E-commerce sales platform background data management method and system

Abstract

The invention relates to the technical field of e-commerce data processing, in particular to an e-commerce sales platform background data management method and system. The method comprises the following steps: S1, collecting a commodity dynamic data stream, a user behavior event stream and a promotion strategy stream in real time, and generating a standardized data stream through time sequence alignment; s2, constructing a dynamic coupling data cube; s3, executing a real-time decision: in response to the payment request, selecting an inventory distribution or replacement commodity pushing strategy based on a commodity-user coupling matrix value; in response to the resource overload state, triggering a resource scheduling strategy; and S4, dynamically adjusting calculation parameters of the coupling matrix according to decision execution feedback. By solving the problems of real-time standardization processing of multi-source heterogeneous data and real-time coupling of dynamic inventory and user behaviors, the data processing capacity, inventory distribution efficiency and recommendation accuracy of the platform are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of e-commerce data processing, and in particular to a backend data management method and system for an e-commerce sales platform. Background Art

[0002] In the backend data management of e-commerce sales platforms, efficient coordination of inventory, user behavior, and computing resources is key to achieving real-time business decisions. As business scale expands, existing technologies face technical bottlenecks in the following specific areas:

[0003] 1. Collaborative decision-making delays for multi-source heterogeneous data:

[0004] Product inventory data (typically stored in relational databases), user behavior logs (stored in NoSQL systems), and promotional strategies (cached in in-memory databases) all belong to independent systems, resulting in inefficient cross-database querying. Especially in high-concurrency scenarios like flash sales, existing technologies (such as offline data warehouses based on scheduled ETL) require minute-level responses and cannot meet real-time inventory allocation needs, resulting in overselling of popular products or sluggish sales of long-tail products. This problem is industry-wide: According to the "2023 China E-commerce Technology White Paper," the order processing delay rate due to data silos on the TOP100 platforms reached 12.7%.

[0005] 2. The static correlation between user behavior and inventory status is flawed:

[0006] Current mainstream solutions (such as recommendation systems based on collaborative filtering) rely on offline user portraits with an update cycle of more than 6 hours. When sudden traffic events occur (such as a product launched by an internet celebrity), the system is unable to dynamically perceive the coupling relationship between user urgency (such as repeatedly searching for similar products and adding them to cart) and inventory decay rate (such as hundreds of items reduced per second), resulting in a disconnect between recommendation results and real-time inventory. This defect is particularly prominent in promotion-sensitive categories (such as consumer electronics). The Q1 2024 financial report of a leading platform shows that the estimated loss of order conversion in such scenarios is 360 million yuan.

[0007] Therefore, there is an urgent need for an e-commerce sales platform backend data management method and system to solve the above problems. Summary of the Invention

[0008] Based on the above objectives, the present invention provides a method and system for managing backend data of an e