Rural e-commerce precision marketing method based on big data

Abstract

The invention relates to the technical field of rural e-commerce marketing, and discloses a rural e-commerce precision marketing method based on big data, and the method comprises the steps: dividing user data sets of all regions based on the geographic positions and consumption records of users, including purchase and browsing behavior data; determining a user type according to the purchase frequency and the commodity preference; generating a marketing strategy set and determining an execution sequence; generating recommended commodities and a promotion activity list in combination with user types, stock and the like; and associating the data to generate a precision marketing scheme. The method further relates to user data boundary determination, user type subdivision, marketing strategy dynamic adjustment, consumption potential grade division, personalized recommendation strategy generation, marketing scheme verification and correction and the like. According to the method, accurate classification and regional differentiation marketing of users are realized through big data mining, the strategy is dynamically optimized, the marketing effect and the resource utilization rate are improved, and the method is suitable for accurate operation of a rural e-commerce platform.

Description

Technical Field

[0001] The present invention relates to the field of rural e-commerce marketing technology, and specifically to a rural e-commerce precision marketing method based on big data. Background Art

[0002] With the popularization of Internet technology and the development of the rural economy, the rural e-commerce market has shown a rapid growth trend, but it still faces many challenges in actual operation, which makes it difficult to improve marketing effects and restricts the further development of rural e-commerce.

[0003] From a user perspective, rural areas are dispersed, with significant differences in consumption habits, economic levels, and product demands across regions. Traditional marketing approaches often employ a one-size-fits-all approach, failing to precisely target the individual needs of users in different regions. For example, some economically developed rural areas may have a high demand for high-end home appliances and fashionable apparel, while less developed regions may be more interested in daily necessities and agricultural supplies. A unified marketing strategy would not only fail to meet the needs of users in different regions, but would also waste marketing resources.

[0004] In terms of data utilization, rural e-commerce platforms have accumulated a vast amount of user data, but this data has not been effectively integrated and deeply mined. Users' location information, consumption history, browsing behavior, and social interaction data are isolated, making it difficult to form a complete user profile. Traditional marketing methods struggle to accurately identify user types and needs using this data, resulting in a lack of scientific basis for developing marketing strategies. For example, it's impossible to categorize users into highly active users, potential users, or churned users based on their purchase frequency and product preferences, making it difficult to develop targeted marketing plans.

[0005] From the perspective of marketing effectiveness evaluation and adjustment, traditional marketing models lack real-time data monitoring and dynamic adjustment mechanisms. Once a marketing plan is developed, it is often executed throughout the marketing cycle, failing to adjust strategies in a timely manner based on market feedback and changes in user behavior. For example, if user interest in a recommended product in a certain region decreases, there's no way to promptly detect and adjust the recommended product list or promotional campaign priorities, resulting in poor marketing results.

[0006] Furthermore, infrastructure and logistics capabilities in rural areas also impact marketing effectiveness. Product accessibility and digital penetration vary across regions, and traditional marketing approaches fail to fully account for these factors. For example, in areas with limited logistics and distribution capabilities, recommending fresh produce that requires fast delivery can negatively impact the user experience due to delivery time constraints, thus reducing marketing effectiveness. Summary of the Invention

[0007] The purpose of the present invention is to provide a rural e-commerce precision marketing method based on big data to solve the problems raised in the above background technology.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a rural e-commerce precision marketing method based on big data, the method comprising:

[0009] Divide user data of the rural e-commerce platform based on user geographic location information and consumption records to obtain user data sets for each region, including purchase behavior data and browsing behavior data;

[0010] Determine the user type in each region based on the purchase frequency and product preference of the user data set in each region;

[0011] Generating a marketing strategy set for each region based on the user type, purchase behavior data, and browsing behavior data of each region, and determining the execution order of the marketing strategy set based on the optimal matching principle;

[0012] Generate a recommended product list for each region based on a preset marketing cycle, user types, and product inventory information for each region, and generate a promotion activity list for each region based on preset promotion rules, user types, and product inventory information for each region;

[0013] Based on the execution order of the marketing strategy set, the recommended product list, promotion activity list and user data set of each region are associated to generate a precision marketing plan for the rural e-commerce platform.

[0014] Preferably, the user data of the rural e-commerce platform is divided based on the user's geographic location information and consumption records to obtain user data sets for each region, including:

[0015] For the current area, the user data boundary of the current area is determined based on the historical total consumption of the current area, the historical total consumption of adjacent areas, and a preset area division threshold;

[0016] extracting a user data subset of the current area according to the user data boundary and a preset time window;

[0017] The user data including consumption records and browsing behaviors in the user data subset is determined as the user data set of the current region.

[0018] Preferably, determining the user type in each region based on the purchase frequency and product preference of the user data set in each region includes:

[0019] For a current region, if the purchase frequency of the user data set in the current region is greater than a preset frequency threshold, determining the current region as a high-activity region;

[0020] If the purchase frequency of the user data set in the current area is not greater than the preset frequency threshold, classifying the product preference data in the current area based on a first preset clustering algorithm to obtain preference data for each product category, and determining a central feature of the preference data for each product category;

[0021] Determining the consumption potential level of the current area according to the geographical location information of the current area;

[0022] When the matching degree between the central feature of the preference data of each commodity category and the preset commodity category is greater than a preset matching threshold, determining the current area as a targeted demand area;

[0023] When the matching degree is not greater than the preset matching threshold, the current area is determined to be a common demand area.

[0024] Preferably, the user data set further includes user portrait data and social interaction data. The user types in each region include high-value users, potential users, and lost users. Based on the user types, purchasing behavior data, and browsing behavior data in each region, a marketing strategy set for each region is generated, including:

[0025] For the current region, clustering the social interaction data of the current region based on a second preset clustering algorithm to obtain data of each user group; for the current user group data, filtering the current user group data based on the active duration and interaction frequency of each user in the current user group data to obtain a target user subset of the current user group data, and determining interest tags for the current user group data based on the average of the consumption preferences and product browsing records of the target user subset;

[0026] According to whether the user type in the current area is a high-value user;

[0027] Preferably, after determining whether the user type in the current area is a high-value user, the method further includes:

[0028] If the user type in the current area is a high-value user, the interest tags of the user group data are matched with the characteristics of the inventory products to generate a personalized recommendation strategy;

[0029] If the user type in the current area is a potential user, generating a targeted promotion strategy based on preset promotion rules and product discount information;

[0030] If the user type in the current area is a lost user, a user recall strategy is generated according to historical consumption intervals and product recall rules.

