Method and system for constructing e-commerce practical training system

Abstract

The invention relates to the technical field of e-commerce, in particular to a construction method and system of an e-commerce practical training system, which quantifies a market environment through a dynamic environment complexity index, realizes dynamic adaptation of environment complexity, constructs a virtual customer agent and a competitor agent, and improves the training efficiency of e-commerce. The virtual customer agent is embedded with personality attributes and is associated with dynamic environment complexity index dynamic adjustment behaviors, the competitor agent adopts a genetic algorithm to optimize a competition strategy, multi-dimensional evaluation and anti-fact attribution analysis are introduced through a standardized business framework and a bidirectional interaction interface, influences of decisions on finance, markets and other dimensions are quantified, and the real-time performance of the competitor agent is improved. An attribution label is marked for a super-threshold negative contribution decision, a dynamic environment complexity index formula and an agent model are adaptively optimized based on an analysis result, the authenticity of a practical training scene is improved, the ability of a student is dynamically matched, a decision short board is accurately positioned, and the practical training effect and the actual combat ability of the student are effectively enhanced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electronic commerce, in particular to a construction method and system of an e-commerce practical training system. BACKGROUND

[0002] Traditional e-commerce practical training systems rely on static scenarios and fixed scripts, and the interaction of intelligent agents is weak, making it difficult to simulate real market dynamic competition and the complexity of student behavior, resulting in a disconnect between practical training and actual combat. With the rapid iteration of the e-commerce industry, enterprises are demanding higher practical decision-making skills from practitioners, and there is an urgent need for a practical training system that can dynamically adapt to the environment and provide accurate feedback. The present method was developed in response to this need.

[0003] The existing construction method and system of an e-commerce practical training system at least have the following technical problems: 1. The existing construction method and system lack a dynamic environment complexity generation mechanism based on real market data and quantification, which can lead to a static and idealized simulation environment that cannot accurately reflect the real market volatility and risks in different categories and cycles, resulting in a disconnect between the practical training content and the real business scenario, and students cannot obtain effective training to cope with complex and changing market environments.

[0004] 2. The existing construction method and system lack an intelligent agent ecosystem driven by artificial intelligence and with autonomous decision-making capabilities, which can result in fixed and predictable behavior patterns of simulated customers and competitors in the system, relying on pre-set scripts for simple interaction and unable to produce unexpected emergent behavior, making the decision-making process of students lack of challenge and difficult to cultivate their dynamic game and real-time response capabilities required in actual business competition.

[0005] 3. The existing construction method and system lack multi-dimensional penetrating diagnosis of the student's operation process and self-evolution ability of the system, which can result in an evaluation system that only focuses on a small number of result-oriented indicators such as final operating performance, cannot trace back the decision-making logic chain and conduct attribution analysis, and the system itself cannot learn and optimize from the common mistakes of students, resulting in a one-sided evaluation of the practical training effect and a lack of targeted teaching guidance, with the system function stagnating for a long time and unable to become more intelligent with use. SUMMARY

[0006] In view of the above-mentioned shortcomings of the prior art, the present application provides a construction method and system of an e-commerce practical training system, which can effectively solve the problems in the background art.

[0007] To solve the above technical problems, the present application adopts the following technical solution: the present application provides a construction method of an e-commerce practical training system, comprising: S1, obtaining a target product for e-commerce practical training, collecting desensitization data of the target product within a set time period, and calculating a dynamic environment complexity index corresponding to the target product.

[0008] S2, an agent construction module is set in the practical training system, and a virtual customer agent and a competitor agent corresponding to the target product are respectively constructed through dynamic simulation simulation.

[0009] S3, an interaction interface connecting the agent and the operation of the student is constructed.

[0010] S4, practical training data of the student in the practical training operation process is obtained, the scores of the student in each core dimension are evaluated, and attribution analysis is performed on the scores of each core dimension.

[0011] S5, the calculation formula of the dynamic environment complexity index, the virtual customer agent and the competitor agent corresponding to the target product are optimized.

[0012] Preferably, the target product of the e-commerce practical training is obtained, and the specific process is as follows: the selected product of each practical training student is taken as each candidate product, the growth rate and market competition degree of each candidate product are collected, and the practical training adaptation index of each candidate product is calculated.

[0013] The candidate product corresponding to the first ranked practical training adaptation index is selected as the target product.

[0014] Through hierarchical strategy, desensitization data of the target product in a set time period is obtained.

[0015] Preferably, the virtual customer agent corresponding to the target product is constructed, and the specific process is as follows: a specific number of virtual customer agents are calculated and generated according to a predetermined proportion function.

[0016] The psychological characteristics of the customer are extracted from the comment data in the desensitization data, and the quantifiable core personality attributes of each virtual customer agent are set.

[0017] When the practical training system generates each virtual customer agent, the stored target product market consumer psychological portrait is called to generate corresponding quantization values for the core personality attributes of each virtual customer agent.

[0018] When the student is performing practical training, the personality attribute value of the virtual customer agent, the target product information of the current interaction and the environment context state are jointly used as inputs.

[0019] If the behavior of the virtual customer agent belongs to consultation, the combination is submitted to the large language model as a prompt word, and a natural language question is generated and output in real time.

[0020] If the behavior of the virtual customer agent belongs to purchase decision, it is input into a preset purchase probability decision function for calculation, and the final purchase behavior of the virtual customer agent is determined according to the function output result.

[0021] Preferably, the constructing the competitor agent corresponding to the target product is specifically as follows: according to the dynamic environment complexity index of the target product, the initial number and the basic competition strength of the competitor agent are calculated.

[0022] According to the initial number of the competitor agent, each competitor agent is generated.

[0023] Each dominant strategy is assigned from a predefined strategy set, and initial operation parameters are configured for each dominant strategy.

[0024] A strategy optimization cycle based on a genetic algorithm is built in each competitor agent, and the basic competition strength is directly mapped to a mutation probability and a mutation range in the genetic algorithm.

