirement value is used to characterize the degree of need for supplementary information in a given scenario. The processor assigns weights to each external parameter sample. Furthermore, the sum of the weights of all samples is 1. In this embodiment, all external parameter samples are given equal weights. The prospect value, support value, and information demand value are weighted and averaged according to their weights to obtain the prospect diagnostic value, support diagnostic value, and information demand diagnostic value corresponding to the m-th financial observation item.

[0103] ,

[0104] ,

[0105] ,

[0106] The summation range is all valid external parameter samples.

[0107] The processor constructs a financial evidence vector based on financial capability evidence, uncertainty and loss burden evidence, low-resolution indicators, change and impact indicators, and financial vulnerability indicators for each financial observation item. The financial evidence vector is grouped according to a preset precision, with financial observation items with consistent values ​​grouped into the same financial feature profile. In this embodiment, each of the five indicators is rounded to three decimal places for grouping precision. The arithmetic mean of the corresponding indicators for all financial observation items within the same financial feature profile is taken to obtain the profile-level financial evidence, profile-level diagnostic value, and profile-level baseline financial evidence score. Based on this, the information demand residual is calculated. The processor uses the profile-level support diagnostic value and the profile-level baseline financial evidence score as independent variables, and the profile-level information demand diagnostic value as the dependent variable, and performs a cross-fit ridge regression. The profile sample is divided into five folds; during each fit, four folds are used to estimate the regression coefficients, and the remaining fold is used to generate a predicted value. The ridge coefficient is... The difference between the actual image-level information demand diagnostic value and the predicted value obtained from cross-fitting is determined as the information demand residual. When the number of valid images is less than five, the processor uses the mean of the information demand diagnostic values ​​in the valid images to generate a replacement predicted value, and then calculates the corresponding residual.

[0108] The processor calculates the load diagnostic value based on the comprehensive parameters. The external parameter samples are divided into high-value and low-value groups based on the median of the comprehensive parameters. Samples with comprehensive parameters not lower than the median are assigned to the high-value group, and the rest are assigned to the low-value group. This grouping only represents the relative position of the external parameter samples on the comprehensive parameters and is not used for classifying the psychological state of the objects to be processed.

[0109] When both groups contain valid samples, for the m-th financial observation, calculate the mean prospect value and mean support value for the high-value group and the low-value group respectively, and calculate the difference in prospect value between the groups by subtracting the mean of the high-value group from the mean of the low-value group. Differences in support values ​​between groups Simultaneously calculate the sample standard deviation of the foreground value under all external parameter samples. and the sample standard deviation of the support values .

[0110] When any group does not contain valid samples, the processor does not calculate inter-group differences, generates an external parameter indicating insufficient grouping, and disables the anxiety-sensitive interpretation gating. Mechanism load diagnostic value. Calculate according to the following formula:

[0111]

[0112] Mechanism load diagnostic values ​​are used to characterize the degree to which the same financial feature causes explanatory differences under different external parameter samples. The processor averages the mechanism load diagnostic values ​​within the same financial feature profile to obtain the profile-level mechanism load diagnostic value.

[0113] The processor further calculates the profiling-level communication burden diagnostic value. The maximum values ​​of the profiling-level low-resolution indicator, change and shock indicator, financial vulnerability indicator, and tail risk proxy indicator are normalized to obtain the profiling-level normalized financial stress value. Calculate the diagnostic value of communication burden at the profile level using the following formula:

[0114] ,

[0115] in, The normalized value of the diagnostic value for profile-level information needs. This is a normalized value for the profile-level mechanism load diagnostic value. The communication load diagnostic value is used to characterize the level of explanatory input required to complete effective risk communication.

[0116] At the financial profile level, the processor calculates communication scores for five categories. Information supplementation scores are calculated using the following formula:

[0117] ,

[0118] in, This represents the positive portion of the information demand residual. The anxiety sensitivity explanatory score is calculated using the following formula:

[0119] ,

[0120] The score for protective financial resilience communication is calculated using the following formula:

[0121] ,

[0122] The risk boundary interpretation score is calculated using the following formula:

[0123] ,

[0124] The standard risk tolerance communication score is calculated using the following formula:

[0125]

[0126] In all of the above formulas, All items are the results of the corresponding profile-level indicators after min-max normalization.