[0031] Preferably, after generating the marketing strategy set for each region, the method further includes:

[0032] Updating the user data sets of each region based on the real-time user behavior data;

[0033] recalculating the user type of each area according to the updated user data set;

[0034] The execution order of the marketing strategy set is dynamically adjusted according to the recalculated user type.

[0035] Preferably, after associating the recommended product lists, promotion activity lists, and user data sets of each region based on the execution order of the marketing strategy set to generate a precision marketing plan for the rural e-commerce platform, the method further includes:

[0036] Verify whether the matching degree between the recommended product list in the precision marketing plan and the user dataset meets the preset compliance conditions;

[0037] If not, the recommended product list is modified according to the preset fault tolerance rules.

[0038] Preferably, determining the consumption potential level of the current area according to the geographical location information of the current area includes:

[0039] Calculating a commodity accessibility index for the current region based on historical logistics data and regional infrastructure distribution;

[0040] Calculating the digital penetration rate of the current area based on population density and mobile terminal coverage;

[0041] The consumption potential level is determined based on the weighted sum of the commodity accessibility index and the digital penetration rate, and the level is divided into three levels: high, medium and low.

[0042] Preferably, when generating a personalized recommendation strategy, the method further includes:

[0043] Extract high-frequency keywords from the social interaction data of the target user subset and associate them with product description text to generate semantic matching tags;

[0044] Filter candidate products in inventory that meet a preset semantic similarity based on the semantic matching tags;

[0045] Priorities are determined based on the inventory turnover rate and user click-through rate of candidate products, and a weighted recommendation sequence is generated.

[0046] Preferably, the dynamically adjusting the execution order of the marketing strategy set includes:

[0047] Real-time monitoring of user stay time on product details pages and add-to-cart rates in each region;

[0048] If the duration of stay is lower than a preset threshold and the add-to-cart rate decreases, the priority of the promotion activity list is increased;

[0049] If the length of stay is higher than a preset threshold and the add-to-purchase rate is stable, the priority of the recommended product list is maintained.

[0050] Compared with the prior art, the present invention has the following beneficial effects:

[0051] In terms of precise user segmentation, this method meticulously segments users of rural e-commerce platforms based on multi-dimensional data such as their geographic location, consumption history, purchase frequency, and product preferences. By setting parameters such as preset frequency thresholds and matching thresholds, combined with a clustering algorithm, it accurately categorizes users in each region into high-activity areas, targeted demand areas, and general demand areas, as well as different types of users: high-value users, potential users, and churned users. This precise user segmentation enables e-commerce platforms to gain a deep understanding of the needs and characteristics of users in different regions and types, providing a solid foundation for developing targeted marketing strategies. For example, for high-value users in high-activity areas, personalized recommendations can be provided based on their interest tags and consumption preferences to further enhance their purchase intention and loyalty. For potential users, targeted promotional strategies can be used to stimulate their consumption potential. For churned users, user recall strategies can be used to attract them back to the platform, effectively improving the refinement of user management.

[0052] To ensure the effectiveness and efficiency of marketing strategies, we generate tailored marketing strategy sets based on user types and regional characteristics, and determine the execution order based on the principle of optimal matching. For high-value users, personalized recommendation strategies are generated by matching interest tags with inventory product characteristics. High-frequency keywords from social interaction data are extracted to generate semantic matching tags, which are then filtered and prioritized to ensure recommendations are more aligned with user needs. This improves the accuracy and relevance of recommendations, thereby increasing user conversion rates. For potential users, we generate targeted promotion strategies based on pre-set promotion rules and product discount information, precisely stimulating their purchasing behavior and effectively increasing their user activity and spending. For lapsed users, we generate user recall strategies based on historical purchase intervals and product recall rules, helping to reactivate these users and expand the platform's user base and market share. Furthermore, by monitoring real-time behavioral data such as user dwell time on product detail pages and add-to-cart rates, we dynamically adjust the execution order of marketing strategy sets, enabling timely optimization based on market feedback, ensuring that marketing strategies consistently achieve optimal results and enhancing the flexibility and adaptability of marketing campaigns.

[0053] In terms of considering regional differences and optimizing resource allocation, the method fully accounts for factors such as the geographical location, infrastructure distribution, population density, and mobile terminal coverage of different rural areas. By calculating the product accessibility index and digital penetration rate, the consumption potential level of each region is determined, and a list of recommended products and promotional activities is generated accordingly. In high-consumption potential areas with strong logistics and distribution capabilities and high digital penetration rates, products with high requirements for delivery timeliness and digital services can be recommended, and a variety of promotional activities can be implemented. In areas with lower consumption potential, the focus is on recommending products with high cost-effectiveness and practicality to avoid wasting marketing resources due to regional infrastructure limitations. This marketing strategy based on regional differences optimizes the allocation of marketing resources, improves resource utilization efficiency, and enhances the shopping experience of users in different regions.

[0054] In terms of data-driven dynamic optimization, by updating user datasets in real time, recalculating user types, and adjusting the order of marketing strategy execution, the entire marketing process forms a closed-loop dynamic optimization system. E-commerce platforms can promptly capture changing trends in user behavior, such as fluctuations in user interest in certain product categories or shifts in consumer habits, and quickly respond by adjusting marketing plans. This data-driven dynamic optimization mechanism ensures that marketing plans remain synchronized with user needs and market changes, enhancing the competitiveness and adaptability of rural e-commerce platforms and helping them stand out in the fierce market competition.

