E-commerce personalized promotion strategy AI intelligent generation method and system

Abstract

The application discloses an e-commerce personalized promotion strategy AI intelligent generation method and system, and relates to the technical field of artificial intelligence e-commerce marketing. Including the following steps: step one: collect platform behavior data set, commodity information data set, historical promotion effect data set and public environment data set within the authorized range of the user, extract features after anonymization preprocessing, and construct an e-commerce promotion comprehensive database; step two: based on the user historical interaction data in the e-commerce promotion comprehensive database, a user promotion sensitivity evaluation model based on machine learning is constructed, and the sensitivity evaluation value of the user to different promotion methods is calculated. The sensitivity evaluation value is obtained by calculation, the real response degree of the user to various promotion methods can be accurately quantified, the data result is used to replace artificial experience judgment, the promotion method with higher user adaptation degree is preferentially pushed, the user click and order conversion are effectively improved, the invalid marketing cost is reduced, and the user preference change is dynamically adapted.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence e-commerce marketing technology, specifically to an AI-powered method and system for generating personalized e-commerce promotional strategies. Background Technology

[0002] With the rapid development of e-commerce platforms and increasingly fierce market competition, personalized promotions have become an important operational tool for platforms to improve user activity and sales volume. Currently, e-commerce platforms commonly employ various promotional methods such as discounts, spending thresholds, free gifts, coupons, and limited-time flash sales. By pushing promotional information to different user groups, they stimulate user behaviors such as clicking, adding to cart, placing orders, and making payments, thereby increasing the platform's overall sales and market share.

[0003] Most existing e-commerce promotional strategies rely on human experience to formulate uniform or simply categorized promotional plans, making it difficult to achieve precise personalized push notifications for individual users. Some platforms only roughly group users based on their spending level or historical order amount, failing to accurately quantify the actual response of users to different promotional methods, which can easily lead to a mismatch between promotional methods and user preferences. Therefore, this paper proposes an AI-powered intelligent generation method and system for personalized e-commerce promotional strategies to solve these problems. Summary of the Invention

[0004] Technical problems to be solved To address the shortcomings of existing technologies, this invention provides an AI-powered intelligent generation method and system for personalized e-commerce promotional strategies, which solves the problems mentioned in the background section.

[0005] Technical solution To achieve the above objectives, the present invention provides the following technical solution: an AI-powered intelligent generation method for personalized e-commerce promotional strategies, comprising the following steps: Step 1: Collect platform behavior datasets, product information datasets, historical promotion effect datasets, and public environment datasets within the scope authorized by users. After anonymization preprocessing, extract features and construct a comprehensive e-commerce promotion database. Step 2: Based on the historical user interaction data in the e-commerce promotion database, construct a machine learning-based user promotion sensitivity assessment model, calculate the user's sensitivity assessment value to different promotion methods, and form a user promotion sensitivity profile; Step 3: Based on the user's promotion sensitivity profile, generate personalized promotional combinations that meet preset constraints; Step 4: Conduct A / B testing, collect anonymized user behavior data and transaction data from each experimental group, evaluate the actual effects of different promotional combinations, and regenerate personalized promotional combinations based on the actual effects of different promotional combinations.

[0006] Preferably, in step one, the specific steps for anonymization preprocessing and constructing the comprehensive e-commerce promotion database are as follows: The user's unique identifier is irreversibly hashed to remove all personally identifiable information. All collected data were deduplicated, missing values ​​were filled, timestamps were aligned, and the data was standardized. Extract effective features from the data, associate and integrate all feature data according to user identifiers, and form a structured comprehensive e-commerce promotion database; The platform behavior dataset includes time-series data of user clicks, add-to-cart, favorites, order placement, and payment behavior, as well as user registration duration data. The product information dataset includes product categories, brands, prices, inventory, and historical sales data; The historical promotional performance dataset includes the type, intensity, cost, number of participating users, and conversion data of the platform's historical promotional activities. Publicly accessible datasets include publicly available holiday and seasonal data.

