Marketing prediction intelligent analysis method and system based on machine learning

Abstract

The invention discloses a marketing prediction intelligent analysis method and system based on machine learning, and the method comprises the steps: generating a competition relation matrix between commodity categories through an ecological niche competition model, and achieving the strategy diversity maintenance through combining the strategy feature coding of a dynamic strategy pool with a red queen adjustment mechanism; and multi-strategy game simulation is carried out by using an ecological game engine, an optimal strategy combination is calculated and output through Nash equilibrium, and finally parameters are optimized based on a real-time feedback closed-loop link. According to the technical scheme of the invention, through a competition-game-feedback dynamic balance mechanism, autonomous emergence of a cross-category co-evolution strategy is realized, the environmental adaptability and evolution ability of a marketing strategy are significantly improved, the problem of strategy stiffness caused by manual rule dependence is overcome, and an intelligent decision-making system with an ecological self-organization characteristic is formed.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and particularly to an intelligent analysis method and system for marketing prediction based on machine learning. Background Art

[0002] At present, most marketing prediction technologies adopt static analysis models, which are difficult to dynamically capture the competitive evolution and symbiotic effects among product categories, and rely on artificial experience to preset marketing rules, resulting in insufficient strategy adaptability. In addition, the cross-category collaborative impact is often ignored when analyzing user behavior, and there is a lack of a diversity guarantee mechanism in the strategy generation process, making it easy to have the problem of homogenized strategy failure. At the same time, there is generally a lack of the ability to optimize strategies driven by real-time feedback, making it difficult to cope with the dynamic changes in the market competition environment, which limits the accuracy and timeliness of marketing decisions. Summary of the Invention

[0003] In view of the above problems, the present invention provides an intelligent analysis method and system for marketing prediction based on machine learning, which solves the problems of static prediction of isolated categories, serious strategy homogenization and inability to dynamically coordinate and optimize.

[0004] To achieve the above object, in the first aspect, the present application provides an intelligent analysis method for marketing prediction based on machine learning, including: Obtaining product information and user behavior data, where the product information includes category attributes, inventory status and competitor data, and the user behavior data includes browsing paths, purchase records and promotion responses; Inputting the product information and the user behavior data into a niche competition model for analyzing the category competition relationship, and generating a competition relationship matrix, where the competition relationship matrix includes substitution coefficients and symbiotic relationships among product categories; Constructing a dynamic strategy pool based on the user behavior data, dividing user groups into multiple strategy types through strategy feature encoding based on feature engineering, and real-time monitoring the strategy propagation information and fitness indicators of each strategy type; Deploying a Red Queen regulation mechanism in the dynamic strategy pool, detecting the degree of strategy homogenization through the Shannon entropy algorithm, and when the strategy diversity index is lower than a preset threshold, triggering an environmental pressure regulation module to generate strategy mutation instructions and environmental pressure parameters; Inputting the competition relationship matrix, the dynamic strategy pool and the environmental pressure parameters into an ecological game engine for multi-strategy game simulation, and outputting an optimal strategy combination through Nash equilibrium calculation; Establishing a real-time feedback closed-loop link, monitoring the execution effect of the optimal strategy combination through buried point data, generating feedback data including conversion rate changes and strategy penetration rates, and re-inputting the feedback data into the niche competition model and the dynamic strategy pool for parameter tuning.

[0005] In some embodiments, commodity information and user behavior data are input into a niche competition model for category competition relationship analysis, and a competition relationship matrix is generated, including: The commodity information and user behavior data are associated in multiple dimensions to generate a category association network, which includes a substitution relationship sub-network, a symbiotic relationship sub-network, and a competition relationship sub-network; According to the substitution relationship sub-network, calculate the real-time substitution coefficient corresponding to the category attribute; According to the symbiotic relationship sub-network, calculate the scenario symbiotic coefficient corresponding to the category attribute; Generate a competition intensity value corresponding to the category attribute based on the real-time substitution coefficient and the scenario symbiotic coefficient; Generate a competition relationship matrix based on the competition intensity value of the category attribute.

[0006] In some embodiments, calculating the real-time substitution coefficient corresponding to the category attribute according to the substitution relationship sub-network includes: Extract the cross-category browsing and jumping sequence set of the user group within a preset time window; Statistically analyze the group jump frequency matrix and the average residence time decay factor between commodity categories; Combined with the historical order substitution purchase rate of the user group, calculate the real-time substitution coefficient through a dynamic weighting formula, which is represented by formula (1), and formula (1) is as follows: ; In formula (1), is the historical order substitution purchase rate of the user group, is the first commodity category in the substitution relationship, is the second commodity category in the substitution relationship, is the commodity category to the commodity category is the normalized jump frequency, is the substitution purchase rate when the commodity category is out of stock and turns to the commodity category , is the residence time decay coefficient, is the dynamically adjusted weight, ; Calculating the scenario symbiotic coefficient corresponding to the category attribute according to the symbiotic relationship sub-network includes: Statistically analyze the multi-category combined purchase frequency in a single transaction of the user group; Analyze the high-frequency scenario association patterns in the group behavior path; Calculate the scenario symbiotic coefficient through the scenario symbiotic formula, which is represented by formula (2), and formula (2) is as follows: ; In formula (2), is the first product category in the symbiotic relationship, is the second product category in the symbiotic relationship, is the product category and the product category of the co-purchase probability, is the scenario association strength factor, is the scenario weight coefficient; Generating the competition intensity value corresponding to the category attribute according to the real-time substitution coefficient and the scenario symbiosis coefficient includes: Normalize the real-time substitution coefficient and the scenario symbiosis coefficient, and calculate the competition intensity value using a dynamic weight allocation strategy, which is represented by formula (3). Formula (3) is as follows: ; In formula (3), is the first product category in the final competition relationship, is the second product category in the final competition relationship, is the dynamic weight function related to time, is the competitor data correction term, all point to the product categories included in the product information in the same database, and , is the total number of product categories in the same database.

[0007] In some embodiments, the fitness indicators include conversion rate fitness, value fitness, and competition fitness. A dynamic policy pool is constructed based on user behavior data. The user group is divided into multiple policy types through policy feature encoding based on feature engineering, and the policy propagation information and fitness indicators of each policy type are monitored in real time, including: According to the user behavior data, the policy feature encoding extracts the user behavior pattern features through a temporal attention mechanism to generate user feature vectors; Adopt a dynamic threshold clustering algorithm to adjust the cluster boundary according to the competition relationship matrix, and perform clustering analysis on the user feature vectors to generate multiple user groups. Each user group corresponds to a policy type, and the policy types include price-sensitive, brand-loyal, and new product tasting; Initialize the dynamic policy pool according to the user group, user feature vector, and policy type; Statistically calculate the user proportion information of each policy type in the dynamic policy pool in real time, and calculate the policy propagation information. The policy propagation information includes the adoption rate of new user policy types, the migration rate of existing user policy types, and the policy diffusion correction value; Calculate the conversion rate fitness, value fitness, and competition fitness in the dynamic policy pool, including: Calculate the relative improvement value of the conversion rate for each user group in the current competitive environment, and perform a discount correction on the relative improvement value of the conversion rate according to the category competition intensity to obtain the conversion rate fitness; Statistically analyze the distribution of the unit price of the same user group. After removing the abnormal fluctuations caused by competition, calculate the premium ability index relative to the category benchmark value, and generate the value fitness; Map the category combinations mainly associated with the strategy types of the current user group, and extract the stability indicators of the corresponding categories in the competition relationship matrix to obtain the competition fitness.

[0008] In some embodiments, according to the user behavior data, the strategy feature encoding extracts the user behavior pattern features through the temporal attention mechanism, and generates a user feature vector including: Perform time series modeling on the browsing paths of users one by one, and capture the behavior dependence relationships at different time steps through the multi-head attention mechanism to obtain the transfer preference intensity of users between different categories; Calculate the response elasticity coefficient of the current user to the price range according to the purchase records, and perform weighted fusion by combining the response elasticity coefficient with the discount usage tendency in the promotion response to obtain the price sensitivity feature; Extract the brand switching frequency and the time delay of the first interaction with new products in the user behavior sequence, and construct a category composite index of brand loyalty and new product acceptance; After performing Min-Max normalization processing on the transfer preference intensity, the price sensitivity feature, and the category composite index, splice them into the user feature vector of the current user; Adopt a dynamic threshold clustering algorithm, adjust the cluster boundaries according to the competition relationship matrix, and perform clustering analysis on the user feature vector to generate multiple user groups including: Initialize the cluster centers, and initialize the cluster centers based on the category association degree in the competition relationship matrix; During the clustering iteration process, dynamically adjust the cluster boundary threshold according to the real-time competition intensity; Evaluate the clustering effect through the silhouette coefficient. When it is detected that the change in the category competition relationship leads to a decline in the clustering quality, automatically trigger the re-initialization of the clustering; Output the user groups divided by the strategy types with clear marketing semantics; Real-time statistically analyze the user proportion information of each strategy type in the dynamic strategy pool, and calculate the strategy propagation information including: Construct a strategy type transfer state matrix, record the migration frequencies of users between various strategy types within a fixed time window, and calculate the change gradient of the strategy penetration rate; Quantify the strategy adoption rate of the new user group, and calculate the strategy diffusion resistance coefficient by combining the competition intensity of the source categories of the new user group; Calculate the attenuation factor for the cross-category propagation of the strategy based on the category correlation in the competition relationship matrix, and correct the original propagation rate; Generate the adoption rate of the new user strategy type, the migration rate of the existing user strategy type, and the strategy diffusion correction value based on the change rate of the proportion of each strategy type in the current user group, the migration activity, and the impact of the competition environment, which is the strategy propagation information.