[0055] The compliance and reliability of marketing plans are ensured by verifying whether the recommended product list matches the user dataset and, if not, making corrections based on pre-set fault-tolerance rules. This not only helps enhance user trust in the platform but also prevents user complaints and brand damage caused by inappropriate recommendations, ensuring the healthy and sustainable development of rural e-commerce platforms. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 This is a working principle diagram of the rural e-commerce precision marketing method based on big data according to the present invention;

[0057] Figure 2 Flowchart for user data filtering based on region boundaries and time windows;

[0058] Figure 3 A flowchart for classifying user types based on purchase frequency and product preferences;

[0059] Figure 4 A flowchart generated for a differentiated marketing strategy based on user type;

[0060] Figure 5Flowchart showing consumption potential assessment based on logistics and digital penetration. DETAILED DESCRIPTION

[0061] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0062] See also Figure 1-Figure 5 The present invention relates to a rural e-commerce precision marketing method based on big data, and the specific implementation steps are as follows:

[0063] The system first obtains user data from rural e-commerce platforms, including geographic location information (such as province, city, county, and village levels), purchasing behavior data (such as purchase time, product category, and purchase amount), and browsing behavior data (such as type of browsed products, duration of stay, and number of clicks). For each region, the system determines the user data boundary based on the historical total consumption of the region, the historical total consumption of adjacent regions, and a preset regional division threshold (such as a 30% difference in total consumption as the boundary). For example, if the historical total consumption of the current region A is 1 million yuan and that of adjacent region B is 600,000 yuan, and the preset threshold is 30%, the user data range is divided according to the geographical boundaries where the consumption data difference is obvious (such as township administrative boundaries). Next, the system extracts a subset of user data for the region based on a preset time window (such as the past 12 months) and determines the user data set for the current region whose consumption records and browsing behavior are both within this subset.

[0064] The system analyzes the purchase frequency and product preferences of user datasets in each region. If the purchase frequency in a region exceeds a preset threshold (e.g., an average of two purchases per person per month), it is identified as a high-activity region. If it does not exceed the threshold, the product preference data is classified using a first preset clustering algorithm (e.g., K-means) to obtain preference data for each product category (e.g., agricultural products, daily necessities, home appliances, etc.), and the central features of each category preference data (e.g., major product categories, price range) are calculated. The system also calculates the product accessibility index based on historical logistics data (e.g., delivery time, number of covered outlets) and regional infrastructure distribution (e.g., road grade, express delivery station density). It also calculates digital penetration based on population density and mobile terminal coverage, and uses a weighted sum of these two factors to determine the consumption potential level (high, medium, or low). If the central features of the product category preference data match a preset product category (e.g., government-supported specialty agricultural products) above a preset matching threshold (e.g., 80%), the region is identified as a targeted demand region; otherwise, it is considered a general demand region.

[0065] The system generates marketing strategies based on user types, purchasing behavior, and browsing data for each region. For example, high-activity regions prioritize personalized recommendations, while targeted demand regions prioritize category-specific promotions. The execution order of marketing strategies is determined based on the optimal match principle (e.g., prioritizing user preferences and strategy coverage), such as prioritizing strategies with a large user base.

[0066] The system generates a list of recommended products for each region based on preset marketing cycles (e.g., quarterly), user types, and product inventory information. For example, products with high inventory turnover and high pageviews are prioritized in high-activity regions; general-purpose products are recommended in areas with average demand. Furthermore, a list of promotional activities is generated based on preset promotional rules (e.g., purchase discounts), user types, and inventory information, such as limited-time discounts for overstocked products in low-activity regions. Following the execution order of the marketing strategy set, the system links each region's recommended product list and promotional activity list with the user dataset to form a precise marketing plan for the rural e-commerce platform. The plan covers each region's specific marketing actions, timelines, and data support to ensure that the strategy aligns with user needs.

[0067] The present invention will be further described below in conjunction with Examples 1 to 5:

[0068] Example 1:

[0069] In the user data area division step, the system optimizes boundary determination, data subset extraction, and data set generation. The specific implementation is as follows:

[0070] The system first obtains raw user data from rural e-commerce platforms, including the geographic location information (province, city, county, township, and village levels) provided during user registration, as well as user purchase behavior data (such as purchase time, product name, product category, quantity, and amount), and browsing behavior data (such as product type, browsing timestamp, page duration, number of clicks, and whether items were added to favorites or added to cart). This data is stored in the platform's database and accessed by the system in real time or periodically through a data interface.

[0071] For the region currently being processed, the system begins determining the user data boundaries for that region. Traditional regional divisions may be based solely on administrative boundaries or simple geographic ranges, but this implementation incorporates multi-dimensional data for comprehensive judgment. First, the system obtains the historical total consumption of the current region. This data is the sum of the purchase amounts of all users in the current region over the past period. It also obtains the historical total consumption of adjacent regions. Adjacent regions refer to administrative regions or custom geographic units that border the current region. The preset regional division threshold is a pre-set ratio used to determine whether the consumption difference between the current region and adjacent regions is significant. For example, it can be set to 30%. When the difference between the historical total consumption of the current region and the adjacent region exceeds this threshold, the system deems that there is a significant difference in consumption characteristics between the two regions and uses the geographical boundary between them as a reference for the user data boundary. For example, if the historical total consumption of the current region A is 1 million yuan and the historical total consumption of the adjacent region B is 600,000 yuan, the difference is 40% (exceeding the preset 30% threshold). In this case, the system will prioritize the administrative boundary between regions A and B (such as the township boundary) as the initial basis for user data boundary division.

[0072] On this basis, the system further introduces geo-fencing technology to improve the accuracy of boundary division. Geofencing technology is based on the user's GPS positioning data or IP address positioning data, and accurately locates the user within a specific geographical range by drawing polygonal areas, circular areas or other custom-shaped areas on the map. For example, in mountainous rural areas with complex terrain, logistics and distribution may be affected by factors such as mountain roads and village distribution, resulting in large differences in user consumption behaviors in different villages within the same administrative area. At this time, the system can set up a geo-fence with a radius of 5 kilometers with the express delivery outlet as the center, and divide the users within the fence into an independent area, because the users within this range have similarities in logistics delivery timeliness, commodity accessibility, etc. At the same time, the system will also refer to the actual logistics distribution area, that is, the coverage of the distribution outlets divided by the express delivery company, to ensure that the user data boundary is consistent with the actual logistics service range, and avoid the failure of marketing strategies due to differences in distribution capabilities.

[0073] After determining the boundaries of user data, the system needs to extract a subset of user data for the current region. The preset time window is a configurable time range that limits the time span of the data. For example, it can be set to the past 12 months, the past 6 months, or the past 3 months. Considering that rural e-commerce consumption may have seasonal characteristics (such as higher consumer demand during the agricultural product harvest season), the system supports dynamic adjustment of the preset time window. For example, during the autumn harvest season for agricultural products (such as September to November), the system will automatically switch the preset time window to the past 3 months to capture the user consumption trends of the current season in a more timely manner; in the non-peak season, it can be restored to the time window of the past 12 months to obtain more comprehensive historical data. Through the dynamic adjustment of the time window, it is ensured that the extracted user data subset is timely and can reflect the real needs of users in the current region.