[0007] Preferably, in step two, the sensitivity assessment value is obtained in the following way: Retrieve historical interaction data of target users participating in various promotional activities from the comprehensive e-commerce promotion database; Calculate the click-through rate, order conversion rate, and average order value increase for each type of promotional activity for target users. Assign preset weights based on historical promotional performance statistics to click-through rate, order conversion rate, and average order value increase; The click-through rate, order conversion rate, and average order value increase are multiplied by their respective weights and then summed to obtain the target user's sensitivity assessment value for the corresponding promotion method.

[0008] Preferably, in step two, the specific steps for forming a user's promotion sensitivity profile are as follows: Iterate through all types of promotional methods and calculate the sensitivity assessment value of the target user for each promotional method; Sort all sensitivity assessment values ​​by promotion method type and combine them into a user sensitivity vector; Based on the user sensitivity vector, a user promotion sensitivity profile is generated, which includes information on the target user's preference for each promotion method.

[0009] Preferably, in step three, the specific steps for generating a personalized promotional combination that meets preset constraints are as follows: Obtain a profile of the target users' sensitivity to promotions; Based on the target users' sensitivity to different promotional methods from high to low, determine the priority of each promotional method and the weight of their combination; Within the limits of the preset constraints, adjust the parameters of each promotion method; Combine different promotional methods and their parameters to generate personalized promotional combinations for target users.

[0010] Preferably, in step four, the specific steps for evaluating the actual effectiveness of different promotional combinations are as follows: Users were randomly divided into a control group and multiple experimental groups. The control group used the platform's general promotional strategy, while the experimental groups used a generated personalized promotional combination. Collect anonymized user behavior data and transaction data from each experimental group in real time; Calculate the core evaluation metrics for each experimental group, including click-through rate, conversion rate, average order value, sales revenue, and return on investment. By comparing the core evaluation indicators of each experimental group with those of the control group, the actual effects of different promotional combinations can be obtained.

[0011] Preferably, the different promotional methods include discounts, spending thresholds reductions, gifts, coupons, and limited-time flash sales.

[0012] Preferably, the parameters of the personalized promotional combination include discount rate, minimum spending requirement, minimum spending amount, type of gift, coupon value, and duration of the promotion.

[0013] Preferably, the preset constraints include a single user promotion cost cap, a total promotion budget cap, a product inventory quantity limit, a platform minimum discount limit, and a limit on the number of times the same user can receive promotional information within the same period.

[0014] An AI-powered system for generating personalized e-commerce promotional strategies, including: The data processing module is used to collect various types of data within the scope authorized by the user, perform anonymization preprocessing, extract features, and build a comprehensive e-commerce promotion database. The user profiling module is used to build a machine learning-based user promotion sensitivity assessment model, calculate the sensitivity assessment value of users to different promotion methods, and form a user promotion sensitivity profile. The strategy generation module is used to generate personalized promotional combinations that meet preset constraints based on user promotion sensitivity profiles. The evaluation and optimization module is used to perform A / B testing, collect anonymized user behavior data and transaction data, evaluate the actual effect of different promotional combinations, and drive the strategy generation module to regenerate personalized promotional combinations based on the evaluation results.

[0015] Beneficial effects The present invention has the following beneficial effects: This AI-powered method and system for generating personalized e-commerce promotional strategies calculates sensitivity assessment values, which can accurately quantify the actual response of users to various promotional methods. By replacing human experience-based judgment with data-driven results, it prioritizes and pushes promotional methods with higher user suitability, effectively improving user click-through and order conversion rates, reducing ineffective marketing costs, and dynamically adapting to changes in user preferences.

[0016] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0017] Figure 1 This is a flowchart of the AI-powered method for generating personalized e-commerce promotional strategies according to the present invention. Figure 2 This is a structural diagram of the AI-powered intelligent generation system for personalized e-commerce promotion strategies of the present invention. Detailed Implementation

[0018] 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 embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] This invention provides a technical solution: an AI-powered intelligent generation method for personalized e-commerce promotional strategies, such as... Figure 1 As shown, it includes the following steps: Step 1: Collect platform behavior datasets, product information datasets, historical promotion effect datasets, and public environment datasets within the scope authorized by users. After anonymization preprocessing, extract features and construct a comprehensive e-commerce promotion database.