[0127] The processor calculates preset quantiles and performs gating judgment only when the current batch contains at least five valid financial feature profiles and the gating variables have valid dispersion. If the above conditions are not met, the processor determines standard risk tolerance communication as the primary communication strategy and generates a manual review flag indicating insufficient batch information. When the conditions are met, the processor determines the primary communication strategy according to a preset priority order, from highest to lowest priority: information supplementation, anxiety-sensitive interpretation, risk boundary interpretation, protective financial resilience communication, and standard risk tolerance communication. If the conditions of the first type of strategy are met, the latter type of strategy will not be used to cover the judgment result. Profiles that do not meet the first four types of conditions will default to the standard risk tolerance communication strategy.

[0128] The specific gating conditions for each strategy are as follows: when at least one of the following conditions is met, the information supplementation strategy is deemed to be valid: the information supplementation score is not lower than its 68th quartile, and the normalized value of the information demand diagnosis value is not lower than its 68th quartile, the normalized value of the positive value of the information demand residual is not lower than its 68th quartile, the normalized value of the low definition index is not lower than its 62nd quartile, or the normalized value of the change and impact index is not lower than its 70th quartile.

[0129] Without generating an external parameter grouping inadequacy indicator, the anxiety-sensitive explanatory strategy is considered valid when the anxiety-sensitive explanatory score is not lower than its 68th quartile, and at least one of the normalized absolute value of the difference between group support values, the normalized standard deviation of support values, or the normalized absolute value of the difference between group prospect values ​​is not lower than its 62nd quartile.

[0130] The risk boundary interpretation strategy is deemed valid when the risk boundary interpretation score is not lower than its 62nd percentile, the normalized value of the benchmark financial evidence score is not lower than its 55th percentile, and the financial resilience stress value is not lower than its 55th percentile. The financial resilience stress value is the normalized value of the maximum value of the financial vulnerability index and the tail risk proxy index.

[0131] A protective financial resilience communication strategy is considered successful when at least one of the following conditions is met: the score of the protective financial resilience communication strategy is not lower than its 55th percentile; the normalized value of the financial vulnerability index is not lower than its 55th percentile; the normalized value of the tail risk proxy index is not lower than its 55th percentile; or the normalized value of the benchmark financial evidence score is not higher than its 40th percentile.

[0132] The standard risk tolerance communication strategy is considered valid when the standard risk tolerance communication score is not lower than its 70th percentile, the normalized value of the benchmark financial evidence score is not lower than its 65th percentile, the financial resilience stress value is not higher than its 55th percentile, and none of the aforementioned four higher priority conditions are met.

[0133] The processor prioritizes the primary strategy by determining the maximum value among the five communication scores. It then generates secondary communication indicators using the 75th percentile of each of the five communication scores as thresholds, and sets the secondary communication indicators corresponding to the primary primary communication strategy as valid, ensuring consistency between the displayed results and the primary primary communication strategy. The processor selects at least one indicator with the largest value from among financial capability evidence, uncertainty and loss burden evidence, low-resolution indicators, change and shock indicators, financial vulnerability indicators, tail risk agency indicators, absolute value of inter-group prospect value difference, standard deviation of prospect value, absolute value of inter-group support value difference, standard deviation of support value, information demand diagnostic value, positive value of information demand residual, mechanism load diagnostic value, and communication burden diagnostic value as the primary driving indicator. Based on the primary primary communication strategy, it generates corresponding strategy rationales, and based on the secondary communication indicators, it generates corresponding precautions. The information supplementation strategy instructs the output end to first explain missing information, complex fields, data sources, and changes and impacts before conveying risk conclusions; the anxiety-sensitive explanation strategy instructs the output end to use step-by-step explanations, scenario comparisons, and neutral wording to reduce misunderstandings caused by differences in interpretation; the risk boundary explanation strategy instructs the output end to clarify liquidity constraints, loss limits, and adverse scenarios while explaining risk tolerance conditions; the protective financial resilience communication strategy instructs the output end to prioritize buffer funds, debt burden, financial vulnerability, and tail risks; and the standard risk tolerance communication strategy instructs the output end to provide routine explanations based on benchmark financial evidence scores and key evidence. The processor will jointly write the profile identifier, the number of observations within the profile, various types of financial evidence, various diagnostic values, five types of communication scores, basic main communication strategy, secondary communication indicators, key driving indicators, strategy rationale, and various anomaly indicators into the financial risk communication strategy record.

