Planning and effect evaluation method and system for enterprise marketing activities

Abstract

The invention provides a planning and effect evaluation method and system for enterprise marketing activities, and the method comprises the steps: generating a marketing strategy according to an enterprise target and a constraint condition based on a conditional generative adversarial network; performing causal effect analysis on the generated marketing strategy through a causal forest model, and evaluating the robustness of the strategy in combination with anti-fact reasoning; the market trend is captured in real time by using an online generative adversarial network, and marketing strategy parameters are dynamically adjusted in combination with a Bayesian optimization algorithm. According to the method, a feasible marketing strategy meeting constraints is efficiently generated based on the conditional generative adversarial network, the effect and stability of the strategy are accurately evaluated through the causal forest model in combination with anti-factual reasoning, the risk cost is reduced, and the strategy dynamically adapts to market changes through the online generative adversarial network and the Bayesian optimization algorithm, so that the marketing efficiency is improved. And the long-term marketing effect is optimized.

Description

Technical Field

[0001] The present invention relates to the technical field of enterprise marketing planning, and in particular to a method and system for planning and effect evaluation of enterprise marketing activities. Background Art

[0002] In today's highly competitive business environment, corporate marketing activities play a vital role in enhancing brand awareness, expanding market share and increasing sales.

[0003] Traditional corporate marketing planning relies primarily on manual experience and market research. Marketers develop marketing strategies based on their industry experience and initial understanding of the market, combined with the limited information obtained from market research.

[0004] However, this approach has many flaws: first, manual experience is subjective and limited, and different marketers have different levels of experience and cognition, resulting in uneven quality of generated marketing strategies; second, market research usually requires a lot of time, manpower and material costs, and the research results may have a certain lag and cannot reflect the latest market dynamics in a timely manner; in addition, traditional methods make it difficult to fully consider the various goals and constraints set by the enterprise, such as budget constraints, time requirements, brand image, etc., resulting in the generated marketing strategy may not meet the actual needs of the enterprise. Summary of the Invention

[0005] Based on this, the purpose of the present invention is to propose a method and system for planning and evaluating the effects of corporate marketing activities to solve the above-mentioned problems.

[0006] According to the present invention, a method for planning and evaluating the effects of corporate marketing activities comprises:

[0007] Generate marketing strategies based on enterprise goals and constraints based on conditional generative adversarial networks;

[0008] The causal effect analysis of the generated marketing strategy is carried out through the causal forest model, and the robustness of the strategy is evaluated by combining counterfactual reasoning;

[0009] Use online generative adversarial networks to capture market trends in real time, and combine them with Bayesian optimization algorithms to dynamically adjust marketing strategy parameters.

[0010] Furthermore, the steps of generating a marketing strategy based on the conditional generative adversarial network according to the enterprise goals and constraints include:

[0011] Define enterprise goals and constraints, and convert them into conditional vectors to be input into the generator of the conditional generative adversarial network;

[0012] The generator of the generative adversarial network generates a marketing strategy vector based on the input condition vector and random noise, including advertising creativity, channel combination, promotion intensity and delivery rhythm;

[0013] The discriminator of the generative adversarial network evaluates the feasibility of the generated marketing strategy and feeds back the gradient to update the generator until a marketing strategy that meets the constraints and has high feasibility is generated.

[0014] Furthermore, the steps of performing causal effect analysis on the generated marketing strategy using the causal forest model and evaluating the robustness of the strategy using counterfactual reasoning include:

[0015] Use the causal forest model to train historical marketing strategy data to quantify the causal effect of each strategy variable;

[0016] Evaluate the expected effectiveness of the generated marketing strategies based on the output of the causal forest model;

[0017] Combine counterfactual reasoning to generate simulation scenarios and test the performance of marketing strategies under different market conditions.

[0018] Furthermore, the step of using the causal forest model to train historical marketing strategy data to quantify the causal effect of each strategy variable includes:

[0019] Collect historical marketing strategy data, including strategy variables, market characteristics, and user behavior data;

[0020] Use the causal forest model to train historical marketing strategy data to learn the impact of each strategy variable on marketing effectiveness;

[0021] Output the estimated causal effect of each strategy variable.

[0022] Furthermore, the steps of using an online generative adversarial network to capture market trends in real time and dynamically adjusting marketing strategy parameters in combination with a Bayesian optimization algorithm include:

[0023] Use online generative adversarial networks to capture and analyze market data in real time, including social media popularity, advertising platform CTR, and competitor dynamics;

[0024] Generate simulated data that matches current market trends based on the output of an online generative adversarial network;

[0025] Use Bayesian optimization algorithms to dynamically adjust marketing strategy parameters based on simulated data and historical observations.