[0009] In some embodiments, deploying the Red Queen regulation mechanism in the dynamic strategy pool includes: Configure a Red Queen regulator in the dynamic strategy pool and establish the following detection mechanisms: Set a policy diversity warning threshold, which is dynamically adjusted according to the category competition intensity in the competition relationship matrix; Deploy a policy type distribution monitor to continuously track the real-time proportion fluctuation of the policy type; Install a policy ecosystem health diagnosis module to regularly output a homogenization risk assessment report.

[0010] In some embodiments, detecting the policy homogenization degree through the Shannon entropy algorithm includes: Real-time scan the user distribution of all policy types in the dynamic strategy pool, record the proportion data of each policy type in the dynamic strategy pool, and obtain the policy type proportion; Statistically analyze the proportion change trend of the policy type within a preset time window; According to the policy type proportion and the proportion change trend, calculate the policy diversity index through the Shannon entropy algorithm, including: Input the policy type proportion as the probability distribution to calculate the entropy value of the policy system at the current moment; Calculate the entropy change rate according to the proportion change trend to generate the policy diversity index; Judge the policy homogenization degree according to the policy diversification index and update the homogenization risk assessment report.

[0011] In some embodiments, when the policy diversity index is lower than the preset threshold, trigger the environmental pressure regulation module to generate a policy mutation instruction and environmental pressure parameters, including: Extract the list of high-competition intensity categories from the competition relationship matrix, denoted as high-competing product categories; Calculate the pressure application intensity of the high-competing product categories and calculate the pressure duration to obtain the category pressure parameters; Extract the target strategy types to be mutated according to the strategy diversity index, and generate a mutated strategy feature combination, which includes core features, non-core features, and new features. The core features are the core effective features of the target strategy type before mutation, the non-core features are the non-core effective features of the target strategy type after random mutation, and the new features are the new features extracted from the current competitive environment; Test the mutated strategy feature combination according to the category pressure parameter until the mutated strategy feature combination meets the preset test index, and record it as the final mutated strategy; Generate a strategy mutation instruction according to the final mutated strategy, and generate an environmental pressure parameter according to the category pressure parameter.

[0012] In some embodiments, input the competition relationship matrix, dynamic strategy pool, and environmental pressure parameter into the ecological game engine for multi-strategy game simulation, and output the optimal strategy combination through Nash equilibrium calculation, including: Map the strategy types in the dynamic strategy pool to game participants; Construct a strategy payoff matrix according to the competition relationship matrix, and use the environmental pressure parameter as a strategy constraint condition; In the strategy payoff matrix, calculate the expected payoff of the strategy types; Generate a strategy adjustment direction according to the expected payoff; Adjust the proportion of the strategy types according to the strategy adjustment direction until the Nash equilibrium condition is met, and generate the final strategy type distribution information. The Nash equilibrium condition includes that the adjustment amplitude of the strategy types is less than the preset convergence threshold, the rate of change of the expected payoff of the strategy types tends to be stable, and the maximum iteration times limit is reached; Generate an implementation plan for the optimal strategy combination according to the strategy type distribution information, including: Extract the proportion of the strategy types in the strategy type distribution information; Calculate the recommended implementation intensity of the strategy types in the strategy type distribution information according to the proportion of the strategy types, and generate a category-level strategy matching comparison table and a strategy adjustment priority list; Obtain the implementation plan for the optimal strategy combination.

[0013] In a second aspect, the present invention provides a marketing prediction intelligent analysis system based on machine learning, which is applicable to the marketing prediction intelligent analysis method based on machine learning in the first aspect. The system includes: A data acquisition unit for obtaining product information and user behavior data. The product information includes category attributes, inventory status, and competitor data, and the user behavior data includes browsing paths, purchase records, and promotion responses; A logical processing unit for inputting product information and user behavior data into a niche competition model for analyzing category competition relationships, generating a competition relationship matrix, where the competition relationship matrix includes substitution coefficients and symbiotic relationships between product categories; constructing a dynamic policy pool based on user behavior data, dividing user groups into multiple policy types through policy feature encoding based on feature engineering, and real-time monitoring the policy propagation information and fitness indicators of each policy type; deploying a Red Queen regulation mechanism in the dynamic policy pool, detecting the degree of policy homogenization through the Shannon entropy algorithm, and when the policy diversity index is lower than a preset threshold, triggering an environmental stress regulation module to generate policy mutation instructions and environmental stress parameters; inputting the competition relationship matrix, dynamic policy pool, and environmental stress parameters into an ecological game engine for multi-strategy game simulation, and outputting an optimal policy combination through Nash equilibrium calculation; A closed-loop feedback unit for establishing a real-time feedback closed-loop link, monitoring the execution effect of the optimal policy combination through buried-point data, generating feedback data including conversion rate changes and policy penetration rates, and re-inputting the feedback data into the niche competition model and dynamic policy pool for parameter tuning.

[0014] Different from the prior art, the above technical solution provides a marketing prediction intelligent analysis method and system based on machine learning. The method generates a competition relationship matrix between product categories through a niche competition model, realizes the maintenance of policy diversity by combining the policy feature encoding of the dynamic policy pool and the Red Queen regulation mechanism, and uses the ecological game engine for multi-strategy game simulation, outputs an optimal policy combination through Nash equilibrium calculation, and finally optimizes the parameters based on the real-time feedback closed-loop link. The above technical solution realizes the autonomous emergence of strategies for cross-category co-evolution through a dynamic balance mechanism of competition-game-feedback, significantly improves the environmental adaptability and evolution ability of marketing strategies, overcomes the problem of policy rigidity caused by relying on artificial rules, and forms an intelligent decision-making system with ecological self-organization characteristics.

[0015] The above relevant records of the invention content are only an overview of the technical solution of this application. In order to enable those of ordinary skill in the art to more clearly understand the technical solution of this application, and then can be implemented according to the content recorded in the description and the drawings, and in order to make the above objects, other objects, features and advantages of this application more easily understood, the following is described in conjunction with the specific implementation manners and drawings of this application. Brief Description of the Drawings

[0016] The drawings are only used to illustrate the principles, implementation methods, applications, features, and effects of the specific implementation manners of the present invention and other related contents, and should not be considered as a limitation to this application.

[0017] In the accompanying drawings of the specification: Figure 1It is a method step diagram of steps S101 to S106 of the intelligent analysis method described in the specific implementation manner; Figure 2 It is a method step diagram of steps S201 to S205 of the intelligent analysis method described in the specific implementation manner; Figure 3 It is a method step diagram of steps S301 to S305 of the intelligent analysis method described in the specific implementation manner; Figure 4 It is a method step diagram of steps S401 to S404 of the intelligent analysis method described in the specific implementation manner; Figure 5 It is a structural schematic diagram of the intelligent analysis system described in the specific implementation manner.

[0018] The descriptions of the reference numerals involved in the above-mentioned respective drawings are as follows: 1. Intelligent analysis system; 11. Data acquisition unit; 12. Logic processing unit; 13. Closed-loop feedback unit. Specific implementation manner

[0019] To describe in detail the possible application scenarios, technical principles, implementable specific solutions, achievable purposes and effects, etc. of the present application, the following is described in detail with reference to the listed specific embodiments and in conjunction with the drawings. The embodiments described herein are only used to more clearly illustrate the technical solutions of the present application, so they are only used as examples and cannot be used to limit the protection scope of the present application.

[0020] Referring to "embodiment" in this article means that the specific features, structures or characteristics described in conjunction with the embodiment may be included in at least one embodiment of the present application. The term "embodiment" that appears in various positions in the specification does not necessarily refer to the same embodiment, nor does it particularly limit its independence or relevance to other embodiments. In principle, in the present application, as long as there is no technical contradiction or conflict, the various technical features mentioned in each embodiment can be combined in any way to form the corresponding implementable technical solutions.

[0021] Unless otherwise defined, the meanings of the technical terms used in this article are the same as those generally understood by those skilled in the technical field to which the present application belongs; the use of the relevant terms in this article is only for describing specific embodiments and is not intended to limit the present application.

[0022] In the description of the present application, the term "and / or" is an expression used to describe the logical relationship between objects, indicating that there can be three relationships. For example, A and / or B means: there is A, there is B, and there is both A and B at the same time. In addition, the character " / " in this text generally represents an "or" logical relationship between the associated objects before and after.

[0023] In the present application, terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual quantitative, primary-secondary, or sequential relationships between these entities or operations.

[0024] Without further limitations, in the present application, the open-ended expressions such as "comprising", "including", "having", or other similar expressions used in a statement are intended to cover non-exclusive inclusion. These expressions do not exclude the possibility that there may be additional elements in the process, method, or product that includes the described elements. Thus, a process, method, or product that includes a series of elements may include not only those defined elements, but also other elements not explicitly listed, or elements inherent to such a process, method, or product.

[0025] Similar to the understanding in the "Examination Guidelines", in the present application, expressions such as "greater than", "less than", "exceeding", etc. are understood not to include the recited number; expressions such as "above", "below", "within", etc. are understood to include the recited number. In addition, in the description of the embodiments of the present application, the meaning of "a plurality of" is two or more (including two). Similar expressions related to "many", such as "multiple groups", "multiple times", etc., are understood in the same way, unless otherwise specifically defined.

