Intelligent business decision analysis and risk early warning system based on large model driving

Abstract

The present application belongs to the technical field of business decision analysis, and specifically relates to an intelligent business decision analysis and risk early warning system based on a large model drive. The present application sets initialization modules, associated business identification modules, cause and effect identification modules, comprehensive evaluation modules and risk identification modules in sequence, first extracts accurate and regular business feature vectors from multi-source marketing data, then completes clustering identification and synergy degree evaluation of associated business combinations, further distinguishes the cause and effect direction between associated businesses and quantifies the cause and effect strength, finally generates business combination health degree scores in combination with business synergy degree indexes and cause and effect influence coefficients, realizes risk grading early warning, breaks through the limitations of traditional methods relying on artificial experience and linear assumptions, can mine hidden associations between complex businesses, effectively distinguishes causality and correlation in the association relationship, avoids misjudging false correlation as business influence, and greatly improves the accuracy of business decision analysis results.

Description

Technical Field

[0001] This invention belongs to the field of business decision analysis technology, specifically relating to an intelligent business decision analysis and risk warning system based on a large model. Background Technology

[0002] With the advancement of digital transformation, current enterprise business systems are constantly expanding, and the relationships between businesses are becoming increasingly complex and hidden. Traditional business decision analysis often relies on human experience to set analytical dimensions, which can only perform superficial analysis on single businesses or simple business combinations. It cannot accurately uncover the hidden relationships and causal logic between businesses, and the resulting analysis results are not accurate enough to provide reliable support for business decisions. Therefore, in order to ensure the healthy operation of enterprise business combinations, there is an urgent need for intelligent analysis and early warning solutions that can adapt to complex business systems.

[0003] While some business risk warning methods based on statistical models exist in the existing technology, these methods mostly rely on pre-set linear correlation assumptions, making it difficult to capture the non-linear hidden correlations between businesses, and also unable to distinguish between causality and correlation in the correlation, which easily leads to false warnings. At the same time, with the rapid growth of business data scale, the processing efficiency and analytical depth of traditional analysis methods can no longer meet the current decision-making needs of enterprises. Based on this, this solution provides an intelligent business decision analysis and risk warning system driven by a large model to solve the above problems. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent business decision analysis and risk warning system based on a large model, which can uncover the relationships and causal logic between businesses, distinguish between correlation and causality in business relationships, and effectively improve the accuracy of business analysis results.

[0005] The specific technical solution adopted by this invention is as follows: A large-model-driven intelligent business decision analysis and risk early warning system includes: The initialization module is used to acquire marketing data for each business in the historical period, and pre-build a multi-dimensional feature extraction framework based on the marketing data to extract business feature vectors under the multi-dimensional feature extraction framework. The related business identification module is used to perform correlation analysis on the feature vectors of various businesses, identify and output related business combinations, and evaluate the synergy effect of each related business combination to generate a business synergy index. The causal identification module is used to identify the causal relationships among the various businesses in the related business combination, output the dominant business and subordinate business, quantify the causal strength of the dominant business and subordinate business, and generate causal influence coefficients. The comprehensive evaluation module is used to comprehensively evaluate the business performance of related business portfolios based on business synergy indicators and causal impact coefficients, and generate a business portfolio health score. The risk identification module is used to classify the risk level of related business portfolios based on the health score of the business portfolio, and to match the corresponding risk warning signal according to the risk level classification result.

[0006] In a preferred embodiment, the initialization module includes: The data acquisition unit is used to collect marketing data from multiple data sources for each business during historical marketing cycles. The feature construction unit is used to extract multi-dimensional features from marketing data based on a preset multi-dimensional feature extraction framework, and generate an initial feature set including statistical features, time-series features and cross features. The feature filtering unit is used to sort the importance of the initial feature set and remove redundancies, filter out the key features that meet the business decision-making objectives, and combine them into a business feature vector.

[0007] In a preferred embodiment, the associated service identification module includes: The correlation analysis unit is used to calculate the correlation between various business feature vectors and output the correlation strength values ​​between different business feature vectors. Clustering units are used to cluster services based on association strength values, identify and output related service combinations; The collaborative evaluation unit is used to evaluate the synergistic effect of the joint performance of businesses within each related business portfolio over a historical period, calculate the complementarity and substitutability between businesses, and generate business synergy indicators.