[0074] When extracting a subset of user data, the system cleans and pre-processes the original data. First, exclude abnormal data, such as invalid browsing records generated by test accounts (such as the same account frequently browsing the same product in a short period of time without purchasing it), false transaction data (such as purchase records generated by fake orders), etc. The identification of abnormal data can be achieved through a rule engine. For example, if a single account browses the same product more than 10 times within 24 hours and there is no collection, add-to-cart or purchase behavior, it will be judged as an invalid browsing record; or train an anomaly detection model through machine learning algorithms to automatically identify anomalies in the data. Secondly, the system deduplicates user data through IP addresses and device IDs to ensure that the behavioral data of the same user on different devices are merged into one record to avoid duplicate statistics. For example, when the same user logs in to the platform using a mobile phone and a computer, the system will integrate their behavioral data under the same user ID through account association to ensure the uniqueness and accuracy of the user data set.

[0075] After completing data cleaning and deduplication, the system will identify the user data whose consumption records and browsing behaviors are within the user data subset as the user data set of the current area. Specifically, for each piece of user data, the system checks whether its geographic location information falls within the user data boundary of the current area, and also checks whether its purchase time and browsing time are within the preset time window. Only user data that meets both conditions will be included in the user data set of the current area. For example, if user A's geographic location falls within the geographic fence of the current area A, and he has purchase records and browsing behaviors in the past three months, user A's data will be included in the user data set of area A; and although user B's geographic location belongs to area A, but he has no consumption or browsing behaviors in the past three months, his data will not be included in the current data set, but will be retained in the historical data for subsequent analysis.

[0076] The system also supports manual adjustment and verification of user data sets. Operations personnel can view the user data boundaries and data set contents of each region through the backend management interface and manually adjust any unreasonable divisions, such as merging adjacent small regions or splitting up large regions with significant consumption differences. The system also generates a data verification report, displaying metrics such as the size of the user data set, data integrity (such as the missing rate of each field), and data consistency (such as the match between geographic location and consumption behavior) for operations personnel to review and ensure that the quality of the user data set meets the requirements of precision marketing.

[0077] Example 2:

[0078] When determining regional user types, the system integrates user portrait data and social interaction data to further refine user types into high-value users, potential users, and churned users. The specific implementation is as follows:

[0079] The system first extracts the user data set for the current region from the rural e-commerce platform database, including purchasing behavior data, browsing behavior data, user profile data, and social interaction data. User profile data includes basic information provided by users during registration (such as age, gender, occupation, income level, and family size), real-name authentication information (such as ID card address, which can assist in verifying geographic location), and user-defined tags (such as "mom" and "farmer"). Social interaction data includes user behavior records such as comments, likes, and shares within the platform. For example, users' evaluations of specific products, replies to other users' comments, records of sharing product links on social media platforms such as WeChat and Weibo, and the frequency and content of topic discussions within the platform community.

[0080] For the current region, the system performs cluster analysis on the social interaction data using a second preset clustering algorithm. This second preset clustering algorithm can use the DBSCAN (Density-Based Spatial Clustering with Noise) algorithm. Based on the concept of density connectivity, this algorithm effectively discovers clusters of arbitrary shapes in a dataset and is suitable for handling uneven distribution of user groups in social interaction data. The specific steps are as follows: First, each piece of social interaction data is converted into a feature vector. Features include user ID, interaction type (e.g., comment, share, like), interaction object (e.g., product ID, user ID), interaction timestamp, and keywords in the interaction content. Then, algorithm parameters are set, such as the neighborhood radius ε and the minimum number of samples, MinPts (for example, ε is set to a spatial distance threshold encompassing 10 feature dimensions, and MinPts is set to 5). The algorithm searches the feature space for densely reachable sample points, forming distinct clusters, each corresponding to a user group. For example, if there are 100 users in a certain area, among which users A, B, C, D, and E frequently interact in the comment section of the same type of goods (such as agricultural products) and share the links of this type of goods to the same WeChat group many times, then the DBSCAN algorithm will cluster these 5 users into a group, defined as the "agricultural products attention group."

[0081] After clustering is completed, the system filters the data of each user group to determine the target user subset. The screening criteria are based on the user's active time and interaction frequency: the active time is measured by the number of times the user logs into the e-commerce platform in the past 30 days and the total online time. For example, the number of logins ≥ 3 times and the total online time ≥ 2 hours are set as the active standard; the interaction frequency is measured by the total number of comments, likes, shares and other behaviors of the user on the platform in the past 30 days. For example, the number of interactions ≥ 5 times is set as the high-frequency interaction standard. The system traverses the user data in each user group and only retains users who meet the conditions of active time and interaction frequency at the same time to form the target user subset. For example, among the 10 users in the "agricultural product attention group", 6 users have logged into the platform more than 4 times in the past 30 days, with a cumulative online time of 3 hours and 8 interactions. These 6 users will be screened as the target user subset, and the remaining 4 users with insufficient activity or interaction frequency will not be included in the analysis for the time being.

[0082] The system calculates the average consumption preferences and product browsing history of the target user subset to generate interest tags. Consumption preference analysis includes statistics on the most frequently purchased product categories, average purchase amounts, and frequently purchased brands within the subset. For example, six users in the target user subset purchased agricultural products 15 times in the past three months, with an average purchase amount of 80 yuan per purchase, and frequently purchased local brands of rice and cooking oil. Their consumption preferences can be summarized as "frequent purchases of local agricultural products, with a preference for mid- to low-priced products." Calculation of the average product browsing history includes statistics on the average browsing time, average number of clicks, and the rate of favorites or add-to-carts for each product category. For example, the average browsing time for agricultural products for this subset of users is 5 minutes per item, with an average of 3 clicks per item and a favorite rate of 20%. However, the average browsing time for daily necessities is only 2 minutes per item, indicating that their interest in agricultural products is significantly higher than that in other categories. Based on consumption preferences and browsing history, the system generates specific interest tags, such as "local agricultural product enthusiasts" and "cost-effective food consumers". The tag generation process can be combined with natural language processing (NLP) technology to extract keywords from consumption data and browsing content and perform semantic integration.