[0020] In Step 1, the specific steps for anonymization preprocessing and constructing the comprehensive e-commerce promotion database are as follows: The platform's behavioral dataset is collected through front-end event tracking. When users browse product pages, click on promotional activities, add products to their cart, favorite products, submit orders, or complete payments, the corresponding behavioral data is automatically recorded. The event tracking locations cover the platform's homepage, product detail pages, promotional activity pages, shopping cart pages, and checkout pages. The collection frequency is real-time, and the data transmission uses an encrypted transmission method.

[0021] The product information dataset is synchronized hourly through the platform's product management system. This synchronization includes basic product information and real-time inventory changes, ensuring data consistency with the actual product status. The historical promotional performance dataset is automatically exported from the platform's marketing system after each promotional activity, covering all types of data throughout the entire activity cycle to ensure data integrity.

[0022] The public environment dataset is obtained once a day through the public calendar interface, mainly collecting holiday dates and seasonal information for subsequent adaptation and adjustment of promotional strategies.

[0023] The unique user identifier is desensitized by irreversible hashing using the SHA256 algorithm. The processed data cannot be reversed to identify a specific natural person, thus protecting user information security.

[0024] All collected data is deduplicated by identifying and deleting duplicate records using both user hash identifiers and timestamps to avoid data redundancy.

[0025] Missing data is filled using the arithmetic mean of corresponding metrics for users of the same type. Users of the same type are categorized by consumption level and shopping preferences to ensure the reasonableness of the filled data. All data timestamps are aligned to the millisecond level, and Z-score standardization is used to standardize numerical data, eliminating the influence of different data magnitudes. Effective features are extracted from the data, and all feature data are linked and integrated according to user hash identifiers to form a structured e-commerce promotion comprehensive database. This database is stored using a relational database, supporting efficient querying and data retrieval.

[0026] It is worth noting that this step ensures the security of users' personal information through compliant data collection and anonymization. At the same time, it provides a standardized and high-quality data foundation for subsequent model training and strategy generation through unified data format and correlation integration, avoiding the impact of data issues on the implementation effect of subsequent technical steps.

[0027] The platform behavior dataset includes time-series data of user clicks, add-to-cart, favorites, orders, and payments, as well as user registration duration data. The time-series data is accurate to milliseconds, and user registration duration is calculated by day.

[0028] The product information dataset includes product categories, brands, prices, inventory, and historical sales data. Product categories are divided according to the platform's unified classification standards, and historical sales data is the cumulative sales over the past 30 days. The historical promotional performance dataset includes the type, intensity, cost, number of participating users, and conversion data of the platform's historical promotional activities. Conversion data includes click-through rate, order conversion rate, and average order value increase data.

[0029] The publicly available dataset includes publicly available holiday data and seasonal data. The holiday data includes statutory holidays, weekends, and platform-defined holidays, while the seasonal data is divided into natural quarters.

[0030] Step 2: Based on the historical user interaction data in the e-commerce promotion database, construct a machine learning-based user promotion sensitivity assessment model, calculate the user's sensitivity assessment value to different promotion methods, and form a user promotion sensitivity profile.

[0031] In step two, the specific steps for constructing a machine learning-based user promotion sensitivity assessment model are as follows: The model inputs are user historical behavior features from the comprehensive e-commerce promotion database, including user click features, add-to-cart features, order features, payment features, and average order value features. Click features include the number of clicks and click duration; add-to-cart features include the number of times added to the cart and the amount of the added items; order features include the number of orders and the order amount; payment features include the number of payments and the payment amount; and average order value features include the historical average order value and the average order value for participation in promotional activities.

[0032] The historical response results of users to various promotional methods are used as the training labels for the model. The response results are labeled according to whether the user placed an order due to the promotional method. A label of 1 indicates a response, and a label of 0 indicates no response.

[0033] A shallow neural network structure is used to build the model framework, with an input layer, hidden layers and an output layer. The number of neurons in the input layer is consistent with the dimension of the input features. There are 1-2 hidden layers with 32-64 neurons in each layer. The number of neurons in the output layer is consistent with the types of promotion methods, and the output layer outputs the predicted value of the user's sensitivity to each promotion method.

[0034] The model was iteratively trained using historical user data, with training batches set to 100-200 batches and a learning rate of 0.01-0.05. The cross-entropy loss function was used to calculate the error between the predicted value and the actual label. The internal weight parameters of the model were continuously adjusted using gradient descent to make the predicted results of the model output more consistent with the actual user promotion response results.