[0008] In a preferred embodiment, the causal identification module includes a causal identification unit, which is used to determine the causal direction between businesses in each related business combination based on the temporal change pattern and interaction path of the business feature vector in the historical period, and output the leading business with the guiding role and the subordinate businesses affected by it. The causal quantification unit is used to numerically measure the causal effect by comparing the response magnitude of subordinate businesses before and after the change in the dominant business, and combining the degree of dependence between subordinate businesses and the dominant business, and to generate a causal influence coefficient.

[0009] In a preferred embodiment, the causal quantification unit includes a heterogeneity correction subunit, which, after generating the causal influence coefficient, performs heterogeneity analysis on the response differences of subordinate businesses at different levels of the dominant business based on the value changes of the dominant business in historical states, and corrects the causal influence coefficient based on the analysis results, outputting the heterogeneously corrected causal influence coefficient.

[0010] In a preferred embodiment, the comprehensive evaluation module includes: The weighting allocation unit is used to allocate weighting factors to business synergy indicators and causal influence coefficients based on the historical marketing objectives and results of related business combinations. The comprehensive scoring unit is used to perform weighted calculations on each business within the related business portfolio based on the weight allocation results of business synergy indicators and causal influence coefficients, and generate a business portfolio health score.

[0011] In a preferred embodiment, the weight allocation unit includes a time-series adjustment subunit, which is used to calculate and output time-series adjustment factors of the business synergy index and the causal influence coefficient through a sliding window mechanism, and to adjust the weight factors of the business synergy index and the causal influence coefficient in real time based on the time-series adjustment factors.

[0012] In a preferred embodiment, the comprehensive scoring unit further includes a trend analysis subunit, which is used to perform trend correction on the business portfolio health score based on the changing trend of the business portfolio health score within a continuous time window, thereby completing the calibration of the business portfolio health score.

[0013] In a preferred embodiment, the risk identification module includes: The risk level classification unit is used to classify related business portfolios into different risk levels based on the numerical range of the business portfolio health score. The early warning matching unit is used to match and output the corresponding risk early warning signal from the preset set of early warning signals based on the risk level classification results.

[0014] The present invention also provides an electronic device, the electronic device comprising: At least one processor; and a memory communicatively connected to the at least one processor; The memory stores a computer program that can be executed by the at least one processor, which enables the at least one processor to execute the aforementioned intelligent business decision analysis and risk warning system based on a large model.

[0015] The technical effects achieved by this invention are as follows: This invention, by sequentially setting up an initialization module, a related business identification module, a causal identification module, a comprehensive evaluation module, and a risk identification module, first extracts precise and well-organized business feature vectors from multi-source marketing data. Then, it completes cluster identification and synergy assessment of related business combinations, further distinguishes the causal directions between related businesses and quantifies the causal strength. Finally, it combines business synergy indicators and causal influence coefficients to generate a business portfolio health score, achieving risk-level early warning. This invention overcomes the limitations of traditional methods that rely on human experience and linear assumptions, enabling the discovery of hidden connections between complex businesses. Simultaneously, it effectively distinguishes between causality and correlation in relationships, avoiding misjudging false correlations as business impacts, and significantly improving the accuracy of business decision analysis results, thus providing reliable intelligent support for enterprise business portfolio management. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the system modules of the present invention; Figure 2 This is a schematic diagram of the electronic device structure of the present invention. Detailed Implementation

[0017] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0018] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0019] Secondly, the term "an embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in a preferred embodiment" appearing in different places throughout this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that mutually excludes other embodiments.