[0026] Furthermore, the step of dynamically adjusting marketing strategy parameters using the Bayesian optimization algorithm based on simulation data and historical observation results includes:

[0027] Define the marketing strategy parameters to be optimized;

[0028] The expected effect of the marketing strategy predicted by the causal forest model is used as the objective function of Bayesian optimization;

[0029] The parameters are iteratively adjusted using the Bayesian optimization algorithm until the objective function converges.

[0030] Furthermore, the step of iteratively adjusting parameters by using the Bayesian optimization algorithm until the objective function reaches convergence includes:

[0031] Initialize the parameter combination and calculate the corresponding objective function value;

[0032] Generate the next set of new parameter combinations based on the recommendations of the Bayesian optimization algorithm;

[0033] Calculate the objective function value for the new parameter combination and update the prior distribution of the Bayesian optimization model;

[0034] The iterative process is repeated until the change in the objective function value is less than the preset threshold or the maximum number of iterations is reached.

[0035] The present invention further proposes a system for planning and evaluating the effects of corporate marketing activities, which is used to implement the above-mentioned method for planning and evaluating the effects of corporate marketing activities. The system includes:

[0036] Strategy Generation Module: Used to generate marketing strategies based on the enterprise goals and constraints based on the conditional generative adversarial network;

[0037] Strategy effect evaluation module: used to analyze the causal effects of the generated marketing strategies through the causal forest model and evaluate the robustness of the strategies in combination with counterfactual reasoning;

[0038] Strategy parameter adjustment module: used to use online generative adversarial networks to capture market trends in real time and dynamically adjust marketing strategy parameters in combination with Bayesian optimization algorithms.

[0039] In summary, the planning and effect evaluation method for enterprise marketing activities of the present invention first converts enterprise goals and constraints into marketing strategy vectors based on conditional generative adversarial networks, thereby efficiently generating marketing strategies that meet the constraints and are feasible, significantly improving the efficiency of strategy formulation. Secondly, the causal forest model is used to conduct causal effect analysis on the generated marketing strategies, and the expected effects of the marketing strategies are accurately evaluated by quantifying the causal effects of each strategy variable; at the same time, simulation scenarios are generated in combination with counterfactual reasoning to test the robustness of the strategies, which can accurately measure the effectiveness of the strategies and their stability, effectively reducing the risks and costs brought about by blindly implementing strategies. Finally, an online generative adversarial network is used to capture and analyze market data in real time to generate simulation data, and then the Bayesian optimization algorithm is used to dynamically adjust the marketing strategy parameters, so that the strategies can dynamically adapt to market changes and optimize long-term marketing effects.

[0040] Additional aspects and advantages of the present invention will be set forth in part in the following description and, in part, will be obvious from the following description, or may be learned through embodiments of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments with reference to the following drawings, in which:

[0042] Figure 1 This is a flow chart of a method for planning and evaluating the effectiveness of corporate marketing activities according to the first embodiment of the present invention;

[0043] Figure 2 This is a system block diagram of a system for planning and evaluating the effects of corporate marketing activities according to the second embodiment of the present invention. DETAILED DESCRIPTION

[0044] To facilitate understanding of the present invention, the present invention will be described more fully below with reference to the accompanying drawings. The drawings illustrate several embodiments of the present invention. However, the present invention may be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and comprehensive understanding of the present invention.

[0045] It should be noted that when an element is referred to as being "fixed to" another element, it may be directly on the other element or there may be an intermediate element. When an element is referred to as being "connected to" another element, it may be directly connected to the other element or there may be an intermediate element. The terms "vertical," "horizontal," "left," "right," and similar expressions used herein are for illustrative purposes only.

[0046] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this invention pertains. The terms used in this specification of the present invention are for the purpose of describing specific embodiments only and are not intended to limit the present invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0047] Example 1

[0048] See also Figure 1 The present invention proposes a method for planning and evaluating the effects of corporate marketing activities, which includes steps S101 to S103:

[0049] S101, based on conditional generative adversarial networks, generates marketing strategies based on corporate goals and constraints.

[0050] Further optionally, the step of generating a marketing strategy based on a conditional generative adversarial network according to enterprise goals and constraints includes:

[0051] Define enterprise goals and constraints, and convert them into conditional vectors to be input into the generator of the conditional generative adversarial network;

[0052] The generator of the generative adversarial network generates a marketing strategy vector based on the input condition vector and random noise, including advertising creativity, channel combination, promotion intensity and delivery rhythm;

[0053] The discriminator of the generative adversarial network evaluates the feasibility of the generated marketing strategy and feeds back the gradient to update the generator until a marketing strategy that meets the constraints and has high feasibility is generated.

[0054] As you can understand, first define the company's goals and constraints, convert them into conditional vectors, and input them into the generator of the conditional generative adversarial network. Companies often have clear goals when conducting marketing activities, such as increasing product sales or increasing market share. At the same time, companies also face various constraints, such as budget constraints and time limits. Clearly define the company's goals and constraints and convert them into a form that computers can process. Suppose the company's goal is to increase product sales, and the constraint is that the budget does not exceed B. Convert the sales target T and the budget constraint B into a conditional vector c = (T, B).

