Search

CN-121998678-A - Marketing scene dynamic simulation and strategy optimization system

CN121998678ACN 121998678 ACN121998678 ACN 121998678ACN-121998678-A

Abstract

The invention relates to the technical field of management systems, and particularly discloses a marketing scene dynamic simulation and strategy optimization system which comprises a multi-source data fusion module, a dynamic scene simulation engine, a multi-agent strategy game module and a strategy iteration optimization and deployment module which are sequentially cooperated to realize the generation and execution of an optimal marketing strategy perceived from a market environment. According to the invention, through a microscopic consumer heterogeneity model and combining a deep survival analysis and a attention mechanism, decision rules of individual consumers and differentiated responses to marketing stimulation can be accurately captured, the limitation of macroscopic analysis of a traditional system is broken through, full-dimensional conversion prediction from individuals to groups is realized, a refined data support is provided for strategy formulation, and the marketing conversion efficiency is effectively improved.

Inventors

  • ZHANG XIAOYU

Assignees

  • 沈阳曼得科技有限公司

Dates

Publication Date
20260508
Application Date
20260118

Claims (10)

  1. 1. The marketing scene dynamic simulation and strategy optimization system is characterized by comprising a multi-source data fusion module, a dynamic scene simulation engine, a multi-agent strategy game module and a strategy iteration optimization and deployment module which are sequentially cooperated to realize closed loop from market environment perception to optimal marketing strategy generation and execution; the multi-source data fusion module is configured to collect and structuralized multi-dimensional data from markets, consumers, competing brands and channels in real time to generate a unified dynamic environment feature vector set; The dynamic scene simulation engine receives the dynamic environment feature vector set and generates an interactive parameterized dynamic virtual market environment based on a macroscopic market dynamics model and a microscopic consumer heterogeneity model; The central game coordinator is configured to schedule the intelligent agent of the my marketing strategy, at least one intelligent agent of the competitor behavior simulation and the intelligent agent of the consumer group simulation to carry out dynamic game deduction, and generate a plurality of candidate strategy sequences of the intelligent agent of the my and simulation effect evaluation thereof; The strategy iterative optimization and deployment module receives the candidate strategy sequences and the simulation effect evaluation thereof, performs strategy screening and parameter optimization based on a multi-objective optimization algorithm and a Bayesian optimization framework, outputs a pareto optimal strategy set, and deploys the highest priority strategy to a real marketing channel for execution.
  2. 2. The marketing scene dynamic simulation and strategy optimization system of claim 1, wherein the microscopic consumer heterogeneity model in the dynamic scene simulation engine adopts an individual decision model based on deep survival analysis and attention mechanism for predicting the conversion probability and conversion time of single consumer under marketing stimulus, and the microscopic consumer heterogeneity model is used for the consumer At the moment of Subject to marketing strategies Transformation risk function of conditions after stimulation The definition is as follows: ; Wherein, the Fitting natural conversion trend of consumers without marketing stimulus by using Weibull distribution as a reference risk function; Is a consumer The dimension is 128-256 dimensions; Is a consumer By the time The interaction history sequence covers browsing, clicking, purchasing, complaint related behavior records and corresponding time stamps; for the attention network, a 3-layer transducer encoder structure is adopted, and the historical interaction behavior and the current strategy are calculated The semantic association degree of (1) outputs a weight value, and the value range is [0,1]; the method adopts a piecewise function form to capture the attenuation or enhancement effect of a marketing strategy along with time as a strategy time-varying effect function, and the formula is Wherein As a result of the initial coefficient of effect, In order to achieve a rate of decay, For the policy critical validation time point, Adjusting coefficients for post critical effects, consumer in time window Probability of transformation within Derived from cumulative risk functions, in particular 。
  3. 3. The marketing scenario dynamic simulation and strategy optimization system of claim 2, wherein the consumer population simulation agent is composed of agent populations driven by the microscopic consumer heterogeneity model, the agent numbers are configured according to a 1:100 ratio of the real consumer scale of the target market, and the feature distribution of the agent populations is ensured to be consistent with the real market through hierarchical sampling, the central game coordinator calculates the my strategy agent adoption strategy by aggregating individual decisions in deduction After that, in the simulation period Group-level key performance indicators within, including total conversion costs New number of clients Customer lifecycle value variation ; Wherein the number of newly added clients is net Is calculated to introduce a competitive loss factor The formula is: ; for a set of potential customers to be targeted, A set of existing clients for my; For my existing customers Contemporaneous exposure to competitor policies The loss probability under the influence is obtained by mapping the competitive strength output by the competitive behavior simulation agent, the mapping relation is fitted by a logistic regression model, and the input is competitive strategy strength, customer loyalty grade and product substitution degree; for loss coefficient, the resource competition relationship between new customer and reserved old customer is obtained, its value is dynamically regulated by enterprise resource allocation proportion, when the resource is inclined to new customer Taking 0.3-0.5, and taking 0.6-0.8 when tilting to the elderly.