[0020] Please see Figure 1 As shown, this invention provides an intelligent business decision analysis and risk early warning system based on a large model, including: The initialization module is used to acquire marketing data for each business in the historical period, and pre-build a multi-dimensional feature extraction framework based on the marketing data to extract business feature vectors under the multi-dimensional feature extraction framework. In multi-business scenarios, it is often necessary to customize differentiated marketing strategies for different business lines. For businesses with good marketing results, more resources are allocated to expand their advantages, while for underperforming businesses, marketing resources need to be reduced to avoid ineffective investment. Of course, for certain product categories, market fluctuations are unique; for example, seasonal consumer goods, sales are significantly affected by holidays or weather, and they are generally allocated fixed marketing resources. In this implementation, the initialization module first collects all marketing data from the historical period, including data such as advertising spending, user click-through rate, conversion rate, and repurchase rate. Then, based on this data, a feature extraction framework is constructed. Specifically, a CNN convolutional neural network is used as the basic architecture to build the feature extraction network, thereby transforming the marketing data into business feature vectors. The system employs a combination of convolutional layers, pooling layers, and fully connected layers. First, it acquires marketing data for each business within a historical period, cleans and standardizes the data, removing outliers and missing values. Then, by constructing time-series samples, it combines marketing data from multiple consecutive time points into an input matrix, which serves as the input to the CNN network. Local temporal features are extracted through convolutional layers, and dimensionality is reduced by pooling layers to retain key information. Finally, a fixed-dimensional business feature vector is output through a fully connected layer. This is essentially a feature mapping process, where historical marketing data is used as both training and validation sets to train the feature extraction network, providing a foundation for subsequent marketing data mapping. This eliminates the need for manual intervention in feature engineering design and selection. The initialization module includes a data acquisition unit, used to collect marketing data from multiple data sources for each business within a historical marketing period. The feature construction unit is used to extract multi-dimensional features from marketing data based on a preset multi-dimensional feature extraction framework, and generate an initial feature set including statistical features, time-series features and cross features. The feature filtering unit is used to sort the importance of the initial feature set and remove redundancies, filter out the key features that meet the business decision-making objectives, and combine them into a business feature vector. For the feature construction unit, during execution, it first cleans and standardizes the raw marketing data to eliminate interference from differences in units. Then, it extracts statistical features of each dimension of the marketing data through preset statistical rules. For numerical marketing metrics, such as ad spend and conversion revenue, it calculates statistical quantities such as mean, variance, and skewness. For time-series data, such as fluctuations in daily active users or click-through rate, it uses a sliding window mechanism to capture periodic fluctuations and trend changes to extract time-series features, such as peak user activity, average activity, and fluctuation range between 8 PM and 10 PM each day. For cross-features, it combines marketing metrics from different dimensions in a non-linear way, such as multiplying ad impressions by user dwell time to quantify the potential impact of content attractiveness on conversion. After the time-series and cross-features are output, further standardization processing will be performed to ensure that all features are of the same magnitude. Specifically, this can be represented as a normalized multidimensional vector matrix. For example, if a business's average daily advertising expenditure in summer is 500,000 yuan, the average daily click-through rate is 2.5%, and weekend activity is 15% higher than weekday activity, then its initial feature set can be represented as a vector of [50, 2.5, 15] and its derived statistics. Afterward, a feature filtering unit will sort and remove redundancies from the initial feature set. The purpose is to filter out features that contribute little to business decision analysis, reduce subsequent computation, and avoid interference from feature redundancy. Specifically, this can be quantitatively evaluated by calculating the mutual information between each dimension of the initial feature set and the target variable. The calculation formula is as follows: ; in, The dimensional features in the initial feature set, For business marketing effectiveness target variables, for and The joint probability distribution, and These are the marginal probability distributions of X and Y, respectively. The greater the mutual information, the higher the contribution of the corresponding dimensional features to the target variable. All features are sorted from largest to smallest based on their mutual information values. A preset percentage of features are then retained after sorting, typically set to the top 30% to 50%, thereby constructing a high signal-to-noise ratio business feature vector. This provides a corresponding preliminary data foundation for subsequent marketing decisions.