[0134] When expert calibration is enabled in the system, optional expert calibration processing is performed. The processor first generates a strictly blinded expert annotation template. The template includes profile identifiers, the number of observations within the profile, six types of financial evidence, the baseline financial evidence score, supporting diagnostic values, information demand diagnostic values, information demand residuals, mechanism load diagnostic values, and data on inter-group differences and standard deviations. It does not include observed risk indicators, audit results, future results, the system-generated basic primary communication strategy, the strategy rationale, or the communication score. Experts annotate the primary communication strategy, secondary communication indicators, rationale, and confidence level based on the template. Prohibited information cannot be viewed during the annotation process. The processor validates valid expert annotations, retaining only those containing profile identifiers and valid primary communication strategies, and rejecting annotations containing observed risk indicators or audit fields. The processor uses a profile-based matching strategy to record and annotate with experts, constructing a calibration feature vector containing twenty-two features: the number of observations within the profile, six types of financial evidence, baseline financial evidence score, support diagnostic value, information demand diagnostic value, information demand residual, mechanism load diagnostic value, absolute value of the difference in prospective value between groups, standard deviation of prospective value, absolute value of the difference in support value between groups, standard deviation of support value, five types of communication scores, and communication burden diagnostic value. The processor fills in the non-finite values ​​in the calibration features using the effective sample median of each feature column, calculates the mean and standard deviation using only the training set, and performs standardization on the entire dataset. The feature standard deviation is less than [value missing]. Set the time to 1.

[0135] The expert-annotated data is pre-divided into training, validation, and test sets. The training set is used to fit a five-class ensemble classifier to predict the primary communication strategy, while five binary classifiers are fitted to predict five secondary communication markers. This embodiment uses bagged decision tree ensemble, with 200 decision trees for the five-class classifier and 100 decision trees for each binary classifier. On the validation set, the decision thresholds for each binary classifier are selected from a threshold grid of 0.05 to 0.95 with a step size of 0.05, based on the principle of optimal F1 score. The test set is never involved in model fitting or parameter selection. The processor saves all model parameters and preprocessing parameters to ensure reproducibility. The calibration features at deployment are input into the trained calibration model to obtain the calibrated primary communication strategy, the probabilities of each strategy, and the probabilities of secondary communication markers. The maximum probability corresponding to the primary communication strategy is determined as the strategy confidence (Conf). The thresholds determined in the validation phase are used to generate the original secondary communication markers, and the secondary communication markers corresponding to the calibrated primary communication strategy are set to the display-oriented valid state. The processor calculates the calibrated strategy priority according to the following formula:

[0136] ,

[0137] When the strategy confidence level falls below a preset threshold, or when the calibrated primary communication strategy conflicts with the original secondary communication flags according to preset rules, a manual review flag is generated. The calibrated primary communication strategy serves as the final financial risk communication strategy output. Throughout the expert calibration phase, observed risk indicators and post-deployment audit results are not used; relevant expert-annotated data is stored only in the second data area.

[0138] After the final financial risk communication strategy is generated, post-deployment verification and auditing are performed. The processor obtains observed risk indicators from the third data area, unifies the value direction of the observed risk indicators, and normalizes them to the [0,1] interval. For indicators with larger values ​​indicating stronger risk tolerance, the normalization direction is maintained; for indicators with larger values ​​indicating higher adverse risks, the normalization value is subtracted from 1, ensuring that the processed observed risk indicators are aligned with the benchmark financial evidence score. The observed risk indicators are then matched with the generated strategy records based on the profile identifiers.