[0035] After training, the model parameters are saved to obtain a usable user promotion sensitivity assessment model. The model can be incrementally trained based on new user data to ensure model adaptability.

[0036] It is worth noting that this step uses machine learning to automatically learn the mapping relationship between user behavior and promotional responses, improving the accuracy and generalization ability of sensitivity assessment. Compared with traditional manual rule judgment, it can more accurately capture users' potential promotional preferences.

[0037] In step two, the sensitivity assessment value is obtained as follows: Historical interaction data of target users participating in various promotional activities are retrieved from the comprehensive e-commerce promotion database. This includes the time when users clicked on promotional activities, the time when they added items to their cart, the time when they placed orders and made payments, and the corresponding order amount. The data retrieval time range is the past 90 days to ensure the data is timely.

[0038] The number of clicks, add-to-carts, and orders placed by target users for each type of promotional activity within the statistical period are counted separately. The average order value for those participating in promotional activities and those not participating in promotional activities are also counted separately. The average order value is calculated by dividing the total order amount by the number of orders within the corresponding period.

[0039] Click-through conversion rate is calculated by dividing the number of orders placed by the number of clicks. Order conversion rate is calculated by dividing the number of payments made by the number of add-to-cart transactions. Average order value increase is calculated by subtracting the average order value of non-promotional participants from the average order value of non-promotional participants, and then dividing by the average order value of non-promotional participants. If no promotional activity is participated in, the average order value increase is calculated as 0.

[0040] Based on historical promotional performance statistics, corresponding weighting coefficients were assigned to click-through rate, order conversion rate, and average order value increase.

[0041] The three dimensions of indicators are multiplied by their respective weighting coefficients and then summed to obtain the target user's sensitivity assessment value to the corresponding promotion method. The assessment value ranges from 0 to 1, with the value being closer to 1 indicating that the user is more sensitive to this type of promotion method.

[0042] It is worth noting that this step calculates users' sensitivity to different promotional methods by comprehensively measuring multiple dimensions of indicators. This can accurately reflect users' responsiveness to various promotional activities, avoid the bias caused by a single indicator, and provide accurate data support for the subsequent generation of personalized promotional combinations.

[0043] The sensitivity assessment values ​​are obtained as follows: In the formula, This represents the target user's sensitivity assessment value to the corresponding promotional method, and is a dimensionless coefficient. This represents the click-through rate of the target user for this type of promotional activity, with a value ranging from 0 to 1. This represents the conversion rate of target users to orders for this type of promotional activity, with a value ranging from 0 to 1. This indicates the increase in average order value for target users in response to this type of promotional activity, with a value ranging from 0 to 1. The weighting coefficient representing the click-through rate is obtained from historical promotional performance statistics. The weighting coefficient representing the order conversion rate is obtained from historical promotional performance statistics. The weighting coefficient representing the increase in average order value is obtained from historical promotional performance statistics. add add It equals 1.

[0044] In step two, the specific steps for creating a user's promotion sensitivity profile are as follows: Iterate through all types of promotional methods, calculate the sensitivity assessment value of the target user for each promotional method, and iterate in a fixed order of discount, full reduction, free gifts, coupons, and limited-time flash sales.

[0045] All calculated sensitivity assessment values ​​are sorted according to the preset promotion method order and combined to form a one-dimensional user sensitivity vector, where each element corresponds to the sensitivity assessment value of a promotion method.

[0046] Based on the values ​​in the user sensitivity vector, the target user's preference information for each promotion method is extracted. Promotion methods with a sensitivity assessment value greater than 0.7 are marked as high preference, those between 0.3 and 0.7 are marked as medium preference, and those less than 0.3 are marked as low preference, thus generating a user promotion sensitivity profile containing user promotion preference information.

[0047] This profile serves as the core basis for generating subsequent personalized promotional packages. The profile data is stored in association with the user hash identifier, facilitating subsequent retrieval and updates.

[0048] It is worth noting that this step, by transforming multi-dimensional sensitivity assessment values ​​into structured user profiles, can provide a direct basis for decision-making in the generation of subsequent personalized promotional combinations, ensuring that promotional pushes are highly matched with user preferences and improving promotional response efficiency.