[0021] The associated business identification module, connected to the initialization module, is used to perform correlation analysis on the feature vectors of various businesses, identify and output associated business combinations, evaluate the synergy effect of each associated business combination, and generate business synergy index. After the business feature vectors are extracted, they can be input into the related business identification module to identify the correlation between different businesses and filter out related business combinations. The purpose of setting up related business combinations is that there are often implicit synergistic relationships between different businesses. For example, users who buy maternal and infant products often pay attention to infant complementary foods. Obviously, there is a corresponding synergistic effect between these two types of businesses. Of course, there are also some weakly correlated related business combinations. For example, users who buy sports shoes have a low correlation with the demand for sports towels. Although there is a potential cross-selling opportunity, the synergistic effect is weak. Therefore, the output of business synergy indicators can clarify the marketing priority of each related business combination. The related business identification module includes a correlation analysis unit, which is used to calculate the correlation of each business feature vector and output the correlation strength value between different business feature vectors. Clustering units are used to cluster services based on association strength values, identify and output related service combinations; The collaborative evaluation unit is used to evaluate the synergistic effect of the joint performance of the businesses within each related business combination over a historical period, calculate the complementarity and substitutability between businesses, and generate business synergy indicators. When the correlation analysis unit is executed, it first calculates the correlation coefficient between any two business feature vectors to characterize the degree of correlation between different businesses. The specific calculation formula is as follows: ; in, This represents the correlation strength value between two business feature vectors. and This represents two business feature vectors used for correlation calculation. and These represent the mean values ​​of the two business feature vectors, respectively. Then, the clustering unit groups the services based on the association strength value. Services with an association strength higher than a preset threshold are aggregated into the same associated service group, completing the automatic identification and grouping of associated services. In practice, an association strength threshold is first set, usually between 0.6 and 0.8, to avoid noise interference. Only service pairs with an association strength value greater than or equal to the threshold are clustered. Starting with each service as an initial cluster, the nearest clusters are gradually merged until all services meeting the criteria are included in the corresponding group, forming the final associated service group. Afterward, the collaborative evaluation unit calculates the complementarity score and the substitutability score of the services within the group. The complementarity score is obtained by calculating the overlap of the user group characteristics covered by different services within the group; the lower the overlap, the higher the complementarity. The substitutability score is obtained by calculating the degree of competition for the same marketing resources by the services within the group. Substitutability score = variance of return on investment for businesses within the same business portfolio on the same marketing resources + overlap rate of user conversion paths; Complementarity score = 1 - the overlap of user group characteristics covered by the related business portfolio, where the overlap of user group characteristics is the ratio of the consumption intersection of the target customer groups between different businesses to the total set. After obtaining the complementarity score and the substitution score, the complementarity score and the substitution score are weighted and integrated. The weighting formula adopts the conventional linear weighting method, that is, business synergy index = a × complementarity score - b × substitution score. Generally speaking, the value of a should be greater than b, in order to highlight the positive benefits brought by the complementarity between businesses. Of course, there are also special scenarios. For example, when the enterprise is in the stage of stock competition and the internal competition for resources is fierce, the weight of b can be appropriately increased to strengthen the early warning of internal resource consumption.