[0026] In the description of the embodiments of the present application, the spatially related expressions used, such as "center", "longitudinal", "transverse", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "perpendicular", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", "axial", "radial", "circumferential", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the specific embodiment or the accompanying drawings. It is only for the convenience of describing the specific embodiments of the present application or facilitating the understanding of the reader, rather than indicating or implying that the device or component referred to must have a specific position, a specific orientation, or be constructed or operated in a specific orientation. Therefore, it should not be construed as a limitation to the embodiments of the present application.

[0027] The processor described in the embodiments of the present application can be implemented through hardware, firmware, software, or a combination thereof. It can use circuits, one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), central processing units (CPUs), controllers, microcontrollers, microprocessors, or at least one of the above, and also includes other physical, biological, or chemical structures that can achieve functions similar to or equivalent to those of the above-listed processors, such as biological neurons, quantum computing units, DNA computing units, etc., so that the processor can execute some steps, all steps, or any combination of the steps mentioned in the computer programs or methods involved in the various embodiments of the present application.

[0028] The computer programs involved in the embodiments can be stored in a computer device-readable storage medium, which includes but is not limited to magnetic disks, magnetic tapes, magnetic cards, floppy disks, flash memories, optical discs, optical cards, read-only memories (ROMs), random access memories (RAMs), erasable programmable ROMs (EPROMs), and electrically erasable programmable ROMs (EEPROMs), etc. It also includes other biological, physical, or chemical structures that can achieve the same or equivalent functions as the above-listed storage media, such as units with information storage capabilities like DNA, RNA, proteins, etc. In a specific embodiment, the storage medium involved can be one of the above medium types or a combination of the above medium types. In different embodiments, the computer programs involved in the embodiments can be stored centrally in a single medium or distributedly in multiple media. The memory containing the computer device-readable storage medium can be a non-volatile memory or a random access memory. These computer device-readable storage media can be built into the device or can be an external device or a part of an external device connected to the device involved in the embodiments. In some embodiments, the memory with the computer device-readable storage medium is deployed locally; in other embodiments, a scheme of deploying the memory away from the processor can also be adopted, such as a network-attached memory accessed via an RF circuit or an external port and a communication network, where the communication network can be the Internet, one or more intranets, local area networks (LANs), wide area wireless networks (WLANs), storage area networks (SANs), etc., or an appropriate combination thereof, as long as the computer device can access the memory. In addition, the computer programs involved in the embodiments can be stored in plaintext / ciphertext form or can be designed as training data and be integrated and recombinantly stored implicitly in the parameter states of deep neural networks or other machine learning models through model training.

[0029] Please refer to Figure 1 , in a first aspect, this embodiment provides an intelligent marketing prediction analysis method based on machine learning, including: S101. Obtain product information and user behavior data. The product information includes category attributes, inventory status, and competitor data, and the user behavior data includes browsing paths, purchase records, and promotion responses; S102. Input the product information and the user behavior data into a niche competition model for category competition relationship analysis to generate a competition relationship matrix, where the competition relationship matrix includes substitution coefficients and symbiotic relationships between product categories; S103. Build a dynamic policy pool based on the user behavior data, divide the user groups into multiple policy types through policy feature encoding based on feature engineering, and real-time monitor the policy propagation information and fitness indicators of each policy type; S104. Deploy the Red Queen regulation mechanism in the dynamic policy pool, detect the degree of policy homogenization through the Shannon entropy algorithm, and trigger the environmental pressure regulation module to generate policy mutation instructions and environmental pressure parameters when the policy diversity index is lower than the preset threshold; S105. Input the competition relationship matrix, dynamic policy pool, and environmental pressure parameters into the ecological game engine for multi-strategy game simulation, and output the optimal policy combination through Nash equilibrium calculation; S106. Establish a real-time feedback closed-loop link, monitor the execution effect of the optimal policy combination through buried-point data, generate feedback data including conversion rate changes and policy penetration rates, and re-enter the feedback data into the niche competition model and dynamic policy pool for parameter tuning.

[0030] In step S101, preferably, the category attribute refers to the hierarchical characteristics in the commodity classification system, including functional attributes, price band attributes, and scenario attributes, which are used to characterize the positioning basis of commodities in market competition; the inventory status includes real-time inventory quantity and replenishment cycle data, which are used to judge the impact of commodity sellability on user decisions; the competitor data covers price fluctuations and promotion strategy information of similar commodities on external platforms, which are used to identify cross-platform competition situations; the browsing path is constructed by recording the user access page jump sequence and stay duration, which is used to restore the attention distribution in the user decision-making process; the promotion response data includes user coupon redemption rate and discount sensitivity indicators, which are used to quantify the triggering efficiency of marketing strategies.

[0031] In step S102, the niche competition model refers to a calculation framework that simulates the resource competition relationship between commodity categories. The substitution coefficient is used to quantify the tendency of users to transfer across categories, and the symbiotic relationship represents the potential for multi-category collaborative sales. Preferably, the substitution coefficient is calculated through the user browsing jump rate and order substitution rate between categories, specifically by counting the frequency of users switching from one category to another and the actual substitution purchase ratio in out-of-stock scenarios; the symbiotic relationship is determined through the co-purchase probability and user behavior sequence matching degree. Among them, the co-purchase probability is calculated based on the co-occurrence frequency of multiple categories within the same order, and the user behavior sequence matching degree is evaluated through a path similarity algorithm to measure the association strength of cross-category browsing logic.

[0032] In step S103, preferably, the dynamic policy pool refers to a policy set library that is dynamically adjusted based on real-time user behavior. The policy feature encoding extracts key differential features of user behavior patterns through a temporal attention mechanism. The policy types include price-sensitive, brand-loyal, and new product tasting types, corresponding to the decision-making preferences of different user groups. Preferably, the policy dissemination information is calculated through the user policy type migration rate and the cross-category diffusion attenuation factor, reflecting the dissemination efficiency of the policy among user groups. The fitness index comprehensively considers the conversion rate improvement degree and the contribution degree of the average order value. Among them, the conversion rate improvement degree is calculated based on the increase in the conversion rate of the target category relative to the benchmark value after the implementation of the policy, and the contribution degree of the average order value is evaluated through the premium ability of the orders associated with the policy.

[0033] In step S104, the Red Queen regulation mechanism refers to an adaptive regulation module that maintains policy diversity, and quantifies the balance of the policy type distribution through the Shannon entropy algorithm.

[0034] Preferably, the detection of policy homogenization degree includes calculating the Shannon entropy value by taking the proportion of policy types as a probability distribution, and generating a policy diversity index in combination with the change trend of the historical entropy value. When the index is lower than the preset threshold, it indicates that the policy concentration is too high and mutation needs to be triggered.

[0035] Preferably, the policy mutation instruction includes requirements for new policy feature combinations. For example, a brand association factor is introduced into the price-sensitive policy. The environmental pressure parameters include a traffic limit coefficient and a promotion attenuation factor, which are used to simulate the constraint effect of external competition pressure on policy evolution in the ecological game.

[0036] In step S105, the ecological game engine refers to a computing platform that simulates the competitive evolution of multiple strategies, and determines the optimal distribution ratio of each policy type through Nash equilibrium calculation. Preferably, the optimal policy combination includes the allocation ratio of differentiated promotion strategies, the priority ranking of category traffic, and the policy failure warning threshold. Among them, the allocation ratio is determined based on the expected revenue weight of the policy type in the equilibrium state, and the traffic priority ranking is dynamically adjusted based on the category stability index in the competition relationship matrix.

[0037] In step S106, a real-time feedback closed-loop link is established. The execution effect of the optimal policy combination is monitored through buried-point data, and feedback data including conversion rate changes and policy penetration rates is generated. The feedback data is then re-input into the niche competition model and the dynamic policy pool for parameter tuning, including: Real-time capture of user behavior response data after the implementation of the optimal policy combination through the buried-point data acquisition module. The user behavior response data includes the fluctuation value of the category conversion rate and the penetration rate of the policy type. Generate a feedback data set based on the user behavior response data. The feedback data set includes a conversion rate change gradient matrix and a policy penetration rate distribution map. Input the conversion rate change gradient matrix into the niche competition model for dynamic correction of the competition intensity value, including: Adjust the dynamic adjustment weight of the real-time substitution coefficient and the scenario weight coefficient of the scenario symbiosis coefficient according to the conversion rate change gradient; Input the policy penetration rate distribution map into the dynamic policy pool for synchronous update of policy propagation information, including: Correct the policy diffusion resistance coefficient and the policy cross-category propagation attenuation factor according to the policy penetration rate distribution map; Based on the corrected competition relationship matrix and the dynamic policy pool, re-execute the policy feature encoding and the Red Queen regulation mechanism detection, trigger the reconstruction of the revenue matrix and the reset of the constraint conditions of the ecological game engine, and complete the closed-loop parameter tuning.

[0038] The real-time feedback closed-loop link captures the user behavior changes after the policy execution through the buried point data, and reversely inputs the feedback data into the model to realize parameter iteration. Specifically, the conversion rate change gradient matrix is used to correct the weight distribution logic of the substitution coefficient and the symbiosis coefficient in the niche competition model, and the policy penetration rate distribution map drives the dynamic policy pool to update the calculation parameters of the policy propagation path. For example, the policy cross-category propagation attenuation factor is adjusted according to the actual penetration rate. The corrected model parameters trigger the policy encoding and game calculation again, forming a policy optimization closed-loop.