[0139] For a financial feature profile p, the set of financial observation items with effective observable risk indicators is denoted as p. The processed observed risk index of the j-th financial observation item is defined as follows: The corresponding non-negative observation weights are denoted as The observation weights satisfy:

[0140]

[0141] When no observation weights are pre-set, the observation weight of each valid financial observation is set to 1. The processor calculates the normalized observation weights according to the following formula:

[0142]

[0143] The processor calculates the weighted mean of the observed risk indicators of the financial feature profile $p$ according to the normalized observation weights and the following formula:

[0144]

[0145] The processor calculates the weighted variance of the observed risk indicators within the financial feature profile $p$ according to the following formula:

[0146]

[0147] in, The observed risk center level of the financial feature profile p is used as the target value for subsequent non-decreasing monotonic calibration of cross-fitting. This indicates the degree of dispersion of observed risk indicators within the same financial characteristic profile, serving as the basis for subsequent heterogeneity audit markers in profile generation. Financial observations lacking observed risk indicators do not belong to the set. It does not participate in the calculation of the weighted mean and weighted variance, nor does it affect the previously generated financial risk communication strategy records.

[0148] In obtaining profiles of various financial characteristics and Subsequently, when the number of valid financial feature profiles is no less than five, the processor takes the profile-level baseline financial evidence score and the profile-level supporting diagnostic value as inputs, and outputs the corresponding... For the target value, perform cross-fit non-decreasing monotonic calibration.

[0149] Missing items for observed risk indicators are not included in the audit statistics and do not affect previously generated strategy records. When there are no fewer than five valid profiles, the baseline financial evidence score and supporting diagnostic value at the profile level are used as inputs, and the weighted mean of the observed risk indicators is used as the target, and a cross-fit non-decreasing monotonic calibration is performed.

[0150] After sorting the portraits by input scores, they are assigned to five folds in turn. Each fold uses samples from the other folds to fit a non-decreasing monotonic mapping using a weighted adjacent violator merging algorithm, generating a calibration prediction value in the current fold. Training portraits with the same input score are first merged by observation weights and then used for fitting. When there are fewer than five training portraits in a fold, the weighted mean of the observed risk indicators within that fold is used to generate the calibration prediction value. When there are fewer than five valid portraits, the weighted mean of the observed risk indicators of all valid portraits is used to generate the calibration prediction value, and an audit sample insufficiency flag is generated. The calibration prediction value is restricted to the interval [0,1], and the difference between the weighted mean of the observed risk indicators and the calibration prediction value is calculated to obtain the calibration residual. The 75th quantile of the variance of the observed risk indicators is set as the variance threshold qVar, and the 75th quantile of the absolute value of the calibration residual is set as the residual threshold qRes. When the number of observations in the portrait is not less than 2, and truth_var and qVar are both finite values, ,and At that time, generate audit markers for heterogeneity in the profile; when Both qRes and qRes are finite values, and qRes > ,and When qVar or qRes is not greater than 1, a residual audit flag is generated. At that time, no audit flag is generated based on the corresponding indicator. The processor generates corresponding audit descriptions based on the two types of audit flags and outputs independent audit records. All observed risk indicators, calibration residuals, and audit flags are only used to evaluate the generated financial risk communication strategies and do not participate in the generation of the current batch of financial risk communication strategies. If the audit results are to be used in subsequent version iterations, the complete dataset must be rebuilt and the model validation must be re-executed.

[0151] like Figure 6This embodiment also provides a financial risk communication strategy generation system for implementing the above method. The system includes a data acquisition and processing module, a financial evidence construction module, a prospect theory parameter construction module, a prospect support calculation module, a mechanism diagnosis module, a communication strategy generation module, an expert calibration module, and a post-deployment audit module.