[0022] The causal identification module, connected to the related business identification module, is used to identify the causal relationship between each business in the related business combination, output the dominant business and subordinate business, and quantify the causal strength of the dominant business and subordinate business to generate causal influence coefficient. Once the related business portfolio is determined, the causal direction of different businesses within the portfolio will be assessed to clarify the priority and dependencies of resource allocation, and to identify the dominant and subordinate businesses. For example, when there is a strong correlation between maternal and infant products and complementary foods for infants, it is necessary to determine whether the increase in sales of maternal and infant products drives the increase in sales of complementary foods, or whether the demand for complementary foods in turn drives the sales of maternal and infant products. Generally speaking, for younger infants, maternal and infant products are often a prior demand (corresponding to the dominant business), while complementary foods are a subsequent derivative consumption (corresponding to the subordinate business). In this case, marketing resources should be prioritized for maternal and infant products. However, for older infants or groups with specific complementary food preferences, a reverse-driven characteristic will emerge. In this case, complementary foods become the dominant business, and maternal and infant products become the subordinate business. Marketing resources should then be tilted towards complementary foods. After the dominant and subordinate businesses are identified, the causal strength will be quantitatively assessed. This is converted into specific numerical values ​​that can characterize the magnitude of causal influence. The causal identification module includes a causal identification unit, which determines the causal direction between businesses in each related business combination based on the temporal variation patterns and interaction paths of business feature vectors within historical periods. It outputs the leading business and the subordinate businesses affected by it. In practice, a time-series dataset between businesses is first constructed, using key indicators (such as sales revenue, user visits, conversion rates, etc.) within a continuous time window as dependent and independent variables. The stationarity of the original observation sequences of each business indicator over time is first tested. The time-series data of each business is differencing until the sequence reaches a stationary state, and the differencing order is determined. Then, an autoregressive model is constructed based on the differencing order, introducing lag terms as explanatory variables. The autoregressive model is expressed as: In the formula, This represents the observation value of the time series at time t. For constant terms, The coefficient for the time trend term. Here, p is the coefficient of the lag term, and p is the lag order. This represents the white noise error term. The lag term coefficients are tested. To determine if a sequence is significantly non-zero, check if it contains a unit root. If the value is significantly non-zero, the null hypothesis of the existence of a unit root is rejected, indicating that the series is stationary. Otherwise, the difference needs to be continued until stationary. After determining that the time series of each business is stationary, the significance of the lagged term of each business on the prediction results of other businesses can be tested in turn. If the lagged term of a certain business A can significantly improve the prediction accuracy of business B, it is determined that business A has a one-way causal effect on business B. If both the two-way test are significant, the master-slave relationship is determined by combining the business attributes and the time sequence. Usually, the business with the demand generated earlier is defined as the master business, and the business with the demand generated later is defined as the subordinate business. The causal quantification unit is used to numerically measure the causal effect by comparing the response magnitude of subordinate businesses before and after the change of the dominant business, and combining the degree of dependence between subordinate businesses and the dominant business, and to generate a causal influence coefficient. The causal quantification unit includes a heterogeneity correction subunit, which, after generating the causal influence coefficient, performs heterogeneity analysis on the response differences of subordinate businesses at different levels of the dominant business based on the value changes of the dominant business in historical states, and corrects the causal influence coefficient based on the analysis results, outputting the heterogeneously corrected causal influence coefficient. Specifically, the causal identification module consists of a causal quantification unit and a causal identification unit. When the causal identification unit is executed, it first performs sliding segmentation of the business feature vector according to a preset time window, and counts the order of changes of the dominant business and subordinate business within each time window: If, within a time window, a leading business indicator undergoes a positive change, and a subordinate business indicator follows suit with a positive change, and their proportion across all time windows exceeds the threshold for the same period's change, then this direction is considered the causal direction. This determines the leading and subordinate relationship. The threshold for the same period's change is set to eliminate spurious correlation interference from random fluctuations, and its value is generally not less than 0.8. After obtaining the causal direction, the causal quantification unit uses the difference-in-differences method to calculate the average treatment effect brought about by the change in the leading business. The period in which the leading business undergoes a significant change is taken as the treatment period (generally, a significant change refers to the month-on-month or year-on-year change in the leading business indicator exceeding twice the standard deviation of the historical mean). Samples of leading business changes during this period are used as the treatment group. Then, the same leading business is taken from historical windows where no significant changes have occurred. Samples with the same business and product category attributes are selected as candidate control groups. Additionally, covariates that were determined before the change in the dominant business and simultaneously affect both the dominant and subordinate businesses are selected. These include market size, market environment, and marketing investment variables. Market size variables correspond to the baseline levels of the dominant and subordinate businesses in the previous window, as well as average order value, reflecting overall market capacity and demand fluctuations. Market environment variables correspond to industry prosperity indices, seasonal factors, and user activity base, controlling for interference from uncontrollable external factors in causal identification. Marketing investment variables correspond to the advertising budget and distribution channels in the previous window, controlling for the endogeneous impact of marketing strategies on business changes. The probability of each candidate control group entering the treatment group is then calculated by estimating propensity scores, using the following formula: ; In the formula, Indicates a preference for scoring. Representing covariates, This represents the intercept term, which is the constant term in logistic regression. The weight coefficients of each covariate represent the weights of the corresponding covariates in the Logistic regression, which are obtained through maximum likelihood estimation. Then, the candidate control group is matched based on the propensity score, and the sample with the closest propensity score to the treatment group is selected as the final control group. Afterwards, the average difference between the subordinate business indicators in the two groups is compared, and the average difference value is the initial causal influence coefficient. The specific calculation formula is as follows: Average causal impact coefficient = Average change in subordinate business of the treatment group - Average change in subordinate business of the control group; The heterogeneity correction subunit then further corrects the results, specifically by calculating the response magnitude of the subordinate business in each growth interval based on the different growth intervals in the history of the leading business. Response magnitude = Change in subordinate business / Change in dominant business; If the response magnitude of subordinate businesses in the low-growth range of the dominant business is significantly higher than that in the high-growth range, it indicates the existence of a diminishing marginal effect. In this case, the initial coefficient needs to be adjusted downward. Conversely, if the response magnitude in the high-growth range is better, the initial causal influence coefficient needs to be adjusted upward. The specific adjustment process can be achieved using the weighted average method, that is, assigning corresponding weights according to the sample size proportion of each growth range, and summing the weighted local influence coefficients of each range after correction, thereby obtaining the causal influence coefficient after heterogeneity correction.