[0039] The step process of this embodiment can be understood as: constructing a "competition analysis - strategy evolution - game equilibrium - feedback tuning" cycle system, quantifying the dynamic relationship between categories through the niche competition model, mapping the user behavior to adjustable strategies by the dynamic policy pool, ensuring the strategy diversity by the Red Queen mechanism to avoid the local optimal trap, solving the equilibrium solution under multiple constraints by the ecological game engine, and finally realizing the synchronous evolution of the model parameters and the market environment through real-time feedback. This embodiment forms a strong coupling through the data flow and parameter correction. For example, the competition relationship matrix affects the calculation of the policy propagation path, and the policy penetration rate feedback reversely corrects the competition intensity value, enabling the system to continuously adapt to market changes.

[0040] Please refer to Figure 2 , in some embodiments, input the commodity information and user behavior data into the niche competition model for category competition relationship analysis, and generate a competition relationship matrix, including: S201. Perform multi-dimensional association on the commodity information and user behavior data to generate a category association network, which includes a substitution relationship sub-network, a symbiosis relationship sub-network, and a competition relationship sub-network; S202. Calculate the real-time substitution coefficient corresponding to the category attribute according to the substitution relationship sub-network; S203. Calculate the scenario symbiosis coefficient corresponding to the category attribute according to the symbiosis relationship sub-network; S204. Generate the competition intensity value corresponding to the category attribute according to the real-time substitution coefficient and the scenario symbiosis coefficient; S205. Generate a competition relationship matrix according to the competition intensity value of the category attribute.

[0041] In step S201, the category association network is a category relationship topological structure constructed by multi-dimensional data mapping. Among them, the substitution relationship sub-network reflects the user transfer behavior between categories and is constructed through the user browsing jump sequence and the substitution purchase record; the symbiotic relationship sub-network reflects the co-purchase behavior between categories and is generated based on the high-frequency co-occurrence category combinations in the same user behavior path; the competition relationship sub-network reflects the resource competition behavior between categories and is calculated through the price overlap degree and inventory extrusion effect in the competitor data.

[0042] In step S202, preferably, the real-time substitution coefficient statistically analyzes the group jump frequency matrix through the user cross-category browsing jump sequence set, and performs dynamic weighted fusion in combination with the historical order substitution purchase rate in the out-of-stock scenario. Specifically, it includes extracting the user cross-category jump path within a preset time window, statistically analyzing the group jump frequency matrix between each commodity category and performing standardization processing, synchronously calculating the actual purchase proportion of users turning to substitute categories when out of stock, and balancing the immediate behavior and the historical substitution effect through the dynamic weight formula.

[0043] In step S203, preferably, the scenario symbiosis coefficient statistically analyzes the co-purchase probability through the multi-category combination purchase frequency of a single user transaction, and calculates the association intensity factor in combination with the high-frequency scenario association pattern in the group behavior path. Specifically, it includes analyzing the logical coherence of cross-category access in the user browsing path, using the path similarity algorithm to quantify the scenario association intensity, and finally fusing the co-purchase probability and the scenario association degree through the weighted formula.

[0044] In step S204, preferably, the competition intensity value is calculated by performing dynamic weight allocation on the standardized real-time substitution coefficient and the scenario symbiosis coefficient, where the dynamic weight function is adaptively adjusted according to the market cycle characteristics, and a competitor data correction term is introduced to eliminate the interference of external price fluctuations.

[0045] In step S205, construct an n-dimensional competition relationship matrix, where both the rows and columns correspond to the category attributes in the commodity information; the diagonal elements of the matrix represent the category self-competition intensity, and the degree of internal resource competition of the category is calculated through the real-time inventory quantity and the replenishment cycle in the inventory status; the non-diagonal elements represent the competition intensity values between categories and are directly mapped and filled by the competition intensity values generated in step S204.

[0046] The step process of this embodiment can be understood as follows: the triple sub-structure of the category association network is used to refine the description of the market competition relationship, the substitution relationship is used to quantify the user transfer risk, the symbiotic relationship is used to mine the co-selling opportunities, and the competition relationship is used to evaluate the intensity of resource competition. The calculation of the competition intensity value integrates real-time behavior data and external competitor interference, enabling the matrix to dynamically reflect market changes. The generated competition relationship matrix provides multi-dimensional competition situation input for the subsequent ecological game engine, supporting the accuracy of strategic games.

[0047] In some embodiments, according to the substitution relationship sub-network, calculating the real-time substitution coefficient corresponding to the category attribute includes: Extracting the cross-category browsing and jumping sequence set of the user group within a preset time window; Statistical group jump frequency matrix and average residence time decay factor between each commodity category; Combined with the historical order substitution purchase rate of the user group, calculate the real-time substitution coefficient through the dynamic weighting formula, which is represented by formula (1), and formula (1) is as follows: ; In formula (1), is the historical order substitution purchase rate of the user group, is the first commodity category in the substitution relationship, is the second commodity category in the substitution relationship, is the commodity category to the commodity category 's standardized jump frequency, is the substitution purchase rate when the commodity category is out of stock and turns to the commodity category , is the residence time decay coefficient, is the dynamically adjusted weight, ; According to the symbiotic relationship sub-network, calculating the scenario symbiotic coefficient corresponding to the category attribute includes: Statistical multi-category combined purchase frequency in a single transaction of the user group; Analyzing the high-frequency scenario association patterns in the group behavior path; Calculate the scenario symbiotic coefficient through the scenario symbiotic formula, which is represented by formula (2), and formula (2) is as follows: ; In formula (2), is the first commodity category in the symbiotic relationship, is the second commodity category in the symbiotic relationship, is the co-purchase probability of the commodity category and the commodity category , is the scene correlation strength factor, is the scene weight coefficient; Generating the competition intensity value corresponding to the category attribute according to the real-time substitution coefficient and the scene symbiosis coefficient includes: Normalize the real-time substitution coefficient and the scene symbiosis coefficient, and calculate the competition intensity value using a dynamic weight allocation strategy, which is represented by formula (3). Formula (3) is as follows: ; In formula (3), is the first commodity category in the final competition relationship, is the second commodity category in the final competition relationship, is a time-related dynamic weight function, is the competitor data correction term, all point to the commodity categories included in the commodity information in the same database, and , is the total number of commodity categories in the same database.

[0048] In this embodiment, the cross-category browsing jump sequence set refers to the time sequence record set of the user's cross-category page jumps within a preset time window, which is used to capture the real-time trend of the user's demand transfer; the group jump frequency matrix is constructed by counting the number of jumps between commodity categories, which represents the group behavior pattern of the user's cross-category transfer; the average stay duration decay factor refers to the influence coefficient of the user's stay duration in the original category on the jump behavior. The shorter the stay duration, the larger the decay factor value, which reflects the enhancement of the user's transfer willingness. Dynamically adjust the weight is used to balance the influence weights of the immediate jump frequency and the historical substitution purchase rate, and is dynamically adjusted according to the market promotion cycle. When the promotions are intensive, the jump frequency weight is increased to strengthen the real-time behavior feedback.

[0049] When calculating the scene symbiosis coefficient, the high-frequency scene association pattern refers to the logically coherent combination of cross-category access in the user behavior path, such as the browsing sequence of "sports shoes - sports socks"; the scene correlation strength factor is calculated by the purchase time interval between categories. The shorter the time interval, the higher the correlation strength, which represents the urgency of the scene-based purchase; the scene weight coefficient is used to adjust the contribution ratio of the co-purchase probability and the scene correlation strength, and is adaptively adjusted according to the category attribute characteristics. For example, in high-frequency consumer categories, the weight of the scene correlation strength is higher.

[0050] When generating the competition intensity value, the normalization process preferably eliminates the dimensional differences between categories through min-max normalization; in the dynamic weight allocation strategy, is a time-related function, which dynamically adjusts the confrontation weights of the substitution coefficient and the symbiosis coefficient according to the market competition heat; the competitor data correction term Calculated through the price fluctuation range of external competing products and the inventory extrusion effect to correct the deviation of internal competition relationships.

[0051] The step process of this embodiment can be understood as follows: integrating real-time user behavior and historical substitution rules through a dynamic weight mechanism, quantifying the dynamic relationship of substitution and symbiosis between categories, and at the same time introducing a competing product data correction term to enhance the ability to capture external competition interference, so that the competition intensity value can more comprehensively reflect the multi-dimensional impact of the market environment and provide accurate competition situation input for the ecological game engine.