[0025] The market monitoring and behavior response function of each competitor agent is activated to autonomously generate competition decisions and behaviors.

[0026] The present application provides, in a second aspect, a construction system of an e-commerce training system, comprising: a complexity quantification module for obtaining a target product of e-commerce training, collecting desensitization data of the target product within a set time period, and calculating a dynamic environment complexity index corresponding to the target product.

[0027] An agent ecological construction module is used to set an agent construction module in the training system, and through dynamic simulation, a virtual customer agent and a competitor agent corresponding to the target product are constructed respectively.

[0028] A two-way interaction channel building module is used to construct an interaction interface connecting the agent and the student operation.

[0029] A multi-dimensional evaluation and root cause analysis module is used to obtain training data of the student in the training operation process, evaluate the scores of the student in each core dimension, and perform attribution analysis on the scores of each core dimension.

[0030] A self-evolution iteration module is used to optimize the calculation formula of the dynamic environment complexity index, the virtual customer agent and the competitor agent corresponding to the target product.

[0031] The technical scheme provided by the present application has the following beneficial effects compared with the known prior art: 1. In the target product selection and data preparation process of this invention, by introducing a training adaptation index and implementing a tiered data collection strategy, it is beneficial to achieve a high degree of alignment between training content and real business prospects. By using quantitative indicators to scientifically select target products from candidate products that have growth potential, a certain amount of practical space, and meet the needs of the job market, the practicality and forward-looking nature of the training are ensured. At the same time, by collecting de-identified data from macro, meso, and micro levels, a solid and comprehensive data foundation is laid for building a high-fidelity simulation environment, avoiding simulation distortion caused by a single data dimension.

[0032] 2. In the virtual economic environment construction process of this invention, a group of intelligent agents is generated based on the dynamic environment complexity index, and they are given core personality attributes and competitive strategies driven by data. This helps to create a highly realistic and controllable dynamic competitive environment. The consultation and purchasing behavior of virtual customer intelligent agents is determined in real time by their personality attributes, and infinitely diverse natural interactions are generated with the help of a large language model. The competing customer intelligent agents simulate the learning and evolution capabilities in market competition through genetic algorithms, so that each training session faces unpredictable challenges, greatly enhancing the realism and challenge of the training, and effectively training the trainees' dynamic decision-making and adaptability.

[0033] 3. In the process of building a business system, this invention constructs a standardized business framework and implements a bidirectional mirror mapping interactive interface. This helps to ensure system stability and scalability while enabling seamless and real-time bidirectional interaction between the learner and the intelligent agent environment. The volatile business logic is encapsulated into standardized microservice modules, ensuring the reliability and maintainability of the system. The innovation lies in the fact that the dynamic integration module acts as a "translator" and "router" between the real world and the virtual world, so that every operation of the learner can accurately affect the virtual environment, and every action of the intelligent agent can be fed back to the learner in real time.

[0034] 4. In the evaluation and attribution analysis process, this invention establishes a multi-dimensional evaluation index system and uses advanced technologies such as decision chain reconstruction and counterfactual analysis for attribution. It not only comprehensively evaluates from four dimensions: financial, market, strategy, and risk, but more importantly, it penetrates the surface data to restore the trainees' decision-making logic chain. It also quantifies the specific contribution of each decision to the result through counterfactual simulation, thereby providing trainees with highly targeted and actionable improvement suggestions, and elevating the training from result assessment to decision-making thinking training.

[0035] 5. In the system self-optimization process, the embodiments of the present invention adjust the weights of the dynamic environment complexity index formula and the agent behavior model based on the data driven by the attribution analysis results. This is conducive to the self-adaptation and self-evolution of the entire training system, so that the training system always maintains the optimal level of challenge that matches the students' abilities, and actively focuses on training the students' weaknesses, ultimately achieving large-scale individualized teaching and continuous improvement of teaching effectiveness. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0037] Figure 1 This is a schematic diagram of the implementation steps of the present invention.

[0038] Figure 2 This is a schematic diagram of the system structure connection of the present invention. Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0040] The present invention will be further described below with reference to embodiments.

[0041] Please see Figure 1 As shown, a method for constructing an e-commerce training system includes at least the following: S1. Obtain the target product for e-commerce training, collect anonymized data of the target product within a set time period, and calculate the dynamic environment complexity index corresponding to the target product.

[0042] In a specific embodiment, the process of obtaining the target product for e-commerce training is as follows: retrieve the selected products of each trainee within a set time period from the database, take the selected products of each trainee as candidate products, connect to a third-party industry data platform, collect the growth rate and market competitiveness of each candidate product, and then calculate the training suitability index of each candidate product using the training suitability index calculation formula.

[0043] The training fit index of each candidate product is sorted in descending order, and the candidate product with the highest training fit index is selected as the target product.

[0044] Data collection was conducted using a tiered strategy, starting from the macro-market level, the meso-competitive level, and the micro-consumer level. Anonymized data of target products within a set time period was obtained from publicly available data sources on various e-commerce platforms by simulating human browsing behavior.

[0045] It should be noted that the growth rate refers to the sales growth rate of the candidate product within a set time period, obtained through a third-party industry data platform, such as publicly available data from e-commerce platforms or industry reports. The sales growth rate is calculated using the formula: (current period sales - base period sales) / base period sales × 100%, and standardized to a dimensionless value between 0 and 1.

[0046] It should be noted that market competitiveness is obtained by calculating and converting the Herfindahl-Hirschman Index (HHHMA) of the market in which the candidate product is located. The specific process is as follows: obtain the market share data of the competitors of each candidate product from a third-party industry data platform, count the market share of each competitor and calculate the Herfindahl-Hirschman Index, which is the sum of the squares of the market shares of each competitor. Then, normalize the Herfindahl-Hirschman Index to the range of 0 to 1. Then, calculate the market competitiveness by subtracting the normalized Herfindahl-Hirschman Index from 1. For example, if the Herfindahl-Hirschman Index of a certain market is 5300, then the normalized value is 5300 / 10000, which equals 0.53. Therefore, the market competitiveness is 1 - 0.53, which equals 0.47.