  4. 4. The marketing scenario dynamic simulation and strategy optimization system of claim 3, wherein the competitor behavior simulation agent adopts a deep reinforcement learning-based antagonism strategy generation network, the network structure comprises 6 layers of full-connection layers and 2 layers of LSTM layers, the input is a market state feature vector and my history strategy sequence, the output is a probabilistic competition strategy distribution, and the objective function thereof Defined as minimizing the core revenue targets of the my policy agent while maximizing its own revenue in the simulated environment: ; Wherein, the Network parameters of the intelligent agent; Is that The market state of moment is provided by the dynamic scene simulation engine; 、 respectively being a competitor agent and a my agent Action taken at the moment; The method is characterized in that the method is to calculate the real-time benefits of competitors in such a way that sales are subtracted by marketing cost and operation cost; for the benefit change of the intelligent agent before and after competition, i.e , Indicating no competing actions; In order to resist the intensity coefficient, the value range is [0.5,1.5], which can be adjusted according to the competition intensity of the industry; As a result of the non-linear scaling factor, Default value is 1.8, which is used for amplifying punishment of my high profit loss; the competitor behavior simulation agent can dynamically generate a competitive competition strategy aiming at the my history strategy mode through simulation learning, an experience playback mechanism is adopted to update the strategy synchronously with the target network in the learning process, the capacity of an experience playback buffer zone is set to 100000, and the target network synchronizes the main network parameters once every 100 steps.
  5. 5. The marketing scenario dynamics simulation and strategy optimization system of claim 4, wherein the central game coordinator of the multi-agent strategy game module calculates strategy game balancing indexes after each round of deduction is completed The method is used for quantifying the stability of the current simulation market state and the degree of interaction of each agent strategy: ; Wherein, the Collecting all agents participating in the game; Is an intelligent body Normalized utility values in the present round of gaming are obtained by mapping the original utility values to the [0,1] intervals, the utility values of my and competitors are calculated based on the benefits, and the utility values of the consumer group agents are calculated based on the satisfaction score; For average utility, i.e ; The KL divergence mean value of each agent strategy distribution and the previous strategy distribution of the current round reflects the strategy fluctuation range; The value range is [0.8,1.2] for the attenuation coefficient, and the default value is 1.0; At the level of the minimum value of the total number of the components, For avoiding zero denominator; When (when) Above a threshold value When the coordinator judges that the game tends to Nash equilibrium, stops the current branch deduction, records the strategy and market result of each party under the state, if 3 rounds of deduction are continuous Are all lower than And (5) triggering a strategy diversity enhancing mechanism, and adding random disturbance items to each agent strategy space.
  6. 6. The marketing scenario dynamic simulation and strategy optimization system of claim 5, wherein the strategy iterates a multi-objective optimization algorithm in the optimization and deployment module to maximize simulated net revenue Maximizing market share growth rate Minimizing policy fluctuation risk Constructing a pareto front for the target; Wherein the net benefit is simulated Synthesized from population-level indicators: ; Wherein, the Is a weight coefficient, satisfies Default values are respectively 0.6, 0.3 and 0.1, and can be adjusted according to enterprise strategic targets; Is a strategy The average life cycle value change of the single customer is calculated by predicting the consumption amount, the frequency of repurchase and the retention rate of 36 months in the future of the customer; the existing client base for my; Is a strategy Is added to the total execution cost of (a); And calculating a punishment value of 0.05-0.2 corresponding to each violation or infeasible item according to a preset business rule base for policy compliance and feasibility punishment items, and directly eliminating the policy when the accumulated punishment value exceeds 0.5.