[0152] The data acquisition and processing module is used to acquire target financial data and external parameter data, and perform numerical transformation, anomaly coding identification, missing value handling, value direction unification and normalization processing; the data acquisition and processing module is also used to store the financial feature data and external parameter sample data available in the strategy generation stage in the first data area, store the expert annotations in the second data area, and store the observed risk indicators and post-deployment audit results in the third data area; the first data area, the second data area and the third data area are isolated from each other.

[0153] The financial evidence construction module is used to calculate financial capability evidence, uncertainty and loss burden evidence, low-resolution indicators, change and shock indicators, financial vulnerability indicators, and tail risk proxy indicators based on the target financial characteristics.

[0154] The prospect theory parameter construction module is used to calculate the risk tolerance multiplier, loss sensitivity multiplier, low-resolution sensitivity multiplier, change sensitivity multiplier, financial vulnerability sensitivity multiplier, and worst-case sensitivity parameter based on external parameter samples.

[0155] The prospect support calculation module is used to calculate the baseline financial evidence score, prospect value, support value, information demand value, support diagnostic value, information demand diagnostic value, and information demand residual.

[0156] The mechanism diagnosis module is used to calculate the differences between parameter groups, the degree of sample dispersion, the diagnostic value of mechanism load, and the diagnostic value of communication burden, and to generate an external parameter grouping insufficiency flag when parameter grouping is incomplete.

[0157] The communication strategy generation module is used to calculate five types of communication scores and generate basic primary communication strategies, secondary communication indicators, main driving indicators and strategy rationales according to preset quantile gating conditions and strategy priority order. When there are insufficient batch samples, a batch information insufficient manual review indicator is generated.

[0158] The expert calibration module is an optional module used to train the calibration model based on blinded expert annotations when expert calibration is enabled, and output the calibrated main communication strategy, strategy confidence and manual review flag.

[0159] The post-deployment audit module is used to calculate profile heterogeneity and calibration residuals based on observed risk indicators after the final financial risk communication strategy is generated, and output independent audit records containing audit flags.

[0160] Each of the above modules can be implemented by a single processor executing program instructions in memory, or by multiple processors working together. Different modules can be deployed on the same computing device or on different computing devices through a controlled interface. Regardless of the deployment method, the communication strategy generation module must not call the observation risk indicators, calibration residuals, or audit flags in the post-deployment audit module.

[0161] In a batch execution instance, the processor first reads an external parameter table containing risk preference parameters, loss sensitivity parameters, selection precision parameters, and scores from three anxiety scales. Following the aforementioned parameter construction rules, it generates an external parameter sample table. The processor checks whether the external parameter samples can form high-value and low-value groups for the comprehensive parameters, and generates an external parameter grouping insufficiency flag if neither group can be formed. The processor then reads a target financial data table containing various financial features and missing proportions, generating a financial evidence table according to the aforementioned preprocessing and evidence construction rules. The processor combines the external parameter sample table with each financial observation item, calculating prospect value, support value, and information demand value. It then summarizes these values ​​according to financial feature profiles to obtain profile-level indicators and diagnostic values, thereby generating a basic financial risk communication strategy record. Even if the target financial table separately stores the observed risk indicator field, the processor does not pass this field to the strategy generation function. When the system obtains rigorously blinded expert annotations, the processor trains and applies an expert calibration model; when the system does not obtain qualified expert annotations, the basic financial risk communication strategy is retained as the final result. Once the observed risk indicators meet the audit conditions, the processor generates an independent audit record. Ultimately, the system uses profile identifiers to associate basic policy records, expert calibration policy records, and audit records, but does not overwrite the original policy records with audit records, ensuring that the three-stage data isolation rules always take effect in batch processing scenarios.

[0162] The above description is only a preferred embodiment of the present invention. It should be noted that those skilled in the art can make several improvements without departing from the principle of the present invention, and these improvements should also be considered within the scope of protection of the present invention.