[0047] In a specific embodiment, the calculation of the dynamic environment complexity index corresponding to the target product is carried out as follows: four core parameters are extracted from the de-identified data: the number of inventory units, the price fluctuation range, the standard deviation of sales volume, and the intensity of marketing activities. Weights are assigned to the four core parameters respectively, and the four core parameters are standardized respectively. Then, the dynamic environment complexity index corresponding to the target product is obtained by weighted calculation.

[0048] It should be noted that the inventory holding unit quantity refers to the total number of specific products with different specifications, models, styles, and other distinguishable characteristics within the category to which the target product belongs. This reflects the richness and segmentation of the category. It is obtained by counting the number of different specifications of all competing products and the product itself within the category to which the target product belongs from the mid-level competitive information of the anonymized data, and removing duplicates. For example, when the target product is a certain type of sports shoe, the total number of sports shoes with different sizes, colors, and materials in that category is counted from the anonymized data. If there are 42 distinguishable specific products, then the inventory holding unit quantity is 42.

[0049] It should be noted that the price fluctuation range is extracted from the price change records in the anonymized data. The price data of the target product within a set time period is statistically analyzed, and the ratio of the highest price minus the lowest price to the average price of that time period is calculated to obtain the price fluctuation range. The sales volume standard deviation is extracted from the transaction volume records in the anonymized data. Daily sales data within a set time period are selected, and the standard deviation of the sales volume data is calculated using statistical methods to reflect the dispersion of sales volume. The marketing activity intensity is extracted from the marketing activity records in the anonymized data. The total number of marketing activities and the cumulative duration of marketing activities within a set time period are statistically analyzed, and the results are obtained through weighted calculation and integration to quantify the activity level of the marketing activities.

[0050] It should also be noted that the specific process for obtaining the intensity of a marketing campaign is as follows: normalize the total number of campaigns by dividing the maximum number of campaigns for that product category within a preset historical period (such as the past 36 months), normalize the cumulative duration by dividing the maximum duration for that product category within the same preset historical period, and then multiply the normalized number of campaigns and duration by weights determined by the expert method and sum them up. The result is the intensity of the marketing campaign, for example, the weight of the number of campaigns is 0.6 and the weight of the duration is 0.4.

[0051] It should also be noted that the process of assigning weights to the four core parameters involves using expert scoring or the analytic hierarchy process to determine the weight ratios. For example, the weight of inventory holding units is 0.3, the weight of price fluctuation range is 0.3, the weight of sales standard deviation is 0.2, and the weight of marketing activity intensity is 0.2. The formula for calculating the dynamic environment complexity index is: the index equals (standardized value of inventory holding units × 0.3) + (standardized value of price fluctuation range × 0.3) + (standardized value of sales standard deviation × 0.2) + (standardized value of marketing activity intensity × 0.2).

[0052] S2. Set up an agent construction module in the training system, and construct virtual customer agents and competitor agents corresponding to the target product through dynamic simulation.

[0053] In a specific embodiment, the process of constructing the virtual customer intelligent agent corresponding to the target product is as follows: based on the dynamic environment complexity index value of the target product, a specific number of virtual customer intelligent agents are calculated and generated according to a predetermined proportion function.

[0054] Based on anonymized data of the target product within a set time period, customer psychological characteristics are extracted from the comment data in the anonymized data. Quantifiable core personality attributes are set for each virtual customer agent, including price sensitivity, brand loyalty, novelty pursuit, and decision-making prudence.

[0055] When the training system generates each virtual customer agent, it calls the stored psychological profile of the target product market consumers and generates corresponding quantitative values ​​for the core personality attributes of each virtual customer agent.

[0056] When trainees are conducting practical training, they will use the personality attribute values ​​of the virtual customer intelligent agent, the target product information of the current interaction, and the environmental context state as input.

[0057] If the behavior of the virtual customer agent is considered to be consulting, it is combined into prompt words and submitted to the large language model to generate and output natural language questions in real time.

[0058] If the behavior of the virtual customer agent is a purchase decision, it is input into a preset purchase probability decision function for calculation, and the final purchase behavior of the virtual customer agent is determined based on the function output.

[0059] It should be noted that the specific process of extracting customer psychological characteristics from the anonymized review data is as follows: The extraction process is implemented through natural language processing technology, based on a predefined feature lexicon (which is constructed by collecting typical words in the field and reviewing them by experts). For example, the price-sensitive lexicon includes words such as 'price', 'expensive', and 'good value'. Keyword matching and word frequency statistics are performed on the reviews, and the review sentiment score is output by a pre-trained sentiment analysis model (such as a text classification model based on the BERT architecture). This sentiment score is a quantitative value with a value range of [-1, +1], where -1 represents extremely negative and +1 represents extremely positive. The initial value of each psychological characteristic is calculated through a quantitative formula. For example, the initial value of price sensitivity = price-related word frequency × (1 - |sentiment score|). The initial values ​​of all reviews are statistically aggregated and mapped to the range of 0 to 10, thereby generating specific quantitative values ​​of the core personality attributes for each virtual customer agent.

[0060] It should be noted that the specific process of calculating and generating a specific number of virtual customer agents according to a predetermined ratio function is as follows: the higher the dynamic environment complexity index, the more virtual customer agents there are. The ratio function is preset to be positively correlated, and the predetermined ratio function is in the form of: number of virtual customer agents = number of basic customers + (dynamic environment complexity index × scaling factor). The specific values ​​of the number of basic customers and the scaling factor are preset during system initialization based on the historical average customer volume of the target product category.

[0061] For categories like "smartphones," which have high sales volume and a wide audience, the historical average monthly unique visitors might be 100,000. To simulate this large market in a virtual environment, high parameters are set, such as a base customer count of 80 and a scaling factor of 2.0. When the dynamic environment complexity index is 60, the number of virtual customers is 200. However, for a relatively niche category like "professional books," the historical average monthly unique visitors might only be 10,000. Accordingly, lower parameters can be set, such as a base customer count of 30 and a scaling factor of 0.8. Under the same environment complexity index of 60, the number of virtual customers is 78.