[0023] The comprehensive evaluation module, connected to the related business identification module and the causal identification module, is used to comprehensively evaluate the business performance of the related business portfolio based on the business synergy index and the causal influence coefficient, and generate a business portfolio health score. To more accurately quantify the overall effectiveness of the business portfolio, a comprehensive evaluation module is used to integrate business synergy indicators and causal influence coefficients to conduct a multi-dimensional health assessment of related business portfolios, thereby generating a business portfolio health score and providing management with an intuitive basis for decision-making. The comprehensive evaluation module includes a weight allocation unit, which is used to allocate weight factors to business synergy indicators and causal influence coefficients based on the historical marketing objectives and results of related business portfolios. The weight allocation unit includes a time-series adjustment subunit, which is used to calculate and output the time-series adjustment factors of the business synergy index and the causal influence coefficient through a sliding window mechanism, and to adjust the weight factors of the business synergy index and the causal influence coefficient in real time based on the time-series adjustment factors. The comprehensive scoring unit is used to perform weighted calculations on each business within the related business portfolio based on the weight allocation results of business synergy indicators and causal influence coefficients, and generate a business portfolio health score. The comprehensive scoring unit also includes a trend analysis subunit, which is used to perform trend correction on the business portfolio health score based on the changing trend of the business portfolio health score within a continuous time window, and complete the calibration of the business portfolio health score. Specifically, during the weight allocation unit's execution, the business synergy index, causal influence coefficient, and final marketing achievement results are first collected from multiple consecutive historical time windows. Specifically, the historical marketing cycle is divided into multiple equal-length sliding windows, and the correlation coefficient between the business synergy index, causal coefficient, and marketing results within each window is calculated. This quantifies the contribution of each index to the marketing results. The specific calculation formula can be found in the correlation coefficient calculation formula mentioned above in the business feature vector section, and will not be repeated here. Finally, initial weights are assigned to the business synergy index and causal influence coefficient based on the contribution ratio. Furthermore, considering the impact of time factors on index stability, a time-series adjustment subunit is executed to correct the initial weights. Specifically, the weight contribution corresponding to each time window is calculated and marked as a time-series adjustment factor. Time windows closer to the current cycle have a higher contribution, thus assigning a higher weight to recent windows. The weight of longer-term windows gradually decreases, ensuring that the weights adapt to the dynamic changes in the business environment and avoiding interference from outdated historical data in current decisions. The calculation formula for the time-series adjustment factor is as follows: ; In the formula, This represents the time-series adjustment factor, where t represents the current period and the [missing information - likely a specific time period]. The time interval between windows, The attenuation coefficient; Subsequently, the comprehensive scoring unit will use time-series-corrected weights to perform a weighted summation of the synergy indicators and causal influence coefficients of various business operations to derive a basic health score. This can be achieved directly using a linear weighting method. However, considering that scores from a single period are susceptible to biases caused by unforeseen factors, the trend analysis subunit will extract health score data from multiple consecutive time windows and fit a curve showing the score's change over time (the corresponding equation for this curve is: ; In the formula, , , These represent the intercept term, linear trend coefficient, and quadratic nonlinear correction coefficient, respectively. The intercept term reflects the baseline health level, the linear trend coefficient reflects the long-term growth or decline rate of the baseline health score, and the quadratic nonlinear correction coefficient reflects the acceleration of the change in the baseline health score. When the baseline health score is higher than the trend fit value and the trend is upward, the baseline health score will be increased to amplify the positive signal. When the baseline score is lower than the trend fit value and the trend is downward, the score will be appropriately reduced to strengthen the risk warning. Finally, the calibrated business portfolio health score will be output.