Claims

1. A method for generating financial risk communication strategies executed by electronic devices, characterized in that: Includes the following steps: Acquire and preprocess the financial characteristic data of the object to be processed, as well as the external parameter sample data; the external parameter sample data includes prospect theory parameters, and also includes comprehensive parameters or values ​​of the generalized anxiety scale, state anxiety scale, and trait anxiety scale used to calculate the comprehensive parameters; Based on the preprocessed financial feature data, financial evidence is constructed, including financial capacity evidence, uncertainty and burden of loss evidence, low-resolution indicators, change and shock indicators, financial vulnerability indicators, and tail risk proxy indicators; a benchmark financial evidence score is calculated based on the difference between the financial capacity evidence and the uncertainty and burden of loss evidence. Based on external parameter sample data, risk tolerance multiplier, loss sensitivity multiplier, low-resolution sensitivity multiplier, change sensitivity multiplier, financial vulnerability sensitivity multiplier, and worst-case sensitivity parameter are constructed. By combining various types of financial evidence with corresponding sensitive parameters, we can obtain the prospect value, support value, and information demand value of each financial observation item corresponding to each external parameter sample. Based on the preset external parameter sample weights, the prospect value, support value, and information demand value are weighted and averaged to obtain the prospect diagnostic value, support diagnostic value, and information demand diagnostic value corresponding to each financial observation item. Based on the comprehensive parameters, the external parameter samples are divided into high-value and low-value groups. For each financial observation item, the mean prospect value and mean support value are calculated for the high-value and low-value groups respectively, and the differences in prospect value and support value between groups are obtained. The standard deviations of prospect value and support value are calculated for all external parameter samples. Based on the differences in prospect value, the differences in support value, the standard deviations of prospect value and support value, the mechanism load diagnostic value corresponding to each financial observation item is calculated. Financial evidence vectors are constructed based on evidence of financial capability, evidence of uncertainty and loss burden, low-definition indicators, change and impact indicators, and financial vulnerability indicators. The financial evidence vectors are then grouped according to a preset precision to obtain multiple financial feature profiles. The average of the corresponding indicators for all financial observation items within the same financial feature profile is used to obtain profile-level financial evidence, profile-level diagnostic value, and profile-level benchmark financial evidence score. Using the profile-level support diagnostic value and the profile-level benchmark financial evidence score as independent variables, the information demand residual is calculated through a ridge regression model with cross-fitting; the communication burden diagnostic value is calculated based on the profile-level information demand diagnostic value, the profile-level mechanism load diagnostic value, and the profile-level financial stress value; the profile-level financial stress is determined by the maximum value of the profile-level low-resolution indicator, the profile-level change and impact indicator, the profile-level financial vulnerability indicator, and the profile-level tail risk proxy indicator. At the level of financial characteristic profiling, the scores are calculated as follows: information supplementation score, anxiety sensitivity explanation score, protective financial resilience communication score, risk boundary explanation score, and standard risk tolerance communication score. Based on the preset quantile gating conditions and strategy priority, the basic main communication strategy corresponding to each financial feature profile is determined, the secondary communication flags corresponding to the basic main communication strategy are set to valid, and the corresponding strategy rationale and main driving indicators are generated. The entire process of generating the basic master communication strategy does not use observed risk indicators, post-deployment audit results, or expert annotations on the test set.

2. The method for generating financial risk communication strategies executed by electronic devices as described in claim 1, characterized in that: The financial characteristic data includes features used to characterize financial capability, liquidity, financial experience, financial knowledge, financial stability, debt stress, financial vulnerability, low financial resilience, financial pressure, and information complexity.

3. The method for generating financial risk communication strategies executed by electronic devices as described in claim 1, characterized in that: The evidence of financial capability is obtained by weighting and summing the normalized values ​​of financial capability, liquidity, financial experience, financial knowledge and financial stability and restricting it to the interval [0,1]. The evidence of uncertainty and loss burden is obtained by weighting and summing the normalized values ​​of debt stress, financial vulnerability, low financial resilience, financial stress and information complexity and limiting it to the interval [0,1]. The low-resolution index is obtained by weighting and summing the normalized value of information complexity, the normalized value of the dispersion of positive and negative financial features within the same observation item, and the proportion of missing original data, and restricting it to the interval [0,1]. The change and shock indicators are obtained by weighted summation of the normalized values ​​of financial stability, financial vulnerability, and financial stress, and restricted to the interval [0,1]. The financial vulnerability index is obtained by weighting and summing the normalized values ​​of financial vulnerability, low financial resilience, and debt stress, and limiting it to the interval [0,1]. The tail risk proxy indicator is obtained by weighting the maximum and mean values ​​of uncertainty and loss burden evidence, low-definition indicators, change and shock indicators, and financial vulnerability indicators, and restricting them to the interval [0,1].