[0052] Please refer to Figure 3 , in some embodiments, the fitness indicators include conversion rate fitness, value fitness, and competition fitness. A dynamic strategy pool is constructed based on user behavior data. Through strategy feature encoding based on feature engineering, user groups are divided into multiple strategy types, and the strategy propagation information and fitness indicators of each strategy type are monitored in real time, including: S301. According to the user behavior data, the strategy feature encoding extracts the user behavior pattern features through a temporal attention mechanism to generate user feature vectors; S302. Adopt a dynamic threshold clustering algorithm to adjust the cluster boundary according to the competition relationship matrix, perform clustering analysis on the user feature vectors to generate multiple user groups, and each user group corresponds to a strategy type. The strategy types include price-sensitive type, brand-loyal type, and new product tasting type; S303. Initialize the dynamic strategy pool according to the user groups, user feature vectors, and strategy types; S304. Statistically calculate the user proportion information of each strategy type in the dynamic strategy pool in real time, and calculate the strategy propagation information. The strategy propagation information includes the adoption rate of new user strategy types, the migration rate of existing user strategy types, and the strategy diffusion correction value; S305. Calculate the conversion rate fitness, value fitness, and competition fitness in the dynamic strategy pool, including: Calculate the relative improvement value of the conversion rate of each user group in the current competition environment, and perform discount correction on the relative improvement value of the conversion rate according to the category competition intensity to obtain the conversion rate fitness; Statistically calculate the distribution of the unit price of the same user group, and after excluding abnormal fluctuations caused by competition, calculate the premium ability index relative to the category benchmark value to generate the value fitness; Map the category combinations mainly associated with the strategy types of the current user group, and extract the stability indicators of the corresponding categories in the competition relationship matrix to obtain the competition fitness.

[0053] In step S301, the temporal attention mechanism refers to a computational method for extracting time-dependent features from the user browsing path. By capturing the behavioral dependencies at different time steps, it identifies the user's decision-making preferences. For example, it determines the price sensitivity by analyzing the time interval between the browsing and purchasing paths of the user.

[0054] In step S302, the dynamic threshold clustering algorithm is a clustering method that dynamically adjusts the cluster division boundary according to the category competition intensity in the competition relationship matrix. For example, when the category competition intensity increases, it expands the boundary threshold of the price-sensitive user group to accommodate more users with similar characteristics.

[0055] In steps S303 to S304, preferably, the policy diffusion correction value is calculated by weighting the category competition intensity in the competition relationship matrix. Specifically, it attenuates and corrects the original propagation rate according to the competition intensity value of the policy-related category. The higher the competition intensity, the greater the resistance to the cross-category diffusion of the policy.

[0056] In step S305, the category benchmark value refers to the median of the historical average customer price of the same category, which is used to eliminate the interference of market fluctuations on the evaluation of the premium ability; the stability index is calculated by the variance of the symbiotic coefficient of the corresponding category in the competition relationship matrix. The smaller the variance, the stronger the co-selling stability of the policy-related category.

[0057] When calculating the fitness index, the conversion rate fitness eliminates the inflated impact of excessive competition on the conversion rate improvement through discount correction. For example, when the category competition intensity exceeds the threshold, it linearly attenuates the conversion rate improvement value; the value fitness calculates the actual premium ability after excluding abnormal orders (such as promotional brushing data) to ensure that the evaluation result reflects the true user value; the competition fitness maps the stability association between the policy type and the category combination. For example, when the brand loyalty strategy is associated with a highly stable category, it obtains a higher score. The three types of fitness indexes are dynamically weighted and fused, and the weight coefficients are automatically adjusted according to the current policy diversity index to ensure that the evaluation result is adapted to the overall ecological environment.

[0058] This embodiment constructs a policy ecosystem through dynamic clustering and competition-aware fitness evaluation: The temporal attention mechanism captures the temporal evolution law of user behavior, the dynamic threshold clustering adjusts the precision of policy group division according to the market competition situation, the policy propagation monitoring corrects the diffusion path in combination with the competition intensity, and the multi-dimensional dynamic fusion of the fitness indexes ensures the synchronous evolution of policy evaluation and the market environment. Finally, a dynamic policy pool with environmental adaptability is formed, providing an evolvable policy input for ecological games.

[0059] In some embodiments, according to the user behavior data, the policy feature encoding extracts the user behavior pattern features through the temporal attention mechanism, and generates a user feature vector including: Perform time series modeling on the browsing paths of users one by one, capture the behavioral dependencies at different time steps through the multi-head attention mechanism, and obtain the transfer preference intensity of users between different categories; Calculate the response elasticity coefficient of the current user for the price range based on the purchase records, and perform weighted fusion by combining the response elasticity coefficient with the discount usage tendency in the promotion response to obtain the price sensitivity feature; Extract the brand switching frequency and the time delay of the first interaction with new products in the user behavior sequence, and construct the category composite index of brand loyalty and new product acceptance; After performing Min-Max normalization processing on the transfer preference intensity, price sensitivity feature, and category composite index, splice them into the user feature vector of the current user; Adopt a dynamic threshold clustering algorithm, adjust the cluster boundary according to the competition relationship matrix, and perform clustering analysis on the user feature vector to generate multiple user groups including: Initialize the clustering center, and initialize the clustering center based on the category association degree in the competition relationship matrix; During the clustering iteration process, dynamically adjust the cluster boundary threshold according to the real-time competition intensity; Evaluate the clustering effect through the silhouette coefficient. When it is detected that the change in the category competition relationship leads to a decline in the clustering quality, automatically trigger the re-initialization of the clustering; Output user groups with a clear marketing semantic strategy type division; Real-time statistics of the proportion information of users of each strategy type in the dynamic strategy pool, and calculate the strategy propagation information including: Construct a strategy type transfer state matrix, record the migration frequency of users between various strategy types within a fixed time window, and calculate the change gradient of the strategy penetration rate; Quantify the strategy adoption rate of the new user group, and calculate the strategy diffusion resistance coefficient by combining the competition intensity of the source category of the new user group; According to the category association degree in the competition relationship matrix, calculate the attenuation factor of the strategy cross-category propagation, and correct the original propagation rate; According to the change rate of the proportion of each strategy type in the current user group, migration activity, and the impact of the competition environment, generate the strategy adoption rate of new users, the strategy migration rate of existing users, and the strategy diffusion correction value, which is the strategy propagation information.

[0060] In this embodiment, time series modeling refers to the process of converting a user's browsing path into an ordered behavior chain according to timestamps, which is used to capture the law of the evolution of user decision-making preferences over time; the multi-head attention mechanism refers to a deep learning module that analyzes the relevance of behaviors at different time steps in parallel. For example, it captures the jump behaviors during price-sensitive periods and the path dependence during regular browsing phases through multiple groups of attention heads respectively; the transfer preference intensity is calculated by the ratio of the jump frequency between categories to the stay duration, which represents the tendency weight of users' cross-category transfers. The price sensitivity feature is constructed by fusing the price response elasticity coefficient and the discount usage tendency. Among them, the response elasticity coefficient is calculated according to the slope of the impact of price range changes on the purchase volume in the user's historical orders, and the discount usage tendency is quantified by the proportion of orders placed after the user receives a coupon. The category composite index is generated by weighted fusion of the brand switching frequency and the reciprocal of the interaction duration with new products. For example, brand loyalty is calculated by the proportion of repeat purchases of the same brand within a unit time, and the acceptance of new products is inversely mapped by the time interval from the first view to the purchase.

[0061] In the dynamic threshold clustering algorithm, the initialization of the clustering center is determined based on the category correlation degree in the competition relationship matrix. For example, the user feature vectors corresponding to categories with high correlation degrees are preferentially used as the initial centers; the adjustment of the cluster boundary threshold is dynamically implemented according to the real-time competition intensity. For users associated with categories with fierce competition, a more relaxed similarity threshold is adopted to expand the strategy coverage range, and a strict threshold is adopted for those with mild competition to improve the accuracy of the strategy; the silhouette coefficient evaluation refers to judging the clustering quality by calculating the difference in the tightness between the user feature vector and the inside and outside of the cluster. When it is detected that the change in the category competition relationship causes the coefficient to be lower than the preset warning value, re-initialization is triggered to ensure that the strategy type division is synchronized with the market environment.

[0062] In the calculation of strategy propagation information, the strategy type transfer state matrix records the migration trajectories of users among different strategy types. For example, the number of transfers from the brand loyalty type to the price sensitivity type; the strategy diffusion resistance coefficient is calculated by the product of the competition intensity of the new user source category and the inventory extrusion effect. The higher the competition intensity, the greater the strategy diffusion resistance; the strategy cross-category propagation attenuation factor is dynamically adjusted according to the reciprocal of the category correlation degree. The lower the correlation degree, the more significant the attenuation.

[0063] This embodiment constructs a dynamic strategy pool through deep feature extraction and competition-aware clustering. Specifically, it captures the time evolution law of user behaviors through time series modeling, strengthens the feature expression of key behavior nodes through the multi-head attention mechanism, adjusts the precision of strategy group division according to the competition situation through dynamic threshold clustering, and corrects the diffusion path by combining category correlation degrees in strategy propagation monitoring. Finally, it forms a strategy type division with environmental adaptability, providing an accurate strategy input basis for ecological games.

[0064] Please refer to Figure 4, in some embodiments, deploying the Red Queen regulation mechanism in the dynamic policy pool includes: S401. Configure a Red Queen regulator in the dynamic policy pool and establish the following detection mechanism: S402. Set a policy diversity warning threshold, which is dynamically adjusted according to the category competition intensity in the competition relationship matrix; S403. Deploy a policy type distribution monitor to continuously track the real-time proportion fluctuation of policy types; S404. Install a policy ecosystem health diagnosis module to regularly output a homogenization risk assessment report.

[0065] In this embodiment, according to the proportion and change trend of policy types, calculate the policy diversity index through the Shannon entropy algorithm; When it is detected that the policy diversity index is lower than the policy diversity warning threshold, activate the pressure application module of the Red Queen regulator; Maintaining the operating state of the Red Queen regulation mechanism includes: Regularly calibrate the policy diversity warning threshold; Verify the data accuracy of the policy type distribution monitor; Update the evaluation dimension of the policy ecosystem health diagnosis module.