[0055] The generator then generates a marketing strategy vector based on the input conditional vector and random noise. Generator G is a neural network whose function is to generate new data based on the input conditional vector and random noise. The introduction of random noise increases the diversity of the generated data and prevents it from being too monotonous. For example, when generating marketing strategies, even for the same corporate objectives and constraints, the generated strategies will vary due to differences in random noise. By learning from a large amount of historical marketing strategy data, the generator can generate parameters related to various marketing strategies based on corporate objectives and constraints, providing a rich sample of strategies for subsequent evaluation and optimization.

[0056] Specifically, the input of the generator G is the condition vector c and random noise z (obeying Gaussian distribution z~N(0,I)), and the output is the marketing strategy vector s=G(c,z). The marketing strategy vector s can include the advertising creativity score s1, the channel combination weight vector s2=(s 21 , s 22 ,…,s 2n )(n is the number of channels), promotion intensity s3 and delivery rhythm time series s4 = (s 41 , s 42 ,…,s 4m )(m is the number of delivery time periods).

[0057] The discriminator evaluates the feasibility of the generated marketing strategies and provides feedback gradients to update the generator until a highly feasible marketing strategy that satisfies the constraints is generated. The discriminator D, also a neural network, distinguishes between the generated marketing strategies and real historical marketing strategies. By providing feedback gradients, it guides the generator to produce more realistic and feasible marketing strategies, thereby improving the quality of the generated strategies. By learning from a large number of real and generated strategies, it can determine whether a strategy aligns with actual market conditions and the company's operating rules. For example, if a generated strategy has a high ad creative score but an unbalanced distribution of channel weights, resulting in an unbalanced budget allocation, the discriminator may deem the strategy infeasible and assign a low probability value.

[0058] Specifically, the input of the discriminator is the generated marketing strategy vector s or the real historical marketing strategy vector s real , the output is a scalar D(s) or D(s real ), which represents the probability that the strategy is a real feasible strategy. The loss function of the discriminator is:

[0059]

[0060] The loss function of the generator is:

[0061] L G =-Ez～N(0,I),c [log D(G(c,z))].

[0062] The discriminator's loss function measures its ability to distinguish between the true and generated policies. It aims to output a higher probability for the true policy and a lower probability for the generated policy. The generator's loss function aims to deceive the discriminator into believing the generated policy is the real policy.

[0063] During training, the generator's parameters are updated using the backpropagation algorithm based on the feedback gradients from the discriminator. If the discriminator deems the generated strategy infeasible, it returns a larger gradient, guiding the generator to adjust its parameters, bringing the generated marketing strategy closer to a realistic and feasible strategy while satisfying both the company's objectives and constraints. This process continues iteratively until the generator is able to generate a highly feasible marketing strategy that satisfies the constraints.

[0064] This embodiment uses a generator to generate parameters related to various marketing strategies based on conditional vectors and random noise, providing a rich set of strategy samples for subsequent evaluation and optimization. A discriminator distinguishes generated strategies from real historical strategies, and feedback gradients guide the generator to produce more realistic and feasible marketing strategies, improving the quality of generated strategies.

[0065] S102, conduct causal effect analysis on the generated marketing strategy through the causal forest model, and combine it with counterfactual reasoning to evaluate the robustness of the strategy.

[0066] Further optionally, the step of performing causal effect analysis on the generated marketing strategy using a causal forest model and evaluating the robustness of the strategy using counterfactual reasoning includes:

[0067] Use the causal forest model to train historical marketing strategy data to quantify the causal effect of each strategy variable;

[0068] Evaluate the expected effectiveness of the generated marketing strategies based on the output of the causal forest model;

[0069] Combine counterfactual reasoning to generate simulation scenarios and test the performance of marketing strategies under different market conditions.

[0070] It is understandable that when analyzing the causal effect of the generated marketing strategy, first, the causal forest model is used to train the historical marketing strategy data to quantify the causal effect of each strategy variable. Assume that the historical marketing strategy data is Historical marketing strategy data includes various marketing strategies implemented by the company in the past and their corresponding effects. i is the strategy variable vector of the i-th sample, including advertising creativity, channel combination, promotion intensity, etc., y iis the corresponding marketing effect indicator, such as sales. The causal forest model constructs multiple decision trees, each tree divides the data, and calculates the average causal effect of the strategy variable on the marketing effect in each divided area. j , the average causal effect in a certain partition area R is estimated to be:

[0071]

[0072] Among them, y i (x ij +δ) represents the predicted marketing effect value after increasing the j-th strategy variable of the i-th sample by δ (which can be obtained through model prediction), where δ is the increment of the strategy variable, and y i (x ij ) is the original marketing effect prediction value, and |R| is the number of samples in region R.