4. The method for generating financial risk communication strategies executed by electronic devices as described in claim 1, characterized in that: The prospect theory parameters include risk preference parameters, loss sensitivity parameters, and selection accuracy parameters; The risk tolerance multiplier is obtained by performing a negative exponential transformation on the standardized value of the risk preference parameter after a negative logarithmic transformation, followed by mean normalization. The loss-sensitive multiplier is obtained by exponential transformation of the standardized logarithmic value of the loss-sensitive parameter and by performing mean normalization. The low-resolution sensitivity multiplier, change sensitivity multiplier, and financial vulnerability sensitivity multiplier are all obtained by exponential transformation of the standardized values ​​of the comprehensive parameters and mean normalization. The worst-case sensitivity parameter is obtained by transforming the standardized value of the comprehensive parameter, the standardized value of the risk preference parameter after negative logarithmic transformation, the standardized value of the loss sensitivity parameter after logarithmic transformation, and the standardized value of the selection precision parameter after logarithmic transformation, and then limiting it to a predetermined range through a logical function transformation. When the external parameter sample data includes values ​​from the General Anxiety Scale, the State Anxiety Scale, and the Trait Anxiety Scale, the electronic device first performs standardization processing on the values ​​of the three scales respectively, then calculates the standardized mean, and finally performs standardization processing on the mean again to obtain the comprehensive parameter.

5. The method for generating financial risk communication strategies executed by an electronic device as described in claim 1, characterized in that: The benchmark financial evidence score is based on calculate; No. The _th external parameter sample corresponds to the _th The prospective value of each financial observation item is based on calculate; Support values ​​according to calculate; Information demand value according to Calculate; where, This is the index number of the external parameter sample. For the index number of the financial observation item; Based on the benchmark financial evidence score, For the first Evidence of financial capability for each financial observation item For the first Evidence regarding the uncertainty and loss burden of individual financial observations. For the first Low-resolution indicators for several financial observation items. For the first Changes and impact indicators of individual financial observation items For the first Financial vulnerability indicators for each financial observation item, For the first Tail risk proxy indicators for each financial observation item; For the first Risk tolerance multiplier corresponding to each external parameter sample For the first Loss-sensitive multipliers corresponding to each external parameter sample For the first The low-resolution sensitive multiplier corresponding to each external parameter sample. For the first The change-sensitive multiplier corresponding to each external parameter sample For the first The financial vulnerability sensitivity multiplier corresponding to each external parameter sample For the first The worst-case sensitive parameter corresponding to each external parameter sample; For the first The external parameter sample corresponds to the th . The prospective value of a financial observation item For the first The external parameter sample corresponds to the th . Support values ​​for each financial observation item For the first The external parameter sample corresponds to the th . Information requirement value for each financial observation item; , , , The weights of low-resolution indicators, change and shock indicators, financial vulnerability indicators, and tail risk proxy indicators in the calculation of support values ​​are as follows: Preset coefficients for the constant term in the formula for calculating information demand. , , , The preset coefficients for low-resolution indicators, change and impact indicators, financial vulnerability indicators, and tail risk proxy indicators in the calculation of information demand value are as follows: The electronic device performs a weighted average of the foreground value, support value, and information demand value based on the weights of the external parameter samples, to obtain the corresponding foreground diagnostic value, support diagnostic value, and information demand diagnostic value.