[0083] After generating interest tags for user groups, the system further determines user types by combining data such as purchase frequency and consumption potential level. Purchase frequency is measured by the average number of purchases per person in the current regional user dataset. For example, a monthly average purchase of ≥2 times per person is considered high frequency, while a monthly average purchase of <1 time is considered low frequency. The consumption potential level is calculated by taking the weighted sum of the product accessibility index and digital penetration rate. The product accessibility index is calculated based on historical logistics data (such as average delivery time, express delivery network coverage density) and regional infrastructure distribution (such as rural road grade and whether there are paved roads). The digital penetration rate is calculated based on population density (such as the number of residents per square kilometer) and mobile terminal coverage (such as the proportion of smartphone users in the region). The weighted sum of the two divides consumption potential into three levels: high, medium, and low.

[0084] The specific rules for determining user types are as follows:

[0085] High-value users: Users are considered high-value users if their spending exceeds the regional average, their active time is long (e.g., ≥5 logins in the past 30 days), their interactions are high (e.g., ≥10 comments and shares), and their region's spending potential is rated "high" or "medium." These users typically possess strong spending power and platform stickiness, making them a priority for platform maintenance. For example, User C, 40 years old with a high income, has made purchases of 2,000 yuan (1,500 yuan above the regional average) in the past three months, logged into the platform eight times, shared product links 12 times, and lives in an area with smooth roads and next-day express delivery (high spending potential), making him a high-value user.

[0086] Potential users: Users who browse frequently (e.g., browsed 10 or more items in the past 30 days) but made few purchases (e.g., only one purchase in the past three months) and whose active time and interaction frequency are moderate (e.g., 2-4 logins, 3-5 interactions) are considered potential users. These users have some interest in the platform but haven't yet established consistent purchasing habits, requiring marketing strategies to stimulate their spending. For example, user D, 25 years old, browsed 15 items, including daily necessities and appliances, in the past 30 days, spending a total of 45 minutes on the platform. However, he only made one daily necessities purchase, logged into the platform three times, and left two comments. Given his region's medium spending potential, he is considered a potential user.

[0087] Lost Users: If a user hasn't logged in to the platform and hasn't made any purchases for a preset period (e.g., 90 days without logging in or spending), they are considered lost. The system monitors the user's last login time and purchase timestamp to determine this. For example, if user E last logged in 120 days ago and made a purchase 150 days ago, with no interaction during that time, they are considered lost and require reactivation through a recall strategy.

[0088] In the process of determining user types, the system supports multi-dimensional data cross-validation. For example, for suspected high-value users, the system will further verify indicators such as the repurchase rate and return and exchange rate of their historical orders to exclude users who made large one-time purchases but had no subsequent active behavior; for potential users, the system will analyze the match between the products they browsed and the hot-selling products in the region to determine whether their needs actually exist. In addition, the system automatically updates user type labels every week to ensure that user status is adjusted in a timely manner as behavior changes. For example, if a potential user completes multiple purchases in a certain week and the frequency of interaction increases, the system will reclassify them as a high-value user or active user in the next update.

[0089] By integrating multi-dimensional data and cluster analysis, the system achieves refined stratification of rural e-commerce users, avoiding the flaws of single-dimensional classification based solely on purchase frequency. This classification approach more accurately identifies the characteristics and needs of different user groups, providing precise target guidance for the subsequent development of differentiated marketing strategies, ensuring that marketing resources are allocated to the most valuable user groups and improving overall marketing efficiency.

[0090] Example 3:

[0091] When generating a marketing strategy set, the system implements differentiated strategies based on user type (high-value users, potential users, churned users). The specific implementation methods are as follows:

[0092] The system first obtains the user type determination results and corresponding user group data for the current area. For high-value user groups, the system focuses on improving their loyalty and consumption frequency, and meets their precise needs through personalized recommendation strategies. The specific steps are as follows: The system matches the interest tags of each user group (such as "organic agricultural product enthusiasts" and "high-end home appliance consumers") with the characteristics of inventory products (including product category, brand, function, price range, origin, etc.). For example, the interest tag of a high-value user group is "local handicraft enthusiasts", and the inventory products include bamboo baskets and embroidered handkerchiefs made by local craftsmen. The system identifies related products through keyword matching (such as "local" and "handmade") to form an initial matching product pool.

[0093] Based on this, the system further extracts high-frequency keywords from the social interaction data of the target user subset. Using natural language processing (NLP) technology, it performs word segmentation and word frequency statistics on user comments and shared content. For example, from the comment "This handmade bamboo basket is very practical, and the weave is very exquisite," the system extracts keywords such as "practical," "weave pattern," and "exquisite" and calculates their frequency of occurrence. The system then semantically associates these high-frequency keywords with the product description text to generate semantically matching tags. For example, the high-frequency keyword "exquisite weave pattern" forms a semantic association with the product description terms "traditional weaving" and "delicate pattern," generating a semantically matching tag of "traditional craftsmanship and refinement."

[0094] The system uses semantic matching tags to filter candidate products from inventory that meet a preset semantic similarity. This preset semantic similarity is calculated using a cosine similarity algorithm. For example, if a threshold of 0.7 is set, products whose semantic similarity between their product description and their semantic matching tags exceeds 0.7 will be included in the candidate set. For example, if the product description of a bamboo basket reads "Using local bamboo, hand-woven by inheritors of intangible cultural heritage, with fine and uniform grain," and its cosine similarity with the semantic matching tag "Refined with traditional craftsmanship" is 0.85, the product will be selected as a candidate.

[0095] To determine recommendation priority, the system comprehensively considers the candidate product's inventory turnover rate and user click-through rate (CTR). The inventory turnover rate is calculated as "number of sales of a product over a certain period / average inventory level," reflecting the product's turnover rate. The CTR is calculated as "number of clicks on a product in a recommended list / number of impressions," reflecting user interest in the product. The system assigns weights to both inventory turnover rate and CTR (e.g., a weight of 0.6 for inventory turnover and 0.4 for CTR). The system then calculates a priority score for each candidate product by summing these scores. A higher score indicates a higher recommendation priority. For example, if candidate product A has an inventory turnover rate of 2 times per month and a CTR of 15%, its score is 2 × 0.6 + 15 × 0.4 = 7.2. If product B has a turnover rate of 1.5 times per month and a CTR of 20%, its score is 1.5 × 0.6 + 20 × 0.4 = 8.9. Therefore, product B is given a higher recommendation priority than product A. Finally, the system generates a weighted recommendation sequence based on the priority scores and displays them to high-value users through channels such as recommended placement on the e-commerce platform's homepage and app notifications.