[0073] Based on the output of the causal forest model, the expected effect of the generated marketing strategy is evaluated. Specifically, for the generated marketing strategy vector s, its strategy variables are substituted into the causal forest model, the causal effect of each strategy variable is calculated, and the expected effect of the strategy is obtained. The expected effect E(s) can be expressed as:

[0074]

[0075] Among them, β0 is the intercept term, β j is the weight of the causal effect of the j-th strategy variable, is the estimated causal effect of the jth strategy variable in the marketing strategy vector s, and k is the number of strategy variables.

[0076] Combined with counterfactual reasoning, simulation scenarios are generated to test the performance of marketing strategies under different market conditions to assess their robustness. Counterfactual reasoning involves varying market condition variables (such as market demand and competitive intensity), assuming these variables take on different values, and then using the causal forest model to predict the effectiveness of the marketing strategy under these simulated scenarios. For example, if the counterfactual scenario is a sudden increase in market demand by 30% or a decrease in competitive intensity by 20%, the marketing effectiveness under these counterfactual scenarios is predicted. Multiple counterfactual scenarios are generated based on the possible range of variation of the market condition variables. For example, market demand can be set to increase by 10%, 20%, or 30%, while competitive intensity can be set to decrease by 10% or 20%. By combining these variables, multiple counterfactual scenarios can be generated. For each counterfactual scenario, the causal forest model is used to predict the effectiveness of the marketing strategy. The robustness of the marketing strategy is then assessed by comparing the marketing effectiveness under different counterfactual scenarios. If a marketing strategy can maintain good results in most counterfactual scenarios, it means that the strategy has strong robustness; conversely, if the results fluctuate greatly, it means that the strategy is more sensitive to changes in market conditions and has poor robustness.

[0077] This embodiment uses a causal forest model to train historical marketing strategy data, quantifying the causal effects of each strategy variable on marketing effectiveness. This helps companies understand the true impact of each strategy variable and provides a scientific basis for strategy evaluation. Based on the output of the causal forest model, the generated marketing strategy is evaluated for its expected effectiveness, allowing companies to understand the potential effects of the strategy in advance and providing a reference for decision-making. Furthermore, by generating different counterfactual scenarios, the marketing strategy's performance under different market conditions is tested, assessing its robustness and ensuring that the strategy maintains its effectiveness across various market environments.

[0078] Further optionally, the step of using the causal forest model to train historical marketing strategy data to quantify the causal effect of each strategy variable includes:

[0079] Collect historical marketing strategy data, including strategy variables, market characteristics, and user behavior data;

[0080] Use the causal forest model to train historical marketing strategy data to learn the impact of each strategy variable on marketing effectiveness;

[0081] Output the estimated causal effect of each strategy variable.

[0082] It is understandable that when using the causal forest model to train historical marketing strategy data to quantify the causal effect of each strategy variable, first, historical marketing strategy data is collected, including strategy variables, market characteristics, and user behavior data. Assume that the collected historical data is Among them, xi is the strategy variable vector, m i is the market characteristic vector (such as market size, market growth rate, etc.), u i is the user behavior data vector (such as user click rate, conversion rate, etc.), y i It is a marketing effectiveness indicator.

[0083] The causal forest model is used to train historical data to learn the impact of each strategy variable on marketing effectiveness. Specifically, the causal forest model learns patterns and relationships in the data by continuously building decision trees. When building each decision tree, the model divides the sample based on strategy variables, market characteristics, and user behavior data, so that the samples in each divided area have similar characteristics. Then, the average causal effect of the strategy variable on the marketing effect is calculated in each divided area. For the strategy variable x j , the average causal effect in a certain partition area R is estimated to be:

[0084]

[0085] Among them, (x i , m i ,u i ) represents the strategy variable vector, market feature vector and user behavior data vector of the i-th sample, y i (x ij +δ) represents the predicted marketing effect value after increasing the j-th strategy variable of the i-th sample by δ (which can be obtained through model prediction), y i (x ij ) is the original marketing effect prediction value, and |R| is the number of samples in region R.

[0086] By training the causal forest model, we can learn the impact of each strategy variable on marketing effectiveness, providing a key basis for subsequent strategy evaluation and optimization. After the training is completed, for each strategy variable x j , calculate the average causal effect estimate over the entire dataset

[0087]

[0088] Among them, N is the total number of samples in the data set, R i is the partition area where sample i is located, is the policy variable x j In region R i The average causal effect estimate within .

[0089] The calculated causal effect estimates reflect the degree and direction of each strategy variable's impact on marketing effectiveness. For example, a positive causal effect estimate for creative advertising suggests that improving creative quality can increase marketing effectiveness. A negative causal effect estimate for promotional effort suggests that excessively increasing promotional effort may negatively impact marketing effectiveness.