[0024] The risk identification module, connected to the comprehensive assessment module, is used to classify the risk level of related business portfolios based on the health score of the business portfolio, and to match the corresponding risk warning signal according to the risk level classification result. The risk identification module categorizes business portfolios into multiple risk levels based on their health scores, such as low, medium, and high risk. When a business portfolio's health score is not lower than a preset health threshold, it is classified as low risk, and only standard business optimization suggestions are output. When the score falls between the health threshold and the risk threshold, it is classified as medium risk, and a mild warning is output, prompting operations personnel to monitor the business portfolio's synergy performance and investigate potential collaboration failures. When the score falls below the preset risk threshold, it is classified as high risk, and a severe warning is output, reminding decision-makers to adjust the business portfolio's resource allocation plan in a timely manner to avoid marketing losses caused by synergy failures. The setting of health thresholds and risk thresholds needs to be dynamically adjusted in conjunction with the distribution characteristics of historical data and industry benchmarks. For example, statistical analysis of historical health score data can be performed to determine the mean and standard deviation. With the mean as the center, the threshold range can be set in combination with the standard deviation. Specifically, the standard deviation can be directly added to or subtracted from the mean to define the threshold boundary. Alternatively, the standard deviation can be multiplied or subtracted to adapt to the risk sensitivity requirements under different business scenarios. The specific settings can be made according to the actual business scenario. The risk identification module includes a risk level division unit, which is used to divide the related business portfolios into different risk levels according to the numerical range of the health score of the business portfolio. The early warning matching unit is used to match and output the corresponding risk early warning signal from the preset set of early warning signals based on the risk level classification results. The risk level classification unit compares the calibrated health score of the business portfolio in real time based on the preset threshold range logic. After determining the corresponding risk level, it transmits it to the early warning matching unit. The early warning matching unit calls the corresponding level of early warning signal and outputs it to the management decision interface according to the preset mapping rules. At the same time, it will also mark the specific abnormal items of business synergy indicators and causal influence coefficients in the related business portfolio, clearly indicating the source of the abnormality, so as to facilitate managers to quickly locate the problem and complete the risk response.

[0025] In summary, the business feature vectors output by the initialization module provide sufficient preliminary data support for subsequent analysis. The related business identification module not only identifies the surface correlation between businesses but also provides candidate combinations to be verified for the causal identification module, avoiding meaningless causal searches across all businesses. The causal identification module further distinguishes the true causal driving relationship based on correlation. Its output causal influence coefficient and the synergy index output by the related business identification module are fused in the comprehensive evaluation module, solving the problem of the one-sidedness of single-dimensional evaluation. Then, the risk identification module transforms the evaluation results into actionable early warning signals, which can be fed back to the initialization module to guide the key direction of the next round of feature extraction, realizing the adaptive optimization of the system. This forms a progressive and mutually input-output coupling relationship, enabling the system to automatically and accurately extract decision signals.

[0026] Please see Figure 2 An electronic device, comprising: At least one processor; and memory that is communicatively connected to at least one processor; The memory stores a computer program that can be executed by at least one processor, which enables the at least one processor to execute the aforementioned intelligent business decision analysis and risk warning system based on a large model.

[0027] The above description is merely a preferred embodiment of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention. Structures, devices, and operating methods not specifically described or explained in this invention are implemented according to conventional methods in the art unless otherwise specified or limited.

Claims

1. A large-model-driven intelligent business decision analysis and risk early warning system, characterized by: include: The initialization module is used to acquire marketing data for each business in the historical period, and pre-build a multi-dimensional feature extraction framework based on the marketing data to extract business feature vectors under the multi-dimensional feature extraction framework. The related business identification module is used to perform correlation analysis on the feature vectors of various businesses, identify and output related business combinations, and evaluate the synergy effect of each related business combination to generate a business synergy index. The causal identification module is used to identify the causal relationships among the various businesses in the related business combination, output the dominant business and subordinate business, quantify the causal strength of the dominant business and subordinate business, and generate causal influence coefficients. The comprehensive evaluation module is used to comprehensively evaluate the business performance of related business portfolios based on business synergy indicators and causal impact coefficients, and generate a business portfolio health score. The risk identification module is used to classify the risk level of related business portfolios based on the health score of the business portfolio, and to match the corresponding risk warning signal according to the risk level classification results.