[0066] In step S401, the Red Queen regulator refers to an adaptive regulation device that maintains policy diversity and quantifies the balance of policy type distribution through the Shannon entropy algorithm.

[0067] In step S402, the policy diversity warning threshold refers to the critical value for triggering policy mutation, which is dynamically adjusted according to the category competition intensity in the competition relationship matrix. For example, when the category competition intensity increases, the threshold is lowered to accelerate the mutation response speed.

[0068] In step S403, the policy type distribution monitor refers to a statistical module that real-time tracks the change in the proportion of users of each policy type, and can calculate the proportion volatility through a sliding time window to identify abnormal concentration trends.

[0069] In step S404, the policy ecosystem health diagnosis module refers to a logic unit that periodically evaluates the policy homogenization risk, and generates a risk assessment report by integrating the policy diversity index, the deviation degree of propagation rate, and the fitness decay rate; the pressure application module is activated when the diversity index is lower than the threshold, and forces policy mutation by injecting environmental pressure parameters.

[0070] In the maintenance mechanism, the calibration of the policy diversity warning threshold is dynamically corrected based on the mean and variance of historical diversity indices. The verification of data accuracy can be achieved by cross-comparing the original data of the policy pool with the output of the monitor. The update of the evaluation dimension is synchronously adjusted according to the dimension expansion of the competition relationship matrix.

[0071] In this embodiment, a health monitoring and self-repair mechanism for the strategy ecosystem is constructed. By setting dynamic thresholds to adapt to changes in market competition intensity, real-time monitoring and risk assessment provide early warnings of the degradation of strategy diversity. The maintenance mechanism ensures that the regulator parameters evolve synchronously with the market environment, and finally forms a stability closed-loop control system for the strategy ecosystem.

[0072] In some embodiments, detecting the degree of strategy homogenization through the Shannon entropy algorithm includes: Real-time scanning of the user distribution of all strategy types in the dynamic strategy pool, recording the proportion data of each strategy type in the dynamic strategy pool, and obtaining the proportion of strategy types; Statistical trend of the proportion change of strategy types within a preset time window; According to the proportion of strategy types and the trend of proportion change, calculating the strategy diversity index through the Shannon entropy algorithm, including: Taking the proportion of strategy types as the probability distribution input and calculating the entropy value of the strategy system at the current moment; Calculating the entropy change rate according to the trend of proportion change and generating the strategy diversity index; Judging the degree of strategy homogenization according to the strategy diversity index and updating the homogenization risk assessment report.

[0073] In this embodiment, the proportion of strategy types is the user distribution ratio of each strategy type in the dynamic strategy pool, which is statistically generated by real-time scanning of the user group attribution status; the entropy value of the strategy system refers to the strategy diversity quantification index calculated based on the Shannon entropy formula. Taking the proportion of strategy types as the discrete probability distribution input, the higher the entropy value, the more balanced the strategy distribution; the entropy change rate is calculated by the difference of the entropy value within the sliding time window, representing the evolution rate of strategy diversity, and the negative change rate indicates that the homogenization risk intensifies.

[0074] The strategy diversity index is generated by superimposing the current entropy value and the entropy change rate. For example, when the entropy value decreases and the change rate is negative, the index decreases significantly; the update of the homogenization risk assessment report includes recording the index change trajectory, marking high-risk strategy types, and predicting the homogenization diffusion path.

[0075] This embodiment quantifies the health of the strategy ecosystem through information entropy theory. Specifically, the current distribution state is reflected by the proportion of strategy types, and the dynamic evolution trend is captured by the entropy change rate. The two are fused to generate a diversity index to achieve risk warning, providing an accurate control trigger basis for the Red Queen regulation mechanism to ensure that the strategy pool continuously maintains a diversity level adapted to market competition.

[0076] In some embodiments, when the strategy diversity index is lower than the preset threshold, the environmental pressure regulation module is triggered to generate strategy mutation instructions and environmental pressure parameters, including: Extract the list of high-competition-intensity categories based on the competition relationship matrix, denoted as high-competition product categories; Calculate the intensity of pressure exerted on the high-competition product categories and calculate the duration of pressure to obtain the category pressure parameter; Extract the target strategy types that need to mutate according to the strategy diversity index, generate a mutated strategy feature combination. The mutated strategy feature combination includes core features, non-core features, and new features. The core features are the core effective features of the target strategy type before mutation, the non-core features are the non-core effective features of the target strategy type after random mutation, and the new features are the new features extracted from the current competitive environment; Test the mutated strategy feature combination according to the category pressure parameter until the mutated strategy feature combination meets the preset test indicators, denoted as the final mutated strategy; Generate a strategy mutation instruction according to the final mutated strategy and generate an environmental pressure parameter according to the category pressure parameter.

[0077] In this embodiment, the high-competition product category refers to the product category whose competition intensity value in the competition relationship matrix exceeds the preset critical value. By traversing the non-diagonal elements of the competition relationship matrix, the category pairs with competition intensity values exceeding the preset critical value are screened out, which are the high-competition product categories. Among them, the preset critical value is determined by the statistical distribution of the competition intensity values between categories in the historical competition relationship matrix, and the specific value is set according to the competition intensity requirements of the platform.

[0078] The intensity of pressure exerted is calculated by multiplying the competition intensity value of the high-competition product category by the reference pressure value, which represents the intervention intensity of applying strategy mutation to this category. Among them, the reference pressure value is deduced from the historical strategy intervention effect data. For example, in the statistical past strategy mutation tests, the minimum pressure value required for every 0.1 unit increase in competition intensity is used as the reference. The reference pressure value is a system-level constant parameter, which is preset during initialization and calibrated regularly through a feedback loop.

[0079] The duration of pressure is linearly mapped according to the deviation degree between the strategy diversity index and the preset threshold. The greater the deviation degree, the longer the duration, but it is restricted by the preset maximum duration.

[0080] The generation of the mutated strategy feature combination can be understood as follows: The core features inherit the key effective attributes of the original strategy type, such as the discount response threshold of the price-sensitive strategy; the non-core features are generated by randomly perturbing the secondary parameters of the original strategy, such as adjusting the cross-category browsing frequency limit of the brand-loyalty strategy; the new features are extracted from the high-frequency associated category features in the current competition relationship matrix, such as introducing the strategy of synchronizing competitor promotions.

[0081] Preferably, the preset test indicators include the conversion rate fitness compliance rate and the competition fitness stability threshold of the mutated strategy in the pressure environment.

[0082] Preferably, the environmental pressure parameters are encapsulated as structured data containing the pressure intensity values and action times of each high-competing product category, and the strategy mutation instruction includes a target strategy type identifier, a mutation feature combination scheme, and a test user ratio parameter. During the test process, the test scope is dynamically adjusted according to the category pressure parameters. The higher the pressure intensity, the larger the test user ratio, so as to accelerate the strategy verification efficiency.

[0083] This embodiment reconstructs the strategy evolution path through the environmental pressure of high-competing product categories: the pressure parameters simulate the situation of intensified market competition, the mutated strategy feature combination integrates historical advantages and innovative elements, the test process verifies the adaptability of the strategy in the pressure environment, and finally the generated instructions and parameters jointly drive the strategy pool to restore the diversity balance, forming an anti-homogenization ecological restoration mechanism.

[0084] In some embodiments, the competition relationship matrix, the dynamic strategy pool, and the environmental pressure parameters are input into the ecological game engine for multi-strategy game simulation. The optimal strategy combination is output through Nash equilibrium calculation, including: Mapping the strategy types in the dynamic strategy pool to game participants; Constructing a strategy payoff matrix according to the competition relationship matrix, and taking the environmental pressure parameters as strategy constraint conditions; Calculating the expected payoff of the strategy types in the strategy payoff matrix; Generating a strategy adjustment direction according to the expected payoff; Adjusting the proportion of the strategy types according to the strategy adjustment direction until the Nash equilibrium conditions are met. The Nash equilibrium conditions include that the adjustment amplitude of the strategy types is less than the preset convergence threshold, the rate of change of the expected payoff of the strategy types tends to be stable, and the maximum iteration times limit is reached; Generating an implementation plan for the optimal strategy combination according to the strategy type distribution information, including: Extracting the strategy type proportion in the strategy type distribution information; Calculating the recommended implementation intensity of the strategy types in the strategy type distribution information according to the strategy type proportion, generating a category-level strategy matching comparison table and a strategy adjustment priority list; Obtaining the implementation plan for the optimal strategy combination.

[0085] In this embodiment, the game participants are the strategy types divided in the dynamic strategy pool, and each strategy type participates in the game as an independent decision-making entity; the strategy payoff matrix refers to the payoff relationship table constructed based on the competition intensity values between categories in the competition relationship matrix. The rows and columns respectively correspond to the strategy types, and the matrix element value represents the expected payoff of a certain strategy type when another strategy type exists. The payoff value is discounted and corrected by superimposing the category pressure intensity in the environmental pressure parameters.

[0086] The expected return calculation can be understood as follows: the basic return is calculated according to the fitness index of the strategy type in the current user group; the environmental pressure correction attenuates the return by the product of the pressure intensity value and the action time; the competition correlation effect amplifies the synergy return according to the symbiosis coefficient of the strategy type-related categories. The strategy adjustment direction is generated by calculating the return growth potential of each strategy type, which is manifested as the probability distribution of the user group migrating to the high-return strategy type. For example, if the return of the brand loyalty strategy increases by 10%, the corresponding proportion of its user migration probability will increase.