[0090] These causal effect estimates can be used for subsequent strategy evaluation and optimization. When evaluating a newly generated marketing strategy, the causal effect estimates of each strategy variable can be used to predict the strategy's effectiveness. When optimizing a strategy, the values of the strategy variables can be adjusted based on the causal effect estimates to improve marketing effectiveness. Obtaining causal effect estimates for each strategy variable facilitates companies to assess the importance of different strategy variables and optimize strategies in a targeted manner.

[0091] S103 uses an online generative adversarial network to capture market trends in real time and dynamically adjusts marketing strategy parameters in combination with the Bayesian optimization algorithm.

[0092] Further optionally, the step of using an online generative adversarial network to capture market trends in real time and dynamically adjusting marketing strategy parameters in combination with a Bayesian optimization algorithm includes:

[0093] Use online generative adversarial networks to capture and analyze market data in real time, including social media popularity, advertising platform CTR, and competitor dynamics;

[0094] Generate simulated data that matches current market trends based on the output of an online generative adversarial network;

[0095] Use Bayesian optimization algorithms to dynamically adjust marketing strategy parameters based on simulated data and historical observations.

[0096] It can be understood that the online generative adversarial network is a generative adversarial network that can generate data in real time. It can capture and analyze market data in real time, including social media popularity, advertising platform click-through rate (CTR) and competitor dynamics.

[0097] Specifically, the generator G of the online generative adversarial network online According to the random noise z online Capture market data in real time and generate simulated market data samples sim =G online (z online ), where z online Is the random noise of the online generative adversarial network. Discriminator D online Used to distinguish generated simulation data from real market data realThrough adversarial training, the generator continuously learns the distribution of real market data, making the generated simulated data closer and closer to the real data. online The loss function is:

[0098]

[0099] The loss function of the generator is:

[0100]

[0101] Loss function of the discriminator This function measures the discriminator's ability to classify real data from generated data. It aims to output a higher probability for real data and a lower probability for generated data. The generator's loss function aims to deceive the discriminator into believing the generated data is real.

[0102] The generator G of the trained online generative adversarial network online Can generate simulation data in real time sim Based on current market conditions and learning from historical data, the generator can capture changes in market trends and generate simulated data that aligns with them. For example, if a topic is currently gaining popularity on social media, the generator might generate simulated data related to that topic, such as more user discussions and higher engagement. This simulated data reflects the underlying trends in the current market, and companies can use it for market forecasting and strategic planning, preparing for countermeasures in advance.

[0103] Then, the Bayesian optimization algorithm is used to dynamically adjust the marketing strategy parameters based on the simulation data and historical observation results to optimize the long-term effect. Specifically, assuming that the marketing strategy parameters to be optimized are θ=(θ1,θ2,…,θ l ), the historical observation result is y obs The Bayesian optimization algorithm estimates the relationship between the objective function f(θ) (such as long-term marketing effect) and the parameter θ by constructing a probabilistic proxy model (such as a Gaussian process model). In each iteration, the next most promising parameter combination θ is selected based on the probabilistic proxy model. new Evaluate and the evaluation result is y new The probabilistic proxy model is then updated based on the evaluation results and the iterative process is repeated. The objective function f(θ) can be defined as:

[0104] f(θ)=αE sim (θ)+(1-α)E obs (θ),

[0105] Among them, E sim (θ) is the marketing effect predicted based on the simulated data, Eobs (θ) is the marketing effect predicted based on historical observations, and α is a weighting factor used to balance the impact of simulated data and historical observations. For example, if a company focuses more on real-time market changes, α can be set larger; if a company trusts the reliability of historical data, α can be set smaller.

[0106] Through continuous iterative optimization, the Bayesian optimization algorithm can dynamically adjust marketing strategy parameters based on simulation data and historical observation results to optimize the objective function (such as long-term marketing effect).

[0107] This embodiment utilizes an online generative adversarial network to capture and analyze market data in real time. Through adversarial training of the generator and discriminator, simulated data that aligns with current market trends is generated, helping companies stay abreast of market dynamics. This simulated data provides data support for the Bayesian optimization algorithm that aligns with current market trends, enabling the optimization process to account for real-time market changes. The Bayesian optimization algorithm also dynamically adjusts marketing strategy parameters based on simulated data and historical observations, optimizing long-term results and improving the adaptability and effectiveness of marketing campaigns.

[0108] Further optionally, the step of dynamically adjusting marketing strategy parameters using a Bayesian optimization algorithm based on simulation data and historical observation results includes:

[0109] Define the marketing strategy parameters to be optimized;

[0110] The expected effect of the marketing strategy predicted by the causal forest model is used as the objective function of Bayesian optimization;

[0111] The parameters are iteratively adjusted using the Bayesian optimization algorithm until the objective function converges.