[0062] It should be noted that psychological characteristics refer to the intrinsic psychological tendencies that play a dominant role in consumer behavior, extracted from anonymized user reviews of target products. For example, tendencies such as "sensitive to price differences," "long-term preference for a certain brand," "like to try new products," and "repeated comparison before purchase" reflected in user reviews are psychological characteristics. These are further transformed into quantifiable core personality attributes such as price sensitivity and brand loyalty, and used to simulate the behavioral logic of virtual customers in practical training.

[0063] It should be noted that the quantification process of core personality attributes is based on the analysis results of comment data. For each comment, the frequency of keywords belonging to the predefined feature word library is counted, such as price, brand, novelty and decision-related words. The sentiment score (i.e. the value range of [-1,+1]) is obtained by the sentiment analysis model, and the word frequency and sentiment score are combined for calculation.

[0064] For example, price sensitivity is calculated by multiplying the frequency of price-related terms by the emotional neutrality (1 minus the absolute value of the emotional score); brand loyalty is calculated by multiplying the frequency of brand-related terms by the positive emotional score (when the score is greater than zero); novelty pursuit is calculated by multiplying the frequency of novelty-related terms by the absolute value of the emotional score; and decision-making caution is calculated by multiplying the frequency of caution-related terms by the intensity of negative emotional sentiment (1 minus the emotional score, when the score is less than zero). Then, the average value of each core personality attribute is calculated for all reviews of the target product. The average value of each core personality attribute is then normalized and linearly mapped to the range of 0 to 10, thus obtaining the final quantitative value of each core personality attribute. For example, when the normalized average value of price sensitivity is 0.5, its final quantitative value is 5 points.

[0065] It should be noted that the training system pre-stores the personality attribute distribution of real users of the target product, such as a price sensitivity average of 5.2 points and brand loyalty of 3 to 7 points accounting for 60%. When generating intelligent agents, specific values ​​that conform to the characteristics are randomly generated with reference to this distribution. For example, after calling the profile, a virtual customer intelligent agent is assigned a price sensitivity value of 6 points, which is higher than the average, and a brand loyalty value of 4 points, which is in the middle range.

[0066] It should be noted that a large language model refers to an artificial intelligence model trained on massive amounts of text that can understand and generate natural language, generating context-appropriate questions based on the input. The calculation process of the purchase probability decision function is as follows: if the behavior of the virtual customer agent belongs to a purchase decision, parameters such as personality attribute values, product information, and environmental state are first standardized into model input features.

[0067] The purchase probability decision function uses a logistic regression model, and its calculation formula is as follows: ,in It is the weighted sum of all input parameters and their corresponding weights, i.e. , For the input parameters, The value of is a positive integer. Intercept term, For the first The feature weights corresponding to the item parameters, No. The specific numerical value corresponding to the item parameter, This logic function, obtained through training on historical data or preset by expert experience, ensures that the weighted sum is calculated. Mapping to the (0,1) interval yields a mathematical probability value; for example, after presetting weights, the calculated value is... If the value is 0.6, the probability of purchasing is approximately 0.65, or 65%. This probability is then compared with a randomly generated threshold to determine the final purchase decision.

[0068] It should be noted that the process of generating natural language questions based on combined prompts is as follows: Input variables such as personality attributes, product information, and environmental state are filled into a preset query template to form a structured instruction. For example, the template is: "You are a consumer with a price sensitivity of 8 and a brand loyalty of 6. You are currently looking at a smartwatch priced at 299 yuan, and you have noticed that three competitors are currently running promotions on similar products. Based on the above personality traits, ask the seller a natural language question that you are most likely to be concerned about." This instruction is submitted to a pre-trained large language model (e.g., OpenAI's GPT-4 model, Baidu's Wenxin Yiyan model, etc., generative pre-trained transformer models) via API. The model then generates a natural language question that fits the context, such as: "This watch is a bit expensive. Are there any discounts recently? If I buy it now, can you guarantee that it is genuine and comes with a warranty?" This example is only for illustrative purposes and is not the only one.

[0069] In a specific embodiment, the process of constructing competitor agents corresponding to the target product is as follows: based on the dynamic environment complexity index of the target product, the initial number and basic competitive intensity of competitor agents are calculated, thereby setting the initial intensity of virtual market competition.

[0070] Generate each competitor's intelligent agent based on the initial number of competitor intelligent agents.

[0071] Assign dominant strategies from a predefined set of strategies and configure initial runtime parameters for each dominant strategy.

[0072] The strategy set includes low-price strategy, differentiation strategy, and focus strategy.

[0073] Each competitor's agent incorporates a strategy optimization loop based on a genetic algorithm and calls upon the calculated basic competitive strength, directly mapping the basic competitive strength to the mutation probability and mutation magnitude in the genetic algorithm.

[0074] Activate the market monitoring and behavioral response functions of each competitor's intelligent agent, enabling it to monitor market dynamics and student operations based on the configured strategies and optimization algorithms, and autonomously generate competitive decisions and behaviors, thereby completing the construction of the competitor's intelligent agent corresponding to the target product.

[0075] It should be noted that the preset dynamic environment complexity index ranges from 0 to 100. The higher the value, the more complex the market environment. When calculating the initial number of competitor agents, a preset positive correlation piecewise function is used: when the dynamic environment complexity index is 0 to 20, the initial number is 2; when it is 21 to 40, it is 3; when it is 41 to 60, it is 4; when it is 61 to 80, it is 5; and when it is 81 to 100, it is 6.