  7. 7. The marketing scenario dynamic simulation and strategy optimization system of claim 6, wherein the strategy iterative optimization and deployment module further comprises a strategy robustness assessment unit for performing pressure testing on each strategy in the pareto optimal strategy set by introducing random disturbance variables into the dynamic scenario simulation engine The random disturbance variable covers market demand fluctuation, raw material price mutation and policy supervision adjustment related types, the occurrence probability of each disturbance type is statistically set according to historical data, limited round game is re-executed, and the robustness score of policy performance is calculated : ; To simulate net gain on average in the presence of disturbances, i.e , Is the first Disturbance variable of the wheel; Is the reference income without disturbance; To the standard deviation coefficient of the gain in the presence of disturbances, i.e ; System priority deployment Highest scoring policies, when multiple policies When the score difference is smaller than 0.03, the execution cost of the strategies is further compared, and the strategy with lower cost is selected.
  8. 8. The marketing scenario dynamic simulation and strategy optimization system of claim 7, wherein after the marketing scenario dynamic simulation and strategy optimization system is deployed, an online learning and feedback closed loop is provided The generated effect data is processed by a multi-source data fusion module and is compared with a simulation predicted value to generate a simulation fidelity error : ; Wherein, the Is the number of key performance indicators; and (3) with The real value and the analog value are respectively; At the level of the minimum value of the total number of the components, For avoiding zero denominator; When (when) Triggering fine adjustment of relevant models in a dynamic scene simulation engine and a multi-agent strategy game module when 3 evaluation periods are higher than a threshold value, wherein the fine adjustment adopts an incremental training mode, the difference between real effect data and simulation data is taken as a loss signal, an optimizer adopts AdamW, and the learning rate is that And simultaneously, supplementing new market phenomena and competition behavior patterns in a real scene to a scene library and a strategy library of the system, and continuously improving the reality and the accuracy of the simulation.
  9. 9. The marketing scene dynamic simulation and strategy optimization system according to claim 8, wherein the multi-source data fusion module integrates an unstructured data understanding unit, analyzes social media graphics, short video content and customer service dialogue recordings by using a multi-mode large model, extracts public emotion tendencies, popular topic heat and competitor new features, and quantifies the features into feature vectors which can be input into a dynamic scene simulation engine; the specific process flow is as follows: s.11, preprocessing data, namely performing format standardization, noise removal and fragment segmentation processing on unstructured data, uniformly scaling image data to 224 multiplied by 224 pixels, converting audio data into a 16kHz mono WAV format, and performing word segmentation and stop word removal on text data; s.12, extracting characteristics of product appearance and scene elements in the image by adopting a visual encoder, extracting characteristics of voice emotion and keywords by adopting an audio encoder, extracting characteristics of semantic information and emotion tendency by adopting a text encoder, wherein the dimension of each modal characteristic vector is 768 dimensions; S.13, cross-modal fusion, namely fusing multi-modal features through an attention mechanism, calculating the association weight of each modal feature and a marketing scene, and generating 256-dimensional unified feature vectors after weighted fusion; S.14, quantitatively mapping, namely mapping the public emotion tendency into an emotion index of a [ -1,1] interval, mapping the popularity topic heat into a heat index of a [0,1] interval, converting the new product characteristics of competitors into characteristic vectors combining Boolean type and numerical type, and finally integrating the characteristic vectors into a component part of a dynamic environment characteristic vector set.
  10. 10. The marketing scenario dynamic simulation and strategy optimization system of claim 9, wherein the marketing scenario dynamic simulation and strategy optimization system provides a strategy explanatory report generating function, and for the outputted optimal strategy, the system can retrospectively simulate the deduction process, locate key game turning points which lead to strategy success, the consumer subdivision groups with core influence and the main coping modes of competitors, and present in the form of visual narrative report; The specific implementation flow is as follows: S.21, positioning key nodes, namely identifying the turn of KPI mutation as a key game turning point by analyzing KPI change curves of each turn in the strategy game process, and extracting strategy actions and market state changes of each party of the turn; S.22, consumer group analysis, namely subdividing consumer agent groups by adopting a clustering algorithm, calculating conversion contribution degree of each subdivision group to an optimal strategy, screening groups with contribution degree Top3 as core influence groups, and outputting characteristic images and behavior patterns of the core influence groups; S.23, analyzing the behaviors of the competitors, namely counting the strategy selection frequency and the response delay time of the competitors in the game process, summarizing the main response modes of the competitors, and analyzing the resisting effect of the optimal strategy on each response mode; And S.24, generating a report, namely integrating the analysis result with the specific content, the expected effect and the execution suggestion of the strategy, and generating a visual report comprising a text description, a data chart and a flow diagram.