[0096] For potential user groups, the system aims to stimulate their purchasing intent through targeted promotional strategies. The system generates promotional plans based on preset promotional rules and product discount information. Preset promotional rules include discounts (e.g., 20 yuan off for purchases over 100 yuan), discounts (e.g., 20% off storewide), buy-one-get-one-free (e.g., buy one, get one free), and limited-time offers (e.g., special offers within 24 hours). These rules can be flexibly configured based on different product types and sales targets. For example, limited-time discounts are preferred for fresh products such as seasonal fruits; for daily necessities, discounts or buy-one-get-one-free strategies can be adopted. Product discount information comes from promotional resources provided by suppliers or promotional activities independently set by the platform. The system synchronizes the discount status of inventory items (e.g., original price, current price, discount period, etc.) in real time.

[0097] Based on potential users' interest tags and browsing history, the system screens for promotional products that match their preferences. For example, if a potential user group has the interest tag "cost-effective daily necessities consumer" and has recently frequently browsed items such as laundry detergent and toilet paper, the system will select discounted items from these categories from its inventory (e.g., a certain brand of laundry detergent originally priced at 50 yuan is now priced at 39 yuan, a 22% discount) and generate a targeted promotion list. Promotional information reaches users through channels such as text messages, app pop-ups, and WeChat official accounts. Product details pages are also highlighted with eye-catching promotional logos (such as "Limited-Time Discount" and "Special Discount Zone") to guide users in placing orders. To increase conversion rates, the system also displays the remaining inventory quantity and discount countdown in promotional information to create a sense of urgency.

[0098] For lapsed user groups, the system develops a user recall strategy and reactivates users through historical consumption data and product recall rules. First, the system analyzes the historical consumption records of lapsed users, extracting their frequently purchased product categories, preferred brands, and information on the last product purchased. For example, a lapsed user purchased agricultural products such as fertilizers and seeds multiple times in the past year, and the last purchase was a certain brand of compound fertilizer. This means that their demand is concentrated in the agricultural sector. Product recall rules are set based on the user's historical consumption intervals and product relevance. For example, if the user's last purchase was more than 90 days ago, the system will push upgraded versions of their historically purchased categories or related products (such as new environmentally friendly fertilizers and smart irrigation equipment), along with recall coupons (such as a 50 yuan discount for purchases over 200 yuan).

[0099] The recall information is sent through SMS, email and other channels, and the content includes personalized greetings (such as "Dear user, you have not logged into the platform recently, we have prepared exclusive discounts for you"), historical preference product recommendations, coupon links, etc. In order to improve the success rate of the recall, the system will stratify lost users: for users with high consumption potential and large historical consumption amounts, higher-value coupons or exclusive customer service follow-up will be provided; for users who have lost due to platform experience problems (such as those with a complaint record), the platform improvement measures (such as optimizing logistics distribution, adding product categories) will be explained in the recall information. In addition, the system will record the feedback of lost users on the recall information (such as whether they clicked on the link, whether they logged into the platform), and for users who did not respond, different forms of recall content (such as changing recommended products, adjusting the form of discounts) will be sent again after a certain interval (such as 7 days) for a second recall.

[0100] During the strategy generation process, the system allows operators to manually adjust strategy parameters. For example, they can modify the matching rules between interest tags and product features, the discount strength of promotional rules, the validity period of recall coupons, and so on. The system also logs the strategy execution process, including the strategy trigger time, the number of users reached, and user response data (such as click-through rate and conversion rate). This allows operators to analyze the strategy's effectiveness and provide data support for subsequent optimization.

[0101] Through this differentiated strategy, the system provides precise marketing interventions for users at different value levels: high-value users receive recommendations tailored to their individual needs, strengthening their reliance on the platform; targeted promotions lower the decision threshold for potential users, encouraging them to complete their first or repeated purchases; and recall strategies reconnect lapsed users, recovering any lost consumer value. This tiered approach avoids a one-size-fits-all marketing model, ensuring efficient use of limited marketing resources and improving the overall marketing effectiveness and user lifetime value of rural e-commerce platforms.

[0102] Example 4:

[0103] After generating a marketing strategy set, the system dynamically optimizes the marketing strategy by monitoring user behavior data in real time, dynamically updating the user data set, and adjusting the strategy execution order. The specific implementation methods are as follows:

[0104] The system first establishes a real-time data monitoring mechanism, using tracking technology to collect behavioral data such as the length of time users spend on product detail pages, add-to-cart rates, conversion rates, and browsing paths in each region. This data is transmitted in real time to the data center in minutes. After cleaning and aggregation, it is stored in a real-time database for system analysis. For example, the length of time users spend on a product detail page is calculated using the timestamps of page load and close. The add-to-cart rate is the ratio of the number of users who add a product to the number of users who browse the product. The conversion rate is the ratio of the number of users who complete payment to the number of users who add a product to the product.

[0105] The system automatically triggers the user dataset update process every hour. For the current region, the system first obtains the latest user behavior data, including newly added browsing records, purchase orders, social interactions, etc., and then merges this data into the original user dataset. During the merging process, the system deduplicates duplicate data. For example, if the same user browses the same product multiple times in a short period of time, only the latest or most complete record will be retained. At the same time, the system will update the user's active status, consumption preferences and other tags. For example, if a user browses home appliances in the past hour, the "Home Appliances Attention" category will be added to their interest tag.

[0106] After the user data set is updated, the system recalculates the user type for each region. The calculation logic is consistent with the initial user type determination, that is, it combines multi-dimensional information such as purchase frequency, product preferences, consumption potential level, social interaction data, etc. For example, a certain area was originally determined to be a "general demand area", but through the latest data, it is found that the matching degree between its product preferences and the preset "agricultural product processing equipment" category has increased from 70% to 85% (exceeding the preset matching threshold of 80%), then the area will be reclassified as a "targeted demand area". For another example, some potential users in a certain user group have recently completed purchases, and their user type may be upgraded from "potential users" to "high-value users" or "active users".