[0076] When calculating the basic competitive intensity, a normalized mapping is used: the basic competitive intensity is equal to the dynamic environment complexity index ÷ 100, and the result ranges from 0 to 1. For example, when the dynamic environment complexity index is 60, the basic competitive intensity is 0.6 (60 ÷ 100), and when the index is 80, the basic competitive intensity is 0.8. This directly quantifies the initial intensity of competition in the virtual market.

[0077] It should also be noted that if the dynamic environment complexity index of the target product is 60, the initial number of competitor agents is calculated to be 4. The system agent construction module generates 4 agents numbered J1 to J4, each with basic attributes such as initial capital of 80,000 yuan and inventory of 400 units. Then, the dominant strategy is allocated from the strategy set: J1 and J2 use the low-price strategy, with initial operating parameters of a 12% price reduction, one price adjustment per day, and a minimum selling price of no less than 1.1 times the cost; J3 uses the differentiation strategy, with parameters of 15% R&D investment and two product selling points updated per month; and J4 uses the focus strategy, with parameters of targeting male users aged 28 to 35 and running ads twice a week.

[0078] It should be noted that after setting the initial number of competitor agents, the system assigns a dominant strategy to each agent based on the dynamic environment complexity index using a weighted random allocation algorithm from a predefined set containing low-price strategies, differentiation strategies, and focus strategies, and configures initial operating parameters for each strategy. The strategy allocation weights and parameter values ​​(such as the benchmark for price reduction for low-price strategies and the R&D investment ratio for differentiation strategies) are obtained through statistical analysis of real operational data from e-commerce platforms and are matched with the current environment complexity.

[0079] It should be noted that the specific process of the strategy optimization loop is as follows: The system encodes the strategy parameters of each competitor's agent into chromosomes to form a strategy population. The strategy parameters include at least product pricing, advertising investment ratio and main selling points. The profit obtained by each strategy in the previous simulation cycle is calculated as its fitness. The profit is calculated based on the pricing, sales volume and commodity costs and operating expenses in the system's virtual environment under the strategy.

[0080] Specifically: Profit = (Product Price - Commodity Cost) × Sales Quantity - Operating Expenses such as Advertising Investment. Based on fitness, a roulette wheel algorithm is used to probabilistically select parent strategies for crossover and mutation operations. The intensity of the mutation operation is dynamically adjusted by the basic competition intensity: Mutation Probability = Basic Competition Intensity × Preset Mutation Probability Coefficient, Mutation Amplitude = Basic Competition Intensity × Preset Mutation Amplitude Coefficient × Original Parameter Value. The initial values ​​of the mutation probability coefficient and the mutation amplitude coefficient are determined during system initialization by querying a preset "Environment Complexity - Genetic Algorithm Parameter Mapping Table" based on the dynamic environmental complexity index of the target product. The higher the environmental complexity, the stronger the strategy exploration should be for the competitor's agent, and therefore a higher mutation coefficient should be configured for it.

[0081] The following example further illustrates the above process. Taking two competing intelligent agents, A and B, as an example, their pricing parameters in their strategy chromosomes are 100 yuan and 90 yuan, respectively. After one simulation cycle, A and B obtain profits of 5000 yuan and 3000 yuan, respectively. During the selection phase, A, which has a higher profit, has a significantly greater probability of being selected as the parent than B. Assuming that the system determines the mutation probability coefficient from the mapping table based on the current high dynamic environment complexity index as 10%, the mutation amplitude coefficient as 0.2, and the current basic competition intensity as 0.6, if A is selected and mutates, its pricing parameter has a 6% (0.6 × 10%) probability of mutating, with a mutation amplitude of ±12 yuan (±(0.6 × 0.2 × 100)). Therefore, A's pricing may mutate to 106 yuan or 95 yuan, thus participating in the competition in the next cycle with an optimized strategy. This example is only for illustrative purposes and is not the only limitation.

[0082] In the process of constructing a virtual economic environment, this invention generates a group of intelligent agents based on the dynamic environment complexity index and endows them with data-driven core personality attributes and competitive strategies. This helps to create a highly realistic and controllable dynamic competitive environment. The consultation and purchasing behavior of virtual customer intelligent agents is determined in real time by their personality attributes, and infinitely diverse natural interactions are generated with the help of a large language model. The competing customer intelligent agents simulate the learning and evolution capabilities in market competition through genetic algorithms, making each training session face unpredictable challenges, greatly enhancing the realism and challenge of the training, and effectively training the trainees' dynamic decision-making and adaptability.

[0083] S3. Construct an interactive interface that connects the intelligent agent with the student's operations.

[0084] In a specific embodiment, the construction of the interactive interface connecting the intelligent agent and the trainee's operation is carried out as follows: Develop a front-end interactive interface for the trainees, which includes store management, product management, marketing promotion, order processing and data dashboard functional modules, and the visual and interactive logic is consistent with the back-end management system of a real e-commerce platform.

[0085] A two-way mirror mapping mechanism is implemented in the interactive interface to capture the student's operations on the interface in real time and convert them into calls to the backend API. The state changes of the intelligent agent's environment caused by the student's operations or its own decisions are pushed back to the front-end interactive interface through the backend API, and the corresponding components are redrawn and data updated in real time, thereby providing the student with immediate and visualized market feedback.

[0086] This is used to construct an interface that serves as the core interaction channel between students and the virtual environment.

[0087] It's important to note that, for example, the developed front-end interface is a webpage. After logging in, students can see five modules: "Store Management" (where they can set the store name and business hours), "Product Management" (where they can upload product images and modify prices), "Marketing Promotion" (where they can create discount activities), "Order Processing" (where they can view and ship orders), and "Data Dashboard" (displaying data such as sales volume and traffic). The button positions and operation flow of the interface are consistent with the backend of a real e-commerce platform (such as the Taobao merchant backend). When a student changes the price of a T-shirt from 99 yuan to 89 yuan in the "Product Management" module, the interface captures this operation in real time and transmits the "reduce price by 10 yuan" instruction to the backend via the backend API (the communication interface connecting the front-end interface and the system backend). After the backend processes this, the virtual customer agent increases purchases due to the price reduction, and competitor agents may follow suit with price reductions. These state changes are transmitted back to the frontend via the backend API, triggering the "Data Dashboard" component to redraw (the sales number changes from 50 to 70), and the "Order Processing" module to add 10 orders awaiting shipment. Students can instantly see the market feedback resulting from their actions. This interface becomes the core channel for interaction between students and the virtual market.