[0112] It's understandable that when using the Bayesian optimization algorithm to dynamically adjust marketing strategy parameters, the first step is to define the marketing strategy parameters to be optimized. These parameters can include budget allocation ratio, number of KOL collaborations, and other parameters. The budget allocation ratio can refer to the proportion of funds allocated between different marketing channels; the number of KOL collaborations can refer to the number of collaborations with key opinion leaders of varying influence. For example, a company can collaborate with 5 top KOLs, 10 mid-range KOLs, and 20 bottom-tier KOLs, resulting in a total of 35 KOL collaborations. Other parameters can include advertising timing, promotional activity intensity, product pricing, and more.

[0113] The expected effect of the marketing strategy, as predicted by the causal forest model, serves as the objective function for Bayesian optimization. The causal forest model can predict the expected effect of a marketing strategy based on the input marketing strategy parameters θ, such as the corresponding return on investment (ROI) or conversion rate (ConversionRate). Assuming the predicted return on investment for the marketing strategy parameters θ is ROI(θ), the objective function for Bayesian optimization can be defined as: f(θ) = ROI(θ). Alternatively, if the conversion rate is of greater interest, the objective function can be defined as: f(θ) = ConversionRate(θ).

[0114] The parameters are then iteratively adjusted using the Bayesian optimization algorithm until the objective function converges. The Bayesian optimization algorithm estimates the relationship between the objective function f(θ) and the parameter combination θ by constructing a probabilistic proxy model (such as a Gaussian process model). In each iteration, the next new parameter combination θ is selected based on the acquisition function (such as the expected improvement function EI(θ)). new Evaluate. The expected improvement function is defined as:

[0115] EI(θ)=E[max(f(θ)-f best,0 )],

[0116] Among them, f best,0 is the currently known best objective function value. This expected improvement function measures the objective function value relative to the currently known best objective function value f after selecting the parameter combination θ. best,0 The algorithm selects the parameter combination with the largest expected improvement function value as the next evaluation point.

[0117] Then calculate the new parameter combination θ new The objective function value f(θ new ). And update the prior distribution of the probability proxy model (such as the Gaussian process model) of the Bayesian optimization model according to the new evaluation data, including the mean function and covariance function.

[0118] Repeat the above process to continuously generate new parameter combinations for evaluation and update the Bayesian optimization model. In each iteration, the current optimal objective function value is compared with the optimal objective function value of the previous iteration. If the change is less than the preset threshold ∈, the iteration is stopped.

[0119] This embodiment determines an objective function, such as return on investment or conversion rate, based on the expected effects of marketing strategies predicted by the causal forest model. This provides a clear optimization direction for the Bayesian optimization algorithm. The Bayesian optimization algorithm then iteratively adjusts parameters to find the optimal parameter combination, optimizing the objective function and thus improving the effectiveness of marketing activities.

[0120] Further optionally, the step of iteratively adjusting parameters by using a Bayesian optimization algorithm until the objective function reaches convergence includes:

[0121] Initialize the parameter combination and calculate the corresponding objective function value;

[0122] Generate the next set of new parameter combinations based on the recommendations of the Bayesian optimization algorithm;

[0123] Calculate the objective function value for the new parameter combination and update the prior distribution of the Bayesian optimization model;

[0124] The iterative process is repeated until the change in the objective function value is less than the preset threshold or the maximum number of iterations is reached.

[0125] It is understandable that in the process of iteratively adjusting parameters through the Bayesian optimization algorithm, first, according to the range of marketing strategy parameters to be optimized, the initial parameter combination θ1, θ2, ..., θ n , where n is the number of initial parameter combinations. For each initial parameter combination, the target function value f(θ i ) for calculation.

[0126] Then, based on the recommendation of the Bayesian optimization algorithm, the next set of new parameter combinations is generated. Specifically, the Bayesian optimization algorithm constructs a probabilistic proxy model (such as a Gaussian process model) based on the currently evaluated parameter combinations and the corresponding objective function values. This model can estimate the distribution of the objective function in the entire parameter space. And use the acquisition function (such as the expected improvement function) to select the next most promising parameter combination θ in the parameter space. new The expected improvement function takes into account the uncertainty of the parameter combination and the potential improvement of the objective function, and selects the new parameter combination that can maximize the objective function value (that is, the parameter combination with the largest EI(θ)) as the next evaluation point.

[0127] Next, the newly generated parameter combination is input into the causal forest model to calculate its corresponding objective function value. Based on the new evaluation data (the new parameter combination and its objective function value), the prior distribution of the probabilistic surrogate model of the Bayesian optimization model is updated. The prior distribution includes the mean function and covariance function. The updated prior distribution can more accurately reflect the relationship between the objective function and the parameters.