[0049] Different promotional methods include discounts, spending threshold reductions, gifts, coupons, and flash sales. Discounts are fixed percentage reductions from the listed price of the product; spending threshold reductions are fixed amount reductions after reaching a specified spending amount; gifts are given as gifts after meeting specific spending conditions; coupons are vouchers that can be used to offset a specified amount; and flash sales are low-price purchase events within a specified time period.

[0050] Step 3: Based on the user's promotion sensitivity profile, generate personalized promotional combinations that meet preset constraints.

[0051] In step three, the specific steps for generating personalized promotional combinations that meet preset constraints are as follows: obtain the promotion sensitivity profile of the target user, extract the user's sensitivity assessment value for different promotional methods, and retrieve the profile data by associating the user hash identifier during the extraction process.

[0052] Users' sensitivity assessment values ​​for different promotional methods are sorted from high to low. The priority of each promotional method is determined based on the sorting results. The higher the sensitivity of the promotional method, the higher the priority. The priority is divided into three levels: first-level, second-level, and third-level. Promotional methods with high preference are first-level priority, medium preference are second-level priority, and low preference are third-level priority.

[0053] At the same time, the combined weight of each promotional method is determined based on the proportion of the sensitivity assessment value of each promotional method to the total sensitivity assessment value of all promotional methods. The weight is calculated by dividing the sensitivity assessment value of a single promotional method by the total sensitivity assessment value of all promotional methods. The combined weight is used to determine the strength ratio of each promotional method in the combination.

[0054] Within the limits of the preset constraints, the parameters of each promotion method are adjusted sequentially according to their adoption priority and combination weight. The parameter adjustment range of the first-priority promotion method is greater than that of the second- and third-priority promotion methods. The higher the combination weight, the greater the parameter discount.

[0055] The adjusted promotional methods and their parameters are combined to generate personalized promotional combinations for target users. Each user's promotional combination is generated separately based on their own sensitivity profile to ensure personalized adaptation.

[0056] It is worth noting that this step automatically generates personalized promotional combinations based on users' promotional preferences, which can effectively improve users' response rate to promotional activities. At the same time, through the restriction of constraints, it ensures that the cost of promotional activities is controllable and avoids problems such as cost overruns and insufficient inventory.

[0057] The parameters for personalized promotional packages include discount rate, minimum spend threshold, minimum spend amount, gift type, coupon value, and promotion duration. The discount rate ranges from 0.7 to 0.95, the minimum spend threshold ranges from 50 yuan to 1000 yuan, the minimum spend amount ranges from 10 yuan to 200 yuan, the gift type matches the user's historical purchase categories, the coupon value ranges from 5 yuan to 100 yuan, and the promotion duration ranges from 1 day to 7 days.

[0058] The preset constraints include a single-user promotion cost cap, a total promotion budget cap, a product inventory quantity limit, a platform minimum discount limit, and a limit on the number of times a single user can receive promotional information within the same period. Among these, the single-user promotion cost cap is no more than 500 yuan, the total promotion budget cap is determined based on the platform's monthly marketing budget, the product inventory quantity limit is no more than the current actual inventory of the products, the platform minimum discount limit is no less than 0.7, and the number of times a single user can receive promotional information within the same period is limited to no more than 3 times, with the period being 7 days.

[0059] Step 4: Conduct A / B testing, collect anonymized user behavior data and transaction data from each experimental group, evaluate the actual effects of different promotional combinations, and regenerate personalized promotional combinations based on the actual effects of different promotional combinations.

[0060] In step four, the specific steps for evaluating the actual effectiveness of different promotional combinations are as follows: All users participating in the promotional activity were randomly divided into a control group and multiple experimental groups. The grouping was carried out by random sampling to ensure that the distribution of the number of users and the distribution of user characteristics in each group were consistent. The number of users in each group was not less than 1,000 to avoid the evaluation results being biased due to the small sample size.

[0061] The control group users were shown the platform's general promotional strategies, which were undifferentiated promotional plans uniformly formulated by the platform. The experimental group users were shown personalized promotional combinations generated, with each experimental group corresponding to one type of personalized promotional combination.

[0062] Anonymized behavioral and transaction data of users in each experimental group are collected in real time. Behavioral data includes the number of clicks, click duration, number of times added to cart, and number of times favorited. Transaction data includes order amount, order quantity, payment amount, and number of payments. The collection frequency is real-time, and the data is stored in encrypted form.

[0063] Calculate the core evaluation indicators for each experimental group, including click-through rate, conversion rate, average order value, sales revenue, and return on investment. Click-through rate is the number of clicks divided by the number of impressions; conversion rate is the number of orders placed divided by the number of clicks; average order value is the total order amount divided by the number of orders placed; sales revenue is the total order amount placed by users in that group; and return on investment is calculated according to a preset formula.

[0064] The core evaluation indicators of each experimental group were compared with those of the control group. The improvement of each indicator was calculated. The actual effect of different promotional combinations was evaluated based on the improvement. An improvement of more than 10% was considered excellent, 5%-10% was considered average, and less than 5% was considered poor.

[0065] Based on the actual effects of different promotional combinations, the parameters of the user promotion sensitivity assessment model are updated. The user sensitivity assessment logic corresponding to the promotional combinations with excellent results is strengthened, while the combinations with poor results are adjusted. Personalized promotional combinations are then regenerated to achieve closed-loop optimization of the strategy.

[0066] It is worth noting that this step objectively evaluates the actual effects of different promotional combinations through A / B testing, and feeds the evaluation results back into the model to form a complete closed-loop optimization system, which can continuously improve the effectiveness of promotional activities and reduce marketing costs.

[0067] The methods for obtaining return on investment are as follows: In the formula, This represents the return on investment for promotional activities. It is a dimensionless coefficient with a positive value range. The larger the value, the better the promotional effect. This represents the sales revenue generated during the promotion, expressed in yuan. It is calculated based on the total order amount placed by the group of users during the promotion period. This represents the total cost of the promotional activity, expressed in yuan. It includes discount costs, full reduction costs, gift costs, and coupon costs. Discount costs are the sum of the differences between the listed price and the discounted price. Full reduction costs are the sum of the full reduction amounts. Gift costs are the total cost of purchasing gifts. Coupon costs are the total amount of coupons deducted.

[0068] AI-powered intelligent generation system for personalized e-commerce promotional strategies, such as Figure 2 As shown, it includes: The data processing module is used to collect various types of data within the scope authorized by the user, perform anonymization preprocessing, extract features, and build a comprehensive e-commerce promotion database. The user profiling module is used to build a machine learning-based user promotion sensitivity assessment model, calculate the sensitivity assessment value of users to different promotion methods, and form a user promotion sensitivity profile. The strategy generation module is used to generate personalized promotional combinations that meet preset constraints based on user promotion sensitivity profiles. The evaluation and optimization module is used to perform A / B testing, collect anonymized user behavior data and transaction data, evaluate the actual effect of different promotional combinations, and drive the strategy generation module to regenerate personalized promotional combinations based on the evaluation results.

[0069] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus.

[0070] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. An AI-powered method for generating personalized e-commerce promotional strategies, characterized in that: Includes the following steps: Step 1: Collect platform behavior datasets, product information datasets, historical promotion effect datasets, and public environment datasets within the scope authorized by users. After anonymization preprocessing, extract features and construct a comprehensive e-commerce promotion database. Step 2: Based on the historical user interaction data in the e-commerce promotion database, construct a machine learning-based user promotion sensitivity assessment model, calculate the user's sensitivity assessment value to different promotion methods, and form a user promotion sensitivity profile; Step 3: Based on the user's promotion sensitivity profile, generate personalized promotional combinations that meet preset constraints; Step 4: Conduct A / B testing, collect anonymized user behavior data and transaction data from each experimental group, evaluate the actual effects of different promotional combinations, and regenerate personalized promotional combinations based on the actual effects of different promotional combinations.

2. The AI-powered intelligent generation method for personalized e-commerce promotion strategies according to claim 1, characterized in that, In step one, the specific steps for anonymization preprocessing and constructing the comprehensive e-commerce promotion database are as follows: The user's unique identifier is irreversibly hashed to remove all personally identifiable information. All collected data were deduplicated, missing values ​​were filled, timestamps were aligned, and the data was standardized. Extract effective features from the data, associate and integrate all feature data according to user identifiers, and form a structured comprehensive e-commerce promotion database; The platform behavior dataset includes time-series data of user clicks, add-to-cart, favorites, order placement, and payment behavior, as well as user registration duration data. The product information dataset includes product categories, brands, prices, inventory, and historical sales data; The historical promotional performance dataset includes the type, intensity, cost, number of participating users, and conversion data of the platform's historical promotional activities. Publicly accessible datasets include publicly available holiday and seasonal data.

3. The AI-powered intelligent generation method for personalized e-commerce promotion strategies according to claim 1, characterized in that, In step two, the sensitivity assessment value is obtained in the following way: Retrieve historical interaction data of target users participating in various promotional activities from the comprehensive e-commerce promotion database; Calculate the click-through rate, order conversion rate, and average order value increase for each type of promotional activity for target users. Assign preset weights based on historical promotional performance statistics to click-through rate, order conversion rate, and average order value increase; The click-through rate, order conversion rate, and average order value increase are multiplied by their respective weights and then summed to obtain the target user's sensitivity assessment value for the corresponding promotion method.

4. The AI-powered intelligent generation method for personalized e-commerce promotion strategies according to claim 1, characterized in that, In step two, the specific steps for forming a user's promotion sensitivity profile are as follows: Iterate through all types of promotional methods and calculate the sensitivity assessment value of the target user for each promotional method; Sort all sensitivity assessment values ​​by promotion method type and combine them into a user sensitivity vector; Based on the user sensitivity vector, a user promotion sensitivity profile is generated, which includes information on the target user's preference for each promotion method.

5. The AI-powered intelligent generation method for personalized e-commerce promotion strategies according to claim 1, characterized in that, In step three, the specific steps for generating personalized promotional combinations that meet preset constraints are as follows: Obtain a profile of the target users' sensitivity to promotions; Based on the target users' sensitivity to different promotional methods from high to low, determine the priority of each promotional method and the weight of their combination; Within the limits of the preset constraints, adjust the parameters of each promotion method; Combine different promotional methods and their parameters to generate personalized promotional combinations for target users.

6. The AI-powered intelligent generation method for personalized e-commerce promotion strategies according to claim 1, characterized in that, In step four, the specific steps for evaluating the actual effectiveness of different promotional combinations are as follows: Users were randomly divided into a control group and multiple experimental groups. The control group used the platform's general promotional strategy, while the experimental groups used a generated personalized promotional combination. Collect anonymized user behavior data and transaction data from each experimental group in real time; Calculate the core evaluation indicators for each experimental group, including click-through rate, conversion rate, average order value, sales revenue, and return on investment. By comparing the core evaluation indicators of each experimental group with those of the control group, the actual effects of different promotional combinations can be obtained.

7. The AI-powered intelligent generation method for personalized e-commerce promotion strategies according to claim 1, characterized in that, The different promotional methods include discounts, spending thresholds reductions, gifts, coupons, and limited-time flash sales.

8. The AI-powered intelligent generation method for personalized e-commerce promotion strategies according to claim 1, characterized in that, The parameters of the personalized promotional package include discount rate, minimum spending requirement, minimum spending amount, type of gift, coupon value, and duration of the promotion.

9. The AI-powered intelligent generation method for personalized e-commerce promotion strategies according to claim 1, characterized in that, The preset constraints include a single user promotion cost cap, a total promotion budget cap, a product inventory quantity limit, a platform minimum discount limit, and a limit on the number of times a single user can receive promotional information within the same period.

10. An AI-powered intelligent generation system for personalized e-commerce promotional strategies, applied to any one of the AI-powered intelligent generation methods for personalized e-commerce promotional strategies described in claims 1-9, characterized in that... include: The data processing module is used to collect various types of data within the scope authorized by the user, perform anonymization preprocessing, extract features, and build a comprehensive e-commerce promotion database. The user profiling module is used to build a machine learning-based user promotion sensitivity assessment model, calculate the sensitivity assessment value of users to different promotion methods, and form a user promotion sensitivity profile. The strategy generation module is used to generate personalized promotional combinations that meet preset constraints based on user promotion sensitivity profiles. The evaluation and optimization module is used to perform A / B testing, collect anonymized user behavior data and transaction data, evaluate the actual effect of different promotional combinations, and drive the strategy generation module to regenerate personalized promotional combinations based on the evaluation results.

Images