[0088] In the process of building a business system, this invention constructs a standardized business framework and implements a bidirectional mirror mapping interactive interface. This facilitates seamless and real-time bidirectional interaction between the learner and the intelligent agent environment while ensuring system stability and scalability. The volatile business logic is encapsulated into standardized microservice modules, ensuring the reliability and maintainability of the system. The innovation lies in the fact that the dynamic integration module acts as a "translator" and "router" between the real world and the virtual world, enabling every operation of the learner to accurately affect the virtual environment, while every action of the intelligent agent can be fed back to the learner in real time.

[0089] S4. Obtain training data from trainees during practical training operations, evaluate trainees' scores in each core dimension, and conduct attribution analysis on the scores in each core dimension.

[0090] In a specific embodiment, the evaluation of trainees' scores in each core dimension is carried out as follows: a multi-dimensional evaluation index system is defined, which includes four core dimensions: financial health, market share, strategy matching degree, and risk control capability, and a corresponding weight coefficient is assigned to each dimension.

[0091] The market results data generated by trainees during the entire training cycle are obtained from the constructed business framework and interactive interface. The market results data includes, but is not limited to, each price adjustment, marketing campaign creation, inventory change record, and the resulting order, traffic and competitor reaction data.

[0092] For financial health and market share scores, quantitative scores are calculated directly based on predefined formulas.

[0093] For strategy matching score, the consistency score between the actual operation and the claimed strategy is calculated by dynamically comparing all subsequent operation logs.

[0094] The risk control strength score is comprehensively evaluated by analyzing market outcome data, including inventory turnover rate, cash flow health, and the effectiveness of responses to competitors' suppressive behaviors.

[0095] It should be noted that the process of assigning corresponding weight coefficients to each dimension is the same as the process of determining the weights of each indicator in the multi-indicator comprehensive evaluation system in the existing technology. For example, by using the expert Delphi method or the analytic hierarchy process, combined with the importance ranking of e-commerce job competency requirements, the weights of financial health (30%), market share (25%), strategy matching degree (25%), and risk control (20%) are set. This will not be elaborated further here.

[0096] It should be noted that the specific process of directly calculating the quantitative score based on the predefined formula is as follows: calculate the original indicator value of each dimension, normalize it to the range of 0-100 points, and then multiply it by the weight coefficient to obtain the weighted score of that dimension. For example: the original value of financial health is (net profit ÷ operating income) × 100. This value is already in percentage form and can be directly used as the score of 0-100 points. If the net profit is 50,000 yuan and the operating income is 200,000 yuan, the original score is 25 points. Then multiply it by 30% weight, and the weighted score is 25 × 30% = 7.5 points; the original value of market share is (sales of student products ÷ total sales in the virtual market) × 100. If the student sales are 800 units and the total sales are 4,000 units, the original score is 20 points. Multiplying by 25% weight, the weighted score is 5 points; the original value of strategy matching degree = (number of consistent operations ÷ total number of operations) × 100. If the consistent operation accounts for 70%, the original score is 70 points, and the weighted score is 17.5 points; the original value of risk control ability is calculated by comprehensively considering sub-indicators such as inventory turnover rate and cash flow health. Assuming the calculated value is 65 points, the weighted score is 65 × 20% = 13 points; the trainee's overall evaluation score is the sum of the weighted scores of each dimension, i.e., 7.5 + 5 + 17.5 + 13 = 43 points. This total score is used to comprehensively measure the trainee's practical training performance.

[0097] It should be noted that the specific process for calculating the consistency score between the actual operation and the claimed strategy for the strategy matching score is as follows: First, natural language processing technology is used to analyze the student's initial input of adopting a differentiation strategy and extract the core elements "enhancing product uniqueness and reducing price competition". Then, the subsequent operation logs are traversed to count the operations that conform to the strategy and those that do not. Finally, the score is calculated as "consistent operation count ÷ total operation count × 100 × weight coefficient". If the consistent operation accounts for 70%, the score is 70 × 25% equal to 17.5 points with a weight of 25%.

[0098] It should be noted that the specific process for comprehensively assessing risk control capability is as follows: First, calculate the inventory turnover rate by dividing the cost of sales by the average inventory. If it scores 60 out of 100, then assess the health of cash flow by the cash inflow-outflow ratio, which scores 70. Finally, analyze the effectiveness of the response to competitor suppression, which scores 65. The three factors are weighted at 30%, 30%, and 40%, respectively, i.e., 60×30%+70×30%+65×40% equals 65 points. Then, multiply by 20% weighting, and finally get 13 points.

[0099] In a specific embodiment, the attribution analysis of the scores of each core dimension is carried out as follows: Based on the trainee's operation log, the trainee's business decisions in the training process are correlated sequentially to restore the complete decision logic chain, thereby transforming the static operation record into a dynamic decision process sequence.

[0100] For each business decision-making stage of the trainees, simulation technology is used to construct counterfactual scenarios. By comparing the simulation results with the actual results, the contribution of each business decision to the scores of each core dimension is quantified.

[0101] When a business decision contributes negatively to the score of a core dimension, and the negative contribution exceeds a preset threshold, the business decision is labeled with an attribution tag.

[0102] It should be noted that the specific calculation process for quantifying the contribution of each business decision to the scores of each core dimension is as follows: For a certain business decision node, first record the actual score change value of each core dimension after the decision is implemented. Then, simulate the counterfactual scenario where the decision was not implemented, and obtain the simulated score change value of each core dimension within the same time period. Finally, subtract the simulated score change value from the actual score change value. The difference is the contribution of the decision to each core dimension. A positive value is a positive contribution, and a negative value is a negative contribution. For example, if the actual score of financial health is 25 points, and the simulated score in the counterfactual scenario where a certain price reduction decision was not implemented is 20 points, then the contribution of the price reduction decision to financial health is +5 points.