6. The method for generating financial risk communication strategies executed by an electronic device as described in claim 1, characterized in that: The mechanism load diagnostic value is obtained by normalizing and weighting the sum of the absolute value of the difference in prospective value between groups, the standard deviation of prospective value, the absolute value of the difference in support value between groups, and the standard deviation of support value; the communication burden diagnostic value is obtained by weighting and summing the normalized values ​​of the profile-level information demand diagnostic value, the normalized values ​​of the profile-level mechanism load diagnostic value, and the normalized values ​​of the profile-level financial stress value.

7. The method for generating financial risk communication strategies executed by an electronic device as described in claim 1, characterized in that: The information supplement score is obtained by weighted summation of the normalized values ​​of the positive-zero portion of the profile-level information demand diagnostic value, low-resolution index, change and impact index, and information demand residual. The anxiety sensitivity explanatory score is obtained by weighted summation of the absolute value of the difference in support values ​​between groups, the standard deviation of support values, the absolute value of the difference in prospect values ​​between groups, and the normalized value of the diagnostic value of mechanism load. The protective financial resilience communication score is obtained by weighted summation of the complement of the normalized values ​​of the financial vulnerability index, the tail risk agency index, and the benchmark financial evidence score. The risk boundary interpretation score is obtained by weighted summation of the baseline financial evidence score and the normalized value of the maximum value of the financial vulnerability index and the tail risk proxy index. The standard risk tolerance communication score is obtained by weighted summation of the baseline financial evidence score, the complement of the normalized value of the information demand diagnosis value, the complement of the normalized value of the financial vulnerability index, and the complement of the normalized value of the tail risk proxy index. The preset quantile gate condition is determined based on whether the differences between each communication score and its related financial evidence, diagnostic value or parameter group reach the corresponding preset quantile threshold. The strategy priority order is as follows: information supplementation takes precedence over anxiety-sensitive interpretation, anxiety-sensitive interpretation takes precedence over risk boundary interpretation, risk boundary interpretation takes precedence over protective financial resilience communication, and protective financial resilience communication takes precedence over standard risk tolerance communication.

8. The method for generating financial risk communication strategies executed by an electronic device as described in claim 1, characterized in that: After determining the basic master communication strategy, the method further includes: A primary communication strategy classification model and a secondary communication indicator classification model are trained using an expert calibration set. Expert annotations in the calibration set are obtained through blind review, where experts do not review observed risk indicators, post-deployment audit results, future results, or the generated basic primary communication strategy during the annotation process. Financial characteristics, diagnostic values, and communication scores generated during the generation of the basic primary communication strategy are input into the trained primary communication strategy classification model and secondary communication indicator classification model to obtain the calibrated primary communication strategy, calibrated secondary communication indicators, strategy confidence level, and manual review indicator. The calibrated primary communication strategy serves as the final financial risk communication strategy. Specifically, the training of the primary communication strategy classification model and the secondary communication indicator classification model uses only the financial features, diagnostic values, and communication scores generated during the generation of the basic primary communication strategy, without using observed risk indicators, post-deployment audit results, future results, and test set expert annotations.

9. The method for generating financial risk communication strategies executed by an electronic device as described in claim 1, characterized in that: After generating the final financial risk communication strategy, the method further includes: obtaining observed risk indicators; evaluating the benchmark financial evidence score based on the observed risk indicators; calculating the mean and variance of the observed risk indicators within the financial feature profile; calculating the calibration prediction value and calibration residual using a cross-fit monotonic calibration method; and generating profile heterogeneity audit indicators and residual audit indicators based on whether the variance of the observed risk indicators and the calibration residual reach a preset quantile threshold. The observed risk indicators, calibration residuals, and audit flags are used only to evaluate the existing financial risk communication strategies and do not participate in the generation of the current financial risk communication strategy.

10. The method for generating financial risk communication strategies executed by an electronic device as described in claim 1, characterized in that: Before acquiring data, the electronic device sets up a first data area, a second data area, and a third data area in its memory that are mutually isolated from each other; the first data area stores financial feature data and external parameter sample data that serve as inputs for the current strategy generation; the second data area stores expert annotations; and the third data area stores observed risk indicators and post-deployment audit results. When generating the basic master communication strategy, the electronic device only reads the first data area and does not read the second and third data areas.

Images