[0107] Based on updated user types and real-time behavior data, the system dynamically adjusts the execution order of marketing strategy sets. The adjustment rules are based on the preset priority judgment logic, as follows:

[0108] Monitoring indicators are linked to strategy priorities: The system presets thresholds for duration of stay on product detail pages (e.g., 3 minutes) and change thresholds for add-to-cart rates (e.g., a 10% month-over-month decrease). If the average duration of stay on recommended products in a particular region falls below the preset thresholds, and the add-to-cart rate decreases month-over-month, indicating low user interest in the current recommendations, the system determines that the recommendation strategy is ineffective and automatically increases the priority of the promotional activity list. For example, if the average duration of stay on recommended clothing products in a particular region is 2 minutes (below the 3-minute threshold), and the add-to-cart rate drops from 15% last week to 10% (a 33% decrease), the system will bring forward the "Apparel Category Full-Discount Promotion" originally scheduled for the next day to the same day, and increase the exposure of the promotion entry on the platform homepage.

[0109] Stable performance maintains the strategic priority: If user dwell time exceeds the preset threshold and the add-to-cart rate is stable (e.g., month-over-month fluctuations within ±5%), the current recommendation strategy meets user expectations, and the system maintains the priority of the recommended products. For example, if the average dwell time for agricultural products recommended in a certain region is 5 minutes (above the threshold) and the add-to-cart rate is stable at around 20%, the system will continue to display the recommended products as planned and gradually expand the recommendations to include similar categories.

[0110] Dynamic Matching of Consumption Potential: For regions with rising consumption potential (e.g., from "low" to "medium"), the system will re-evaluate resource allocation for marketing strategies. For example, if a region's consumption potential rises from "low" to "medium" due to the addition of new express delivery outlets, the system will advance the execution order of the region's marketing strategy set and increase the variety of recommended products to accommodate the potential increase in consumer demand.

[0111] The system also supports intelligent predictive adjustments to the order of policy execution. Machine learning models (such as recurrent neural networks) analyze historical behavioral data and policy execution results to predict the conversion rate probabilities of different policy combinations in future time periods. For example, if the model analysis finds that a promotional strategy executed during weekday evenings in a certain region has a 20% higher conversion rate than the recommended strategy, the system will automatically adjust the policy execution order for that time period on subsequent weekdays, prioritizing the promotional activity list.

[0112] During dynamic adjustments, the system ensures a smooth transition in policy execution, preventing frequent changes from disrupting the user experience. For example, if a region needs to switch from a recommendation strategy to a promotion strategy, the system will gradually insert promotional information into the existing recommendations rather than directly replacing all recommendations. The system also records the reason and time of each adjustment, the policy sequence before and after the adjustment, and changes in user behavior data, creating a policy optimization log for operations personnel to review and analyze.

[0113] In addition, the system features a manual intervention interface, allowing operators to manually adjust the order of policy execution through the backend management interface. For example, if a sudden natural disaster causes unsold agricultural products in a certain region, operators can forcibly increase the priority of the agricultural product promotion strategy for that region, even if current user behavior data doesn't trigger automatic adjustment conditions. Manual intervention operations are recorded by the system and complement the automatic adjustment logic, ensuring that marketing strategies are both intelligently optimized based on data and responsive to specific business needs.

[0114] Through real-time monitoring, dynamic updates, and intelligent adjustments, the system achieves adaptive optimization of marketing strategy sets, enabling rural e-commerce platforms' marketing activities to promptly respond to changes in user demand and fluctuations in the external environment. This dynamic mechanism avoids the lag inherent in static strategies, ensuring that high-activity areas receive continuous, accurate recommendations, that potential user groups are promptly exposed to promotional incentives, and that users at risk of churn are effectively retained, thereby improving overall marketing efficiency and user satisfaction and achieving optimal resource allocation.

[0115] Example 5:

[0116] After generating a precision marketing plan, the system uses a verification mechanism to ensure that the recommended product list matches the user dataset and makes corrections if the conditions are not met. The specific implementation is as follows:

[0117] The system first verifies the compliance of the generated precision marketing plan, focusing on verifying the match between the recommended product list and the user dataset. The match calculation is based on the user's historical purchase and browsing behavior data, and is achieved through the following steps:

[0118] Extracting user-preferred categories: We count the frequency of each product category from the user's purchase and browsing behavior data to determine the user's most frequently preferred categories. For example, if agricultural products account for 40% of purchase records and 50% of browsing records in a user dataset for a certain region, then "agricultural products" is considered a most frequently preferred category in that region.

[0119] Calculate the overlap of recommended categories: Compare the product categories in the recommended product list with the user's frequently preferred categories, and calculate the proportion of overlapping categories as a matching indicator. The matching calculation formula is: in, Indicates the matching degree, A collection of categories representing the recommended product list. Represents the user's high-frequency preference category set, Indicates the number of intersection categories between the two.

[0120] Preset compliance condition judgment: The system presets the matching compliance threshold (such as 60%). , then the matching degree is determined to meet the requirements; if , then the fault-tolerance correction process is triggered.

[0121] When the matching degree does not meet the preset compliance conditions, the system will modify the recommended product list according to the preset fault tolerance rules. The modification process is as follows:

[0122] Supplementing users' browsed but unpurchased products: We filter out products viewed 2 or more times but not purchased from the user's browsing behavior data set and sort them in descending order by browsing duration to form a pool of candidate supplementary products. For example, if a user browsed "electric sprayer" multiple times but didn't purchase it, this product will be prioritized for inclusion in the candidate pool.

[0123] Referencing the preferences of similar user groups in the same region: Cluster analysis is used to identify other user groups with similar characteristics to users in the current region (such as location, spending potential, and interest tags). Product categories that these groups frequently purchase but are not currently listed in the recommended list are extracted and added to the recommended list. For example, if users in the adjacent region "Targeted Demand Area A" frequently purchase "agricultural drones," while this category is not listed in the current recommended list, and both have a "high" spending potential, then "agricultural drones" will be added to the recommended list.

[0124] Adjust the priority of recommended products: Recalculate the priority of existing recommended products and supplementary products. The priority is determined based on user click-through rate, inventory turnover rate and category matching. The calculation formula is: in, represents the priority score, Indicates user click rate (percentage value), represents the inventory turnover rate (times / month), Indicates category matching (the degree of overlap between the current category and the user's preferred category, as a percentage); 、 、 is the weight coefficient ( , for example, , , ). The higher the score, the higher the ranking of the product in the recommendation list.

[0125] After the correction is completed, the system recalculates the match score until the preset compliance conditions are met. For example, if the match score of the initial recommendation list is 50%, by adding the user's browsing but not purchase of "Agricultural Product Processing Equipment" (viewed 3 times, matching the category "Agricultural Products") and referring to the "Agricultural Fertilizer" category in similar areas, the match score is recalculated to 65%, meeting the compliance requirements.