[0103] It should be noted that the specific construction process of simulating the counterfactual scenario where the decision was not executed is as follows: At the decision node, the environmental state of the virtual market (including the current state of the virtual customer agent and competitor agents, product inventory and market parameters) is saved as a snapshot. Based on the completely replicated environmental snapshot, the student decision to be evaluated is removed, and all other conditions remain unchanged. The simulation module of the training system is then rerun to simulate the market evolution results if the decision had not been made during that time period, thereby obtaining the simulated score changes of each core dimension.

[0104] In the evaluation and attribution analysis process, this invention establishes a multi-dimensional evaluation index system and employs advanced technologies such as decision chain reconstruction and counterfactual analysis for attribution. It not only comprehensively evaluates from four dimensions—financial, market, strategy, and risk—but more importantly, it penetrates the surface data to reconstruct the trainees' decision-making logic chain. Through counterfactual simulation, it quantifies the specific contribution of each decision to the outcome, thereby providing trainees with highly targeted and actionable improvement suggestions, elevating the training from outcome assessment to decision-making thinking training.

[0105] S5. Optimize the calculation formula for the dynamic environment complexity index, the virtual customer intelligent agent corresponding to the target product, and the competitor intelligent agent.

[0106] In a specific embodiment, the calculation formula for optimizing the dynamic environment complexity index, the virtual customer agent corresponding to the target product, and the competitor agent are specifically processed as follows: When the attribution analysis results in step S4 indicate that the current simulation environment does not match the trainee's ability, the weight parameters in the dynamic environment complexity index formula, the personality attribute distribution of the virtual customer agent, and the strategy probability distribution of the competitor agent are dynamically adjusted through a data-driven algorithm to adaptively optimize the simulation environment difficulty and agent behavior.

[0107] It should be noted that "dynamic adjustment through data-driven algorithms" specifically refers to an adaptive process with the optimization goal of minimizing the total number of negative attribution labels for trainees and improving their scores in the core dimensions of their weaknesses. The system has a built-in rule-based control logic that directly maps the common decision-making error patterns found in the attribution analysis results into adjustment instructions for the system's core parameters.

[0108] For example, when analysis shows that trainees generally have a high error rate in making decisions when dealing with "high price-sensitive customers" (reflected in a surge of related negative attribution labels and a decrease in "strategy matching" scores), the system automatically executes an optimization sequence: first, reducing the weight of "price fluctuation range" in the dynamic environment complexity index formula to generate a stable market environment; second, reducing the distribution ratio of the "high price-sensitive" personality attribute in the virtual customer agent; and third, reducing the probability of competitor agents adopting a "vicious price war" strategy to alleviate competitive pressure.

[0109] In the system self-optimization process, this invention adjusts the weights of the dynamic environment complexity index formula and the agent behavior model based on attribution analysis data. This facilitates the self-adaptation and self-evolution of the entire training system, ensuring that the training system always maintains the optimal level of challenge that matches the trainees' abilities and proactively focuses on training the trainees' weaknesses. Ultimately, this achieves large-scale individualized instruction and continuous improvement in teaching effectiveness.

[0110] Please see Figure 2As shown, an e-commerce training system construction system includes the following modules: a complex quantification module, an intelligent agent ecosystem construction module, a two-way interaction channel construction module, a multi-dimensional evaluation and root cause analysis module, a self-evolutionary iteration module, and a database.

[0111] The complex quantification module is connected to the agent ecosystem construction module and the database, respectively. The agent ecosystem construction module is connected to the two-way interaction channel construction module, which is connected to the multi-dimensional evaluation and root cause analysis module and the database, respectively. The multi-dimensional evaluation and root cause analysis module is connected to the self-evolution iteration module.

[0112] The complexity quantification module is used to acquire the target product for e-commerce training, collect anonymized data of the target product within a set time period, and calculate the dynamic environment complexity index corresponding to the target product.

[0113] The intelligent agent ecosystem building module is used to set up intelligent agent building modules in the training system. Through dynamic simulation, virtual customer intelligent agents and competitor intelligent agents corresponding to the target product are built respectively.

[0114] The two-way interactive channel building module is used to construct an interactive interface that connects the intelligent agent with the student's operations.

[0115] The multi-dimensional assessment and root cause analysis module is used to acquire training data of trainees during practical training operations, assess trainees' scores in each core dimension, and perform attribution analysis on the scores of each core dimension.

[0116] The self-evolutionary iteration module is used to optimize the calculation formula of the dynamic environment complexity index, the virtual customer intelligent agent corresponding to the target product, and the competitor intelligent agent.

[0117] In the target product selection and data preparation process, this invention introduces a training suitability index and implements a tiered data collection strategy, which facilitates a high degree of alignment between training content and real business prospects. By using quantitative indicators, target products with growth potential, practical application space, and employment market demand are scientifically selected from candidate products, ensuring the practicality and forward-looking nature of the training. At the same time, by collecting anonymized data from macro, meso, and micro levels, a solid and comprehensive data foundation is laid for building a high-fidelity simulation environment, avoiding simulation distortion caused by a single data dimension.

[0118] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the protection scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing an e-commerce training system, characterized in that, include: S1. Obtain the target product for e-commerce training, collect anonymized data of the target product within a set time period, and calculate the dynamic environment complexity index corresponding to the target product. S2. Set up an intelligent agent construction module in the training system, and construct virtual customer intelligent agents and competitor intelligent agents corresponding to the target product through dynamic simulation. S3. Construct an interactive interface that connects the intelligent agent with the student's operations; S4. Obtain training data from trainees during practical training operations, evaluate trainees' scores in each core dimension, and conduct attribution analysis on the scores in each core dimension. S5. Optimize the calculation formula for the dynamic environment complexity index, the virtual customer intelligent agent corresponding to the target product, and the competitor intelligent agent.