2. The intelligent business decision analysis and risk early warning system based on large model-driven system according to claim 1, characterized in that: The initialization module includes: The data acquisition unit is used to collect marketing data from multiple data sources for each business during historical marketing cycles. The feature construction unit is used to extract multi-dimensional features from marketing data based on a preset multi-dimensional feature extraction framework, and generate an initial feature set including statistical features, time-series features and cross features. The feature filtering unit is used to sort the importance of the initial feature set and remove redundancies, filter out the key features that meet the business decision-making objectives, and combine them into a business feature vector.

3. The intelligent business decision analysis and risk early warning system based on large model-driven according to claim 1, characterized in that: The associated business identification module includes: The correlation analysis unit is used to calculate the correlation between various business feature vectors and output the correlation strength values ​​between different business feature vectors. Clustering units are used to cluster services based on association strength values, identify and output combinations of related services; The collaborative evaluation unit is used to evaluate the synergistic effect of the joint performance of businesses within each related business portfolio over a historical period, calculate the complementarity and substitutability between businesses, and generate business synergy indicators.

4. The intelligent business decision analysis and risk early warning system based on large model-driven system according to claim 1, characterized in that: The causality identification module includes: The causal identification unit is used to determine the causal direction between businesses in each related business combination based on the temporal change pattern and interaction path of the business feature vector in the historical period, and output the leading business with the guiding role and the subordinate businesses affected by it. The causal quantification unit is used to numerically measure the causal effect by comparing the response magnitude of subordinate businesses before and after the change in the dominant business, and combining the degree of dependence between subordinate businesses and the dominant business, and to generate a causal influence coefficient.

5. The intelligent business decision analysis and risk early warning system based on large model-driven according to claim 1, characterized in that: The causal quantification unit includes a heterogeneity correction subunit, which, after generating the causal influence coefficient, performs heterogeneity analysis on the response differences of subordinate businesses at different levels of the dominant business based on the value changes of the dominant business in historical states, and corrects the causal influence coefficient based on the analysis results, outputting the heterogeneously corrected causal influence coefficient.

6. The intelligent business decision analysis and risk early warning system based on large model driving according to claim 1, characterized in that: The comprehensive evaluation module includes: The weighting allocation unit is used to allocate weighting factors to business synergy indicators and causal influence coefficients based on the historical marketing objectives and results of related business combinations. The comprehensive scoring unit is used to perform weighted calculations on each business within the related business portfolio based on the weight allocation results of business synergy indicators and causal influence coefficients, and generate a business portfolio health score.

7. The intelligent business decision analysis and risk early warning system based on large model driving according to claim 6, characterized in that: The weight allocation unit includes a time-series adjustment subunit, which is used to calculate and output the time-series adjustment factors of the business synergy index and the causal influence coefficient through a sliding window mechanism, and to adjust the weight factors of the business synergy index and the causal influence coefficient in real time based on the time-series adjustment factors.

8. The intelligent business decision analysis and risk early warning system based on large model driving according to claim 6, characterized in that: The comprehensive scoring unit also includes a trend analysis subunit, which is used to perform trend correction on the business portfolio health score based on the changing trend of the business portfolio health score within a continuous time window, thereby completing the calibration of the business portfolio health score.

9. The intelligent business decision analysis and risk early warning system based on large model-driven as described in claim 1, characterized in that: The risk identification module includes: The risk level classification unit is used to classify related business portfolios into different risk levels based on the numerical range of the business portfolio health score. The early warning matching unit is used to match and output the corresponding risk early warning signal from the preset set of early warning signals based on the risk level classification results.

10. An electronic device, characterized in that: The electronic device includes: At least one processor; and a memory communicatively connected to the at least one processor; The memory stores a computer program that can be executed by the at least one processor, which is then executed by the at least one processor to enable the at least one processor to execute the intelligent business decision analysis and risk warning system based on a large model as described in any one of claims 1 to 9.

Images