[0128] The iterative process is repeated, continuously generating new parameter combinations for evaluation and updating the Bayesian optimization model. In each iteration, the algorithm selects the next parameter combination based on the current prior distribution, calculates its objective function value, and updates the prior distribution. Furthermore, in each iteration, the algorithm compares the current optimal objective function value with the optimal objective function value from the previous iteration and calculates the magnitude of the change. If the magnitude of the change is less than a preset threshold ∈, the iteration is terminated.

[0129] For the generated multiple marketing strategies, they can be sorted according to the effect indicators predicted by the causal forest model. Assume that the generated marketing strategy set is S = {s1, s2, ..., s k}, each strategy s i Corresponding to a prediction effect index f(s i ). According to f(s i ) Sort the strategies from largest to smallest, and select the top-ranked strategy that meets the enterprise requirements as the optimal strategy for implementation.

[0130] This embodiment initializes parameter combinations, providing an initial evaluation point for the Bayesian optimization algorithm, enabling it to begin building a probabilistic proxy model. Based on the recommendations of the acquisition function, it selects more promising parameter combinations for evaluation, improving optimization efficiency. The new evaluation data is then used to update the prior distribution of the Bayesian optimization model, enabling the model to more accurately estimate the relationship between the objective function and the parameters.

[0131] In summary, the planning and effect evaluation method for enterprise marketing activities of the present invention first converts enterprise goals and constraints into marketing strategy vectors based on conditional generative adversarial networks, thereby efficiently generating marketing strategies that meet the constraints and are feasible, significantly improving the efficiency of strategy formulation. Secondly, the causal forest model is used to conduct causal effect analysis on the generated marketing strategies, and the expected effects of the marketing strategies are accurately evaluated by quantifying the causal effects of each strategy variable; at the same time, simulation scenarios are generated in combination with counterfactual reasoning to test the robustness of the strategies, which can accurately measure the effectiveness of the strategies and their stability, effectively reducing the risks and costs brought about by blindly implementing strategies. Finally, an online generative adversarial network is used to capture and analyze market data in real time to generate simulation data, and then the Bayesian optimization algorithm is used to dynamically adjust the marketing strategy parameters, so that the strategies can dynamically adapt to market changes and optimize long-term marketing effects.

[0132] Example 2

[0133] See also Figure 2 The present invention proposes a system for planning and evaluating the effects of corporate marketing activities, the system comprising:

[0134] Strategy Generation Module: Used to generate marketing strategies based on the enterprise goals and constraints based on the conditional generative adversarial network;

[0135] Strategy effect evaluation module: used to analyze the causal effects of the generated marketing strategies through the causal forest model and evaluate the robustness of the strategies in combination with counterfactual reasoning;

[0136] Strategy parameter adjustment module: used to use online generative adversarial networks to capture market trends in real time and dynamically adjust marketing strategy parameters in combination with Bayesian optimization algorithms.

[0137] Further optionally, the strategy generation module is further configured to:

[0138] Define enterprise goals and constraints, and convert them into conditional vectors to be input into the generator of the conditional generative adversarial network;

[0139] The generator of the generative adversarial network generates a marketing strategy vector based on the input condition vector and random noise, including advertising creativity, channel combination, promotion intensity and delivery rhythm;

[0140] The discriminator of the generative adversarial network evaluates the feasibility of the generated marketing strategy and feeds back the gradient to update the generator until a marketing strategy that meets the constraints and has high feasibility is generated.

[0141] Further optionally, the strategy effect evaluation module is further configured to:

[0142] Use the causal forest model to train historical marketing strategy data to quantify the causal effect of each strategy variable;

[0143] Evaluate the expected effectiveness of the generated marketing strategies based on the output of the causal forest model;

[0144] Combine counterfactual reasoning to generate simulation scenarios and test the performance of marketing strategies under different market conditions.

[0145] Further optionally, the strategy effect evaluation module is further configured to:

[0146] Collect historical marketing strategy data, including strategy variables, market characteristics, and user behavior data;

[0147] Use the causal forest model to train historical marketing strategy data to learn the impact of each strategy variable on marketing effectiveness;

[0148] Output the estimated causal effect of each strategy variable.

[0149] Further optionally, the policy parameter adjustment module is further configured to:

[0150] Use online generative adversarial networks to capture and analyze market data in real time, including social media popularity, advertising platform CTR, and competitor dynamics;

[0151] Generate simulated data that matches current market trends based on the output of an online generative adversarial network;

[0152] Use Bayesian optimization algorithms to dynamically adjust marketing strategy parameters based on simulated data and historical observations.

[0153] Further optionally, the policy parameter adjustment module is further configured to:

[0154] Define the marketing strategy parameters to be optimized;

[0155] The expected effect of the marketing strategy predicted by the causal forest model is used as the objective function of Bayesian optimization;

[0156] The parameters are iteratively adjusted using the Bayesian optimization algorithm until the objective function converges.