Description

Marketing scene dynamic simulation and strategy optimization system Technical Field The invention relates to the technical field of management systems, in particular to a marketing scene dynamic simulation and strategy optimization system. Background In the field of digital marketing, the market environment presents the remarkable characteristics of dynamics, complexity and multi-main game, the consumer demand heterogeneity is strong, the competitor strategy is changeable, the multi-source data form is multi-source, and the factors have extremely high requirements on the accuracy and the adaptability of the marketing strategy. The traditional marketing mode relies on experience judgment or static data analysis to generate a strategy, lacks the capability of real-time perception and simulation on dynamic changes of the market, is difficult to accurately capture the association rule between marketing stimulus and consumer behavior, and causes the problems of blindness, easy resource waste, low conversion efficiency and the like in strategy formulation; The prior marketing simulation and optimization technology tries to introduce a modeling method, but has a plurality of limitations that firstly, most systems focus on macroscopic market trend analysis, neglect heterogeneity of individual decisions of consumers at a microscopic level and cannot accurately predict differentiated responses of different consumers to marketing stimulation; secondly, the effective simulation of the multi-main-body dynamic game process is lacking, the influence of a competitor strategy on the marketing effect of the my is difficult to quantify, so that the strategy has insufficient anti-competition capability, thirdly, the data fusion capability is weak, the excavation and the utilization of unstructured data such as social media graphics and texts, customer dialogue recordings and the like are insufficient, the market environment characteristics cannot be comprehensively depicted; In addition, the prior art has insufficient attention to the interpretation and robustness of the strategy, the optimized strategy is difficult to trace the effective logic of the strategy, and is easy to fail when facing sudden market fluctuation, and the core requirements of enterprises on the scientificity, reliability and interpretability of the marketing strategy cannot be met. Therefore, the research and development of an integrated system capable of realizing multi-source data depth fusion, dynamic scene accurate simulation, multi-main-body game deduction and strategy iterative optimization becomes a technical problem to be solved in the current marketing technical field. Disclosure of Invention Aiming at the defects of the prior art, the invention provides a marketing scene dynamic simulation and strategy optimization system, which solves the problems in the background art. The marketing scene dynamic simulation and strategy optimization system comprises a multi-source data fusion module, a dynamic scene simulation engine, a multi-agent strategy game module and a strategy iteration optimization and deployment module which are sequentially cooperated to realize the generation and execution of the optimal marketing strategy perceived from the market environment; the multi-source data fusion module is configured to collect and structuralized multi-dimensional data from markets, consumers, competing brands and channels in real time to generate a unified dynamic environment feature vector set; The dynamic scene simulation engine receives the dynamic environment feature vector set and generates an interactive parameterized dynamic virtual market environment based on a macroscopic market dynamics model and a microscopic consumer heterogeneity model; The central game coordinator is configured to schedule the intelligent agent of the my marketing strategy, at least one intelligent agent of the competitor behavior simulation and the intelligent agent of the consumer group simulation to carry out dynamic game deduction, and generate a plurality of candidate strategy sequences of the intelligent agent of the my and simulation effect evaluation thereof; The strategy iterative optimization and deployment module receives the candidate strategy sequences and the simulation effect evaluation thereof, performs strategy screening and parameter optimization based on a multi-objective optimization algorithm and a Bayesian optimization framework, outputs a pareto optimal strategy set, and deploys the highest priority strategy to a real marketing channel for execution. Preferably, the microcosmic consumer heterogeneity model in the dynamic scene simulation engine adopts an individual decision model based on deep survival analysis and attention mechanism and is used for predicting the conversion probability and conversion time of single consumer under marketing stimulusAt the moment ofSubject to marketing strategiesTransformation risk function of conditions after stimulat