2. The method for constructing an e-commerce training system according to claim 1, characterized in that, The specific process for obtaining the target product for e-commerce training is as follows: The selected products of each trainee are used as candidate products. The growth rate and market competitiveness of each candidate product are collected, and the training suitability index of each candidate product is calculated. The candidate product with the highest-ranked training fit index was selected as the target product. Data is collected using a tiered strategy to obtain anonymized data of the target product within a specified time period.

3. The method for constructing an e-commerce training system according to claim 2, characterized in that, The specific process for calculating the dynamic environment complexity index corresponding to the target product is as follows: Four core parameters are extracted from the anonymized data: inventory holding unit quantity, price fluctuation range, sales standard deviation, and marketing activity intensity. Each of the four core parameters is assigned a weight and then standardized. Finally, the dynamic environment complexity index corresponding to the target product is obtained through weighted calculation.

4. The method for constructing an e-commerce training system according to claim 3, characterized in that, The specific process for constructing the virtual customer intelligent agent corresponding to the target product is as follows: A specific number of virtual customer agents are calculated and generated according to a predetermined proportional function; Psychological characteristics of customers are extracted from the de-identified data and quantifiable core personality attributes are set for each virtual customer agent. When the training system generates each virtual customer agent, it calls the stored psychological profile of consumers in the target product market and generates corresponding quantitative values ​​for the core personality attributes of each virtual customer agent. When trainees are conducting practical training, they will use the personality attribute values ​​of the virtual customer intelligent agent, the target product information of the current interaction, and the environmental context state as input. If the behavior of the virtual customer agent is a consultation, it is combined into prompt words and submitted to the large language model to generate and output natural language questions in real time. If the behavior of the virtual customer agent is a purchase decision, it is input into a preset purchase probability decision function for calculation to determine the final purchase behavior of the virtual customer agent.

5. The method for constructing an e-commerce training system according to claim 4, characterized in that, The specific process for constructing the competitor's intelligent agent corresponding to the target product is as follows: Based on the dynamic environmental complexity index of the target product, calculate the initial number of competitor agents and the basic competitive intensity; Generate each competitor's intelligent agent based on the initial number of competitor intelligent agents; Assign each dominant strategy from the predefined set of strategies and configure initial running parameters for each dominant strategy; Each competing agent incorporates a strategy optimization loop based on a genetic algorithm, and the basic competitive intensity is directly mapped to the mutation probability and mutation magnitude in the genetic algorithm. Activate the market monitoring and behavioral response functions of each competitor's intelligent agent to autonomously generate competitive decisions and behaviors.

6. The method for constructing an e-commerce training system according to claim 5, characterized in that, The specific process of constructing the interactive interface connecting the intelligent agent and the student's operations is as follows: Develop a front-end interactive interface for trainees and implement a two-way mirror mapping mechanism in the interactive interface; The system captures and converts the student's actions on the interface in real time into calls to the backend API, and pushes the state changes of the intelligent agent's environment caused by the student's actions or its own decisions back to the frontend interactive interface through the backend, thereby constructing an operating interface that serves as the core interaction channel between the student and the virtual environment.

7. The method for constructing an e-commerce training system according to claim 6, characterized in that, The specific process for evaluating trainees' scores across each core dimension is as follows: Define a multi-dimensional evaluation index system, which includes four core dimensions: financial health, market share, strategy matching degree, and risk control capability, and assign corresponding weight coefficients to each core dimension. Obtain market results data generated by trainees during the training period from the interactive interface; For financial health score and market share score, quantitative scores are calculated directly; For strategy matching score, the student's operation log in the market results data is dynamically compared to calculate the consistency score between the actual operation and the claimed strategy. The risk control effectiveness score is obtained by comprehensively evaluating market results data.

8. The method for constructing an e-commerce training system according to claim 7, characterized in that, The specific process of performing attribution analysis on the scores of each core dimension is as follows: Based on the trainees' operation logs, the trainees' business decisions during the training process are linked sequentially to restore the complete decision-making logic chain, thereby transforming static operation records into a dynamic decision-making process sequence. For each business decision-making stage of the trainees, simulation technology is used to construct counterfactual scenarios. By comparing the simulation results with the actual results, the contribution of each business decision to the scores of each core dimension is quantified. When a business decision contributes negatively to the score of a core dimension, and the negative contribution exceeds a preset threshold, the business decision is labeled with an attribution tag.

9. The method for constructing an e-commerce training system according to claim 8, characterized in that, The calculation formula for optimizing the dynamic environment complexity index, the virtual customer intelligent agent corresponding to the target product, and the competitor intelligent agent are described in detail below: When the attribution analysis results of step S4 indicate that the current simulation environment does not match the trainee's ability, the weight parameters in the dynamic environment complexity index formula, the distribution of the personality attributes of the virtual client agent, and the strategy probability distribution of the competitor agent are dynamically adjusted through a data-driven algorithm to adaptively optimize the simulation environment difficulty and agent behavior.

10. A construction system for implementing the construction method of the e-commerce training system according to any one of claims 1-9, characterized in that, Includes the following modules: The complexity quantification module is used to obtain the target product for e-commerce training, collect anonymized data of the target product within a set time period, and calculate the dynamic environment complexity index corresponding to the target product. The intelligent agent ecosystem building module is used to set up intelligent agent building modules in the training system, and to build virtual customer intelligent agents and competitor intelligent agents corresponding to the target product through dynamic simulation. The two-way interactive channel building module is used to construct an interactive interface that connects the intelligent agent and the student's operation; The multi-dimensional assessment and root cause analysis module is used to acquire training data of trainees during the practical training process, assess trainees' scores in each core dimension, and perform attribution analysis on the scores of each core dimension. The self-evolutionary iteration module is used to optimize the calculation formula of the dynamic environment complexity index, the virtual customer intelligent agent corresponding to the target product, and the competitor intelligent agent.

Images