[0157] Further optionally, the policy parameter adjustment module is further configured to:

[0158] Initialize the parameter combination and calculate the corresponding objective function value;

[0159] Generate the next set of new parameter combinations based on the recommendations of the Bayesian optimization algorithm;

[0160] Calculate the objective function value for the new parameter combination and update the prior distribution of the Bayesian optimization model;

[0161] The iterative process is repeated until the change in the objective function value is less than the preset threshold or the maximum number of iterations is reached.

[0162] The above-described embodiments merely illustrate several implementations of the present invention, and while their descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

Claims

1. A method for planning and evaluating the effects of corporate marketing activities, characterized in that: The method comprises: Generate marketing strategies based on enterprise goals and constraints based on conditional generative adversarial networks; The causal effect analysis of the generated marketing strategy is carried out through the causal forest model, and the robustness of the strategy is evaluated by combining counterfactual reasoning; Use online generative adversarial networks to capture market trends in real time, and combine them with Bayesian optimization algorithms to dynamically adjust marketing strategy parameters.

2. The method for planning and evaluating the effects of corporate marketing activities according to claim 1, characterized in that: The steps of generating a marketing strategy based on a conditional generative adversarial network according to enterprise goals and constraints include: Define enterprise goals and constraints, and convert them into conditional vectors to be input into the generator of the conditional generative adversarial network; The generator of the generative adversarial network generates a marketing strategy vector based on the input condition vector and random noise, including advertising creativity, channel combination, promotion intensity and delivery rhythm; The discriminator of the generative adversarial network evaluates the feasibility of the generated marketing strategy and feeds back the gradient to update the generator until a marketing strategy that meets the constraints and has high feasibility is generated.

3. The method for planning and evaluating the effects of corporate marketing activities according to claim 1, characterized in that: The steps of performing causal effect analysis on the generated marketing strategy using the causal forest model and evaluating the robustness of the strategy using counterfactual reasoning include: Use the causal forest model to train historical marketing strategy data to quantify the causal effect of each strategy variable; Evaluate the expected effectiveness of the generated marketing strategies based on the output of the causal forest model; Combine counterfactual reasoning to generate simulation scenarios and test the performance of marketing strategies under different market conditions.

4. The method for planning and evaluating the effects of corporate marketing activities according to claim 3, characterized in that: The steps of using the causal forest model to train historical marketing strategy data to quantify the causal effect of each strategy variable include: Collect historical marketing strategy data, including strategy variables, market characteristics, and user behavior data; Use the causal forest model to train historical marketing strategy data to learn the impact of each strategy variable on marketing effectiveness; Output the estimated causal effect of each strategy variable.

5. The method for planning and evaluating the effects of corporate marketing activities according to claim 1, characterized in that: The steps of using an online generative adversarial network to capture market trends in real time and dynamically adjusting marketing strategy parameters in combination with a Bayesian optimization algorithm include: Use online generative adversarial networks to capture and analyze market data in real time, including social media popularity, advertising platform CTR, and competitor dynamics; Generate simulated data that matches current market trends based on the output of an online generative adversarial network; Use Bayesian optimization algorithms to dynamically adjust marketing strategy parameters based on simulated data and historical observations.

6. The method for planning and evaluating the effects of corporate marketing activities according to claim 5, characterized in that: The step of dynamically adjusting marketing strategy parameters using the Bayesian optimization algorithm based on simulation data and historical observation results includes: Define the marketing strategy parameters to be optimized; The expected effect of the marketing strategy predicted by the causal forest model is used as the objective function of Bayesian optimization; The parameters are iteratively adjusted using the Bayesian optimization algorithm until the objective function converges.

7. The method for planning and evaluating the effects of corporate marketing activities according to claim 6, characterized in that: The step of iteratively adjusting parameters by using the Bayesian optimization algorithm until the objective function reaches convergence includes: Initialize the parameter combination and calculate the corresponding objective function value; Generate the next set of new parameter combinations based on the recommendations of the Bayesian optimization algorithm; Calculate the objective function value for the new parameter combination and update the prior distribution of the Bayesian optimization model; The iterative process is repeated until the change in the objective function value is less than the preset threshold or the maximum number of iterations is reached.

8. A system for planning and evaluating the effects of corporate marketing activities, used to implement the method for planning and evaluating the effects of corporate marketing activities according to any one of claims 1 to 7, characterized in that: The system comprises: Strategy Generation Module: Used to generate marketing strategies based on the enterprise goals and constraints based on the conditional generative adversarial network; Strategy effect evaluation module: used to analyze the causal effects of the generated marketing strategies through the causal forest model and evaluate the robustness of the strategies in combination with counterfactual reasoning; Strategy parameter adjustment module: used to use online generative adversarial networks to capture market trends in real time and dynamically adjust marketing strategy parameters in combination with Bayesian optimization algorithms.

Images