[0087] In the Nash equilibrium condition, the preset convergence threshold is dynamically set according to the historical game iteration data. Preferably, when the adjustment amplitude of the strategy type proportion is less than this threshold for three consecutive times, it is determined to converge; preferably, the return change rate tends to be stable means that the return volatility of adjacent iteration cycles is lower than 1%; the maximum number of iterations is a system protection parameter to prevent infinite loops.

[0088] Preferably, the recommended implementation intensity is calculated by the weighted product of the strategy type proportion and its fitness index. For example, the implementation intensity of a strategy type with a proportion of 30% and a conversion rate fitness of 0.8 is 24%.

[0089] This embodiment solves the optimal strategy ecosystem through game equilibrium: the strategy return matrix integrates the market competition intensity and environmental pressure constraints, the expected return calculation quantifies the confrontation and synergy effects between strategies, the iterative adjustment process simulates the natural migration of the user group, and the finally output implementation plan reflects the optimal ratio of strategy types under dynamic balance, guiding the precise marketing resource investment through category-level matching and priority ranking.

[0090] Please refer to Figure 5 , in the second aspect, the present invention provides a machine learning-based marketing prediction intelligent analysis system 1, which is applicable to the machine learning-based marketing prediction intelligent analysis method in the first aspect. The system includes: A data acquisition unit 11 for obtaining product information and user behavior data. The product information includes category attributes, inventory status, and competitor data, and the user behavior data includes browsing paths, purchase records, and promotion responses; The logical processing unit 12 is configured to input product information and user behavior data into the niche competition model for analyzing the category competition relationship, generate a competition relationship matrix, where the competition relationship matrix includes substitution coefficients and symbiotic relationships between product categories; construct a dynamic strategy pool based on user behavior data, divide user groups into multiple strategy types through strategy feature encoding based on feature engineering, and real-time monitor the strategy propagation information and fitness indicators of each strategy type; deploy a Red Queen regulation mechanism in the dynamic strategy pool, detect the degree of strategy homogenization through the Shannon entropy algorithm, and when the strategy diversity index is lower than a preset threshold, trigger the environmental pressure regulation module to generate strategy mutation instructions and environmental pressure parameters; input the competition relationship matrix, the dynamic strategy pool, and the environmental pressure parameters into the ecological game engine for multi-strategy game simulation, and output the optimal strategy combination through Nash equilibrium calculation; The closed-loop feedback unit 13 is configured to establish a real-time feedback closed-loop link, monitor the execution effect of the optimal strategy combination through buried-point data, generate feedback data including conversion rate changes and strategy penetration rates, and re-input the feedback data into the niche competition model and the dynamic strategy pool for parameter tuning.

[0091] In this embodiment, the logical processing unit 12 includes a collaborative computing architecture of a niche competition model, a dynamic strategy pool, and an ecological game engine. Among them, the niche competition model generates a competition relationship matrix through the category association network defined in the foregoing embodiment, the dynamic strategy pool constructs a strategy ecosystem based on the feature encoding and clustering mechanism in the foregoing embodiment, and the ecological game engine executes the Nash equilibrium calculation in the foregoing embodiment; the closed-loop feedback unit 13 captures the strategy execution effect in real time through buried-point data, and the process of the feedback data inversely correcting the model parameters follows the closed-loop tuning logic of step S106 in the foregoing embodiment, forming a complete link of "data collection - competition game - strategy generation - feedback evolution".

[0092] The intelligent analysis system 1 of this embodiment executes the intelligent analysis method described in the first aspect. For specific content, refer to the foregoing technical solutions and will not be elaborated here. This system provides original behavior features through the data collection unit 11, the logical processing unit 12 realizes the quantification of competition relationships and the evolution of the strategy ecosystem, and the closed-loop feedback unit 13 ensures the synchronization of system parameters with market dynamics, ultimately forming an intelligent decision-making system with self-evolution ability.

[0093] By adopting the above technical solutions, the present invention is different from the prior art and has the following beneficial effects: The present invention dynamically constructs a competition relationship matrix between product categories through a niche competition model, accurately quantifies substitution and symbiotic effects, and breaks through the limitations of traditional static prediction. Based on feature engineering, a dynamic strategy pool is constructed. A time-series attention mechanism is used to classify user strategy types and monitor propagation information in real time. Combined with the Red Queen regulation mechanism, the strategy diversity is automatically maintained to avoid strategy rigidity caused by manual rule setting. The ecological game engine simulates the co-evolution process of multiple strategies, and uses the Nash equilibrium to calculate and output the optimal strategy combination to achieve dynamic balanced allocation of cross-category marketing resources. The established real-time feedback closed-loop link continuously monitors the execution effect of the strategy and reversely tunes the model parameters to form an adaptive cycle system of "competition analysis - strategy evolution - game equilibrium - feedback tuning". The above technical solutions significantly improve the environmental adaptability and evolution ability of the marketing strategy, effectively solve the problems of strategy homogenization failure, single-category optimization imbalance and market response lag, and achieve intelligent decision-making with ecological self-organization characteristics.

[0094] Finally, it should be noted that although the above embodiments have been described in the text and drawings of the specification of this application, the patent protection scope of this application cannot be limited thereby. Any technical solutions obtained by equivalent structure or equivalent process substitution or modification based on the substantial concept of this application and using the content recorded in the text and drawings of the specification of this application, as well as those directly or indirectly implementing the technical solutions of the above embodiments in other related technical fields, are all included in the patent protection scope of this application.

Claims

1. An intelligent analysis method for marketing prediction based on machine learning, characterized in that, Including: Obtain product information and user behavior data, where the product information includes category attributes, inventory status, and competitor data, and the user behavior data includes browsing paths, purchase records, and promotional responses; Input the product information and user behavior data into a niche competition model for category competition relationship analysis to generate a competition relationship matrix, where the competition relationship matrix includes substitution coefficients and symbiotic relationships between product categories; Construct a dynamic strategy pool based on the user behavior data, divide user groups into multiple strategy types through strategy feature encoding based on feature engineering, and real-time monitor the strategy propagation information and fitness indicators of each strategy type; Deploy a Red Queen regulation mechanism in the dynamic strategy pool, detect the degree of strategy homogenization through the Shannon entropy algorithm, and when the strategy diversity index is lower than a preset threshold, trigger the environmental pressure regulation module to generate strategy mutation instructions and environmental pressure parameters; Input the competition relationship matrix, dynamic strategy pool, and environmental pressure parameters into an ecological game engine for multi-strategy game simulation, and output an optimal strategy combination through Nash equilibrium calculation; Establish a real-time feedback closed-loop link, monitor the execution effect of the optimal strategy combination through buried point data, generate feedback data including conversion rate changes and strategy penetration rates, and re-input the feedback data into the niche competition model and dynamic strategy pool for parameter tuning.

2. The intelligent analysis method for marketing prediction based on machine learning according to claim 1, characterized in that Input the product information and user behavior data into a niche competition model for category competition relationship analysis, and the generated competition relationship matrix includes: Perform multi-dimensional association of product information and user behavior data to generate a category association network, where the category association network includes a substitution relationship sub-network, a symbiotic relationship sub-network, and a competition relationship sub-network; Calculate the real-time substitution coefficient corresponding to the category attribute according to the substitution relationship sub-network; Calculate the scenario symbiotic coefficient corresponding to the category attribute according to the symbiotic relationship sub-network; Generate a competition intensity value corresponding to the category attribute according to the real-time substitution coefficient and scenario symbiotic coefficient; Generate the competition relationship matrix according to the competition intensity value of the category attribute.

3. The intelligent analysis method for marketing prediction based on machine learning according to claim 2, wherein Calculating the real-time substitution coefficient corresponding to the category attribute according to the substitution relationship sub-network includes: Extract the cross-category browsing jump sequence set of the user group within a preset time window; Statistical group jump frequency matrix and average residence time decay factor between each product category; Combined with the historical order substitution purchase rate of the user group, calculate the real-time substitution coefficient through a dynamic weighting formula, which is represented by formula (1), and formula (1) is as follows: ； In formula (1), is the historical order substitution purchase rate of the user group, is the first product category in the substitution relationship, is the second product category in the substitution relationship, is the product category to the product category of the standardized jump frequency, is the product category when out of stock, it turns to the substitution purchase rate of the product category , is the dwell time decay coefficient, is the dynamic adjustment weight, ; Calculating the scenario symbiotic coefficient corresponding to the category attribute according to the symbiotic relationship sub-network includes: Statistical multi-category combination purchase frequency in a single transaction of the user group; Analyze the high-frequency scenario association patterns in the group behavior path; Calculate the scenario symbiotic coefficient through the scenario symbiotic formula, which is represented by formula (2), and formula (2) is as follows: ； In formula (2), is the first product category in the symbiotic relationship, is the second product category in the symbiotic relationship, is the product category and the product category of the co-purchase probability, is the scenario association strength factor, is the scenario weight coefficient; Generating the competition intensity value corresponding to the category attribute according to the real-time substitution coefficient and scenario symbiotic coefficient includes: Standardize the real-time substitution coefficient and the scenario symbiosis coefficient, and calculate the competition intensity value using a dynamic weight allocation strategy, which is represented by formula (3), and the formula (3) is as follows: ； In formula (3), is the first product category in the final competition relationship, is the second product category in the final competition relationship, is the dynamic weight function related to time, is the competitor data correction term, all point to the product categories included in the product information in the same database, and , is the total number of product categories in the same database.

4. The intelligent analysis method for marketing prediction based on machine learning according to claim 1, characterized in that, The fitness indicators include conversion rate fitness, value fitness, and competition fitness. Based on the user behavior data, a dynamic policy pool is constructed. Through policy feature encoding based on feature engineering, user groups are divided into multiple policy types, and the policy propagation information and fitness indicators of each policy type are monitored in real time, including: According to the user behavior data, the policy feature encoding extracts user behavior pattern features through a temporal attention mechanism to generate user feature vectors; Adopt a dynamic threshold clustering algorithm to adjust the cluster boundary according to the competition relationship matrix, and perform clustering analysis on the user feature vectors to generate multiple user groups. Each user group corresponds to a policy type, and the policy types include price-sensitive type, brand-loyal type, and new product tasting type; Initialize the dynamic policy pool according to the user groups, user feature vectors, and policy types; Statistically count the user proportion information of each policy type in the dynamic policy pool in real time, and calculate the policy propagation information, where the policy propagation information includes the adoption rate of new user policy types, the migration rate of existing user policy types, and the policy diffusion correction value; Calculate the conversion rate fitness, value fitness, and competition fitness in the dynamic policy pool, including: Calculate the relative improvement value of the conversion rate of each user group in the current competition environment, and perform discount correction on the relative improvement value of the conversion rate according to the category competition intensity to obtain the conversion rate fitness; Statistically count the distribution of the unit price per customer in the same user group, and calculate the premium ability index relative to the category benchmark value after removing abnormal fluctuations caused by competition to generate the value fitness; Map the category combinations mainly associated with the policy types of the current user group, and extract the stability indicators of the corresponding categories in the competition relationship matrix to obtain the competition fitness.

5. The intelligent analysis method for marketing prediction based on machine learning according to claim 4, wherein According to the user behavior data, the policy feature encoding extracts user behavior pattern features through a temporal attention mechanism to generate user feature vectors, including: Perform time series modeling on the browsing paths of users one by one, and capture the behavior dependence relationships at different time steps through a multi-head attention mechanism to obtain the transfer preference intensity of users between different categories; Calculate the response elasticity coefficient of the current user to the price range according to the purchase record, and perform weighted fusion of the response elasticity coefficient and the discount usage tendency in the promotion response to obtain the price-sensitive feature; Extract the brand switching frequency and the time delay of the first interaction with new products in the user behavior sequence, and construct a category composite index of brand loyalty and new product acceptance; Perform Min-Max normalization processing on the transfer preference intensity, price-sensitive feature, and category composite index, and then splice them into the user feature vector of the current user; Adopt a dynamic threshold clustering algorithm to adjust the cluster boundary according to the competition relationship matrix, and perform clustering analysis on the user feature vectors to generate multiple user groups, including: Initialize the clustering center, and initialize the clustering center based on the category correlation degree in the competition relationship matrix; During the clustering iteration process, dynamically adjust the cluster boundary threshold according to the real-time competition intensity; Evaluate the clustering effect through the silhouette coefficient. When it is detected that the change in the category competition relationship leads to a decline in the clustering quality, automatically trigger the re-initialization of the clustering; Output user groups divided by strategy types with clear marketing semantics; Statistically calculate the proportion of users of each strategy type in the dynamic strategy pool in real time, and calculate the strategy propagation information including: Construct a strategy type transition state matrix, record the migration frequency of users among various strategy types within a fixed time window, and calculate the change gradient of the strategy penetration rate; Quantify the strategy adoption rate of the newly added user group, and calculate the strategy diffusion resistance coefficient in combination with the competition intensity of the source category of the newly added user group; Calculate the attenuation factor for cross-category propagation of the strategy according to the category association degree in the competition relationship matrix, and correct the original propagation rate; Generate the strategy adoption rate of the newly added user strategy type, the migration rate of the existing user strategy type, and the strategy diffusion correction value according to the proportion change rate, migration activity, and competition environment impact of each strategy type in the current user group, which is the above-mentioned strategy propagation information.

6. The intelligent analysis method for marketing prediction based on machine learning according to claim 1, characterized in that Deploy the Red Queen regulation mechanism in the dynamic strategy pool, including: Configure a Red Queen regulator in the dynamic strategy pool and establish the following detection mechanisms: Set a strategy diversity warning threshold, which is dynamically adjusted according to the category competition intensity in the competition relationship matrix; Deploy a strategy type distribution monitor to continuously track the real-time proportion fluctuation of strategy types; Install a strategy ecosystem health diagnosis module to regularly output a homogenization risk assessment report.

7. The intelligent analysis method for marketing prediction based on machine learning according to claim 6, wherein Detect the strategy homogenization degree through the Shannon entropy algorithm, including: Real-time scan the user distribution of all strategy types in the dynamic strategy pool, record the proportion data of each strategy type in the dynamic strategy pool, and obtain the strategy type proportion; Statistically calculate the proportion change trend of strategy types within a preset time window; According to the strategy type proportion and the proportion change trend, calculate the strategy diversity index through the Shannon entropy algorithm, including: Input the strategy type proportion as the probability distribution to calculate the entropy value of the strategy system at the current moment; Calculate the entropy change rate according to the proportion change trend, and generate the strategy diversity index; Judge the strategy homogenization degree according to the strategy diversification index, and update the homogenization risk assessment report.

8. The intelligent analysis method for marketing prediction based on machine learning according to claim 1, wherein When the strategy diversity index is lower than the preset threshold, trigger the environmental pressure regulation module to generate a strategy mutation instruction and environmental pressure parameters, including: Extract the high-competition intensity category list from the competition relationship matrix and record it as high-competition products; Calculate the pressure application intensity of high-competition products and calculate the pressure duration to obtain the category pressure parameters; Extract the target strategy type that needs to mutate according to the strategy diversity index, and generate a mutated strategy feature combination, which includes core features, non-core features, and new features. The core features are the core effective features of the target strategy type before mutation, the non-core features are the non-core effective features of the target strategy type after random mutation, and the new features are the new features extracted from the current competition environment; Test the combination of mutation strategy features according to the category pressure parameter until the combination of mutation strategy features meets the preset test index, which is recorded as the final mutation strategy; Generate the strategy mutation instruction according to the final mutation strategy, and generate the environmental pressure parameter according to the category pressure parameter.

9. The intelligent analysis method for marketing prediction based on machine learning according to claim 1, characterized in that Input the competition relationship matrix, dynamic strategy pool and environmental pressure parameter into the ecological game engine for multi-strategy game simulation, and output the optimal strategy combination through Nash equilibrium calculation, including: Map the strategy types in the dynamic strategy pool to game participants; Construct a strategy payoff matrix according to the competition relationship matrix, and use the environmental pressure parameter as a strategy constraint condition; Calculate the expected payoff of the strategy type in the strategy payoff matrix; Generate a strategy adjustment direction according to the expected payoff; Adjust the proportion of the strategy type according to the strategy adjustment direction until the Nash equilibrium condition is met, and generate the final strategy type distribution information. The Nash equilibrium condition includes that the adjustment amplitude of the strategy type is less than the preset convergence threshold, the rate of change of the expected payoff of the strategy type tends to be stable, and the maximum iteration times limit is reached; Generate an implementation plan for the optimal strategy combination according to the strategy type distribution information, including: Extract the proportion of the strategy type in the strategy type distribution information; Calculate the recommended implementation intensity of the strategy type in the strategy type distribution information according to the proportion of the strategy type, and generate a category-level strategy matching comparison table and a strategy adjustment priority list; Obtain the implementation plan for the optimal strategy combination.

10. An intelligent analysis system for marketing prediction based on machine learning, characterized in that, Applicable to the machine learning-based marketing prediction intelligent analysis method according to any one of claims 1-9, the system includes: A data collection unit for obtaining product information and user behavior data. The product information includes category attributes, inventory status and competitor data, and the user behavior data includes browsing paths, purchase records and promotion responses; A logic processing unit for inputting the product information and user behavior data into a niche competition model for category competition relationship analysis to generate a competition relationship matrix, where the competition relationship matrix includes substitution coefficients and symbiotic relationships between product categories; constructing a dynamic strategy pool based on the user behavior data, dividing user groups into multiple strategy types through strategy feature coding based on feature engineering, and real-time monitoring the strategy propagation information and fitness indicators of each strategy type; deploying a Red Queen regulation mechanism in the dynamic strategy pool, detecting the degree of strategy homogenization through the Shannon entropy algorithm, and triggering an environmental pressure regulation module to generate a strategy mutation instruction and an environmental pressure parameter when the strategy diversity index is lower than a preset threshold; inputting the competition relationship matrix, dynamic strategy pool and environmental pressure parameter into the ecological game engine for multi-strategy game simulation, and outputting the optimal strategy combination through Nash equilibrium calculation; A closed-loop feedback unit for establishing a real-time feedback closed-loop link, monitoring the execution effect of the optimal strategy combination through buried-point data, generating feedback data including conversion rate changes and strategy penetration rates, and re-inputting the feedback data into the niche competition model and the dynamic strategy pool for parameter tuning.

Images