[0126] During the verification and revision process, the system retains the original recommendation plan as a backup, allowing for comparative analysis of the differences between the revised and unrefined strategies. Operations personnel can access matching calculation details, the list of candidate supplementary products, and priority adjustment records through the backend, allowing them to perform manual intervention (such as removing inappropriate products) as needed. This entire process operates automatically, eliminating the need for real-time human monitoring and ensuring the effectiveness and stability of precision marketing plans.

[0127] Through this verification and correction mechanism, the system ensures that recommended products closely align with actual user needs, avoiding wasted marketing resources due to data bias or strategic errors. This process, combining data-driven automatic correction with manual intervention, improves recommendation accuracy while maintaining flexibility in strategic adjustments, providing reliable assurance for the efficient operation of rural e-commerce platforms.

[0128] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0129] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A rural e-commerce precision marketing method based on big data, characterized by: The method comprises: Divide user data of the rural e-commerce platform based on user geographic location information and consumption records to obtain user data sets for each region, including purchase behavior data and browsing behavior data; Determine the user type in each region based on the purchase frequency and product preference of the user data set in each region; Generating a marketing strategy set for each region based on the user type, purchase behavior data, and browsing behavior data of each region, and determining the execution order of the marketing strategy set based on the optimal matching principle; Generate a recommended product list for each region based on a preset marketing cycle, user types, and product inventory information for each region, and generate a promotion activity list for each region based on preset promotion rules, user types, and product inventory information for each region; Based on the execution order of the marketing strategy set, the recommended product list, promotion activity list and user data set of each region are associated to generate a precision marketing plan for the rural e-commerce platform.

2. The rural e-commerce precision marketing method based on big data according to claim 1 is characterized in that: The user data of the rural e-commerce platform is divided based on the user's geographic location information and consumption records, and the user data set of each region is obtained, including: For the current area, the user data boundary of the current area is determined based on the historical total consumption of the current area, the historical total consumption of adjacent areas, and a preset area division threshold; extracting a user data subset of the current area according to the user data boundary and a preset time window; The user data including consumption records and browsing behaviors in the user data subset is determined as the user data set of the current region.

3. The rural e-commerce precision marketing method based on big data according to claim 1 is characterized in that: Determining the user type in each region based on the purchase frequency and product preference of the user data set in each region includes: For a current region, if the purchase frequency of the user data set in the current region is greater than a preset frequency threshold, determining the current region as a high-activity region; If the purchase frequency of the user data set in the current area is not greater than the preset frequency threshold, classifying the product preference data in the current area based on a first preset clustering algorithm to obtain preference data for each product category, and determining a central feature of the preference data for each product category; Determining the consumption potential level of the current area according to the geographical location information of the current area; When the matching degree between the central feature of the preference data of each commodity category and the preset commodity category is greater than a preset matching threshold, determining the current area as a targeted demand area; When the matching degree is not greater than the preset matching threshold, the current area is determined to be a common demand area.

4. The rural e-commerce precision marketing method based on big data according to claim 3 is characterized in that: The user data set also includes user portrait data and social interaction data. The user types in each region include high-value users, potential users, and lost users. Based on the user types, purchasing behavior data, and browsing behavior data in each region, a marketing strategy set for each region is generated, including: For the current region, clustering the social interaction data of the current region based on a second preset clustering algorithm to obtain data of each user group; for the current user group data, filtering the current user group data based on the active duration and interaction frequency of each user in the current user group data to obtain a target user subset of the current user group data, and determining interest tags for the current user group data based on the average of the consumption preferences and product browsing records of the target user subset; According to whether the user type in the current area is a high-value user.

5. The rural e-commerce precision marketing method based on big data according to claim 4 is characterized in that: After determining whether the user type in the current area is a high-value user, the method further includes: If the user type in the current area is a high-value user, the interest tags of the user group data are matched with the characteristics of the inventory products to generate a personalized recommendation strategy; If the user type in the current area is a potential user, generating a targeted promotion strategy based on preset promotion rules and product discount information; If the user type in the current area is a lost user, a user recall strategy is generated according to historical consumption intervals and product recall rules.

6. The rural e-commerce precision marketing method based on big data according to claim 5 is characterized in that: After generating the marketing strategy set for each region, the method further includes: Updating the user data sets of each region based on the real-time user behavior data; recalculating the user type of each area according to the updated user data set; The execution order of the marketing strategy set is dynamically adjusted according to the recalculated user type.

7. The rural e-commerce precision marketing method based on big data according to claim 1 is characterized in that: After associating the recommended product lists, promotion activity lists, and user datasets of each region based on the execution order of the marketing strategy set to generate a precision marketing plan for the rural e-commerce platform, the method further includes: Verify whether the matching degree between the recommended product list in the precision marketing plan and the user dataset meets the preset compliance conditions; If not, the recommended product list is modified according to the preset fault tolerance rules.

8. The rural e-commerce precision marketing method based on big data according to claim 3 is characterized in that: Determining the consumption potential level of the current area according to the geographical location information of the current area includes: Calculating a commodity accessibility index for the current region based on historical logistics data and regional infrastructure distribution; Calculating the digital penetration rate of the current area based on population density and mobile terminal coverage; The consumption potential level is determined based on the weighted sum of the commodity accessibility index and the digital penetration rate, and the level is divided into three levels: high, medium and low.

9. The rural e-commerce precision marketing method based on big data according to claim 5 is characterized in that: The generation of the personalized recommendation strategy also includes: Extract high-frequency keywords from the social interaction data of the target user subset and associate them with product description text to generate semantic matching tags; Filter candidate products in inventory that meet a preset semantic similarity based on the semantic matching tags; Priorities are determined based on the inventory turnover rate and user click-through rate of candidate products, and a weighted recommendation sequence is generated.

10. The rural e-commerce precision marketing method based on big data according to claim 6 is characterized in that: The dynamically adjusting the execution order of the marketing strategy set includes: Real-time monitoring of user stay time on product details pages and add-to-cart rates in each region; If the duration of stay is lower than a preset threshold and the add-to-cart rate decreases, the priority of the promotion activity list is increased; If the length of stay is higher than a preset threshold and the add-to-purchase rate is stable, the priority of the recommended product list is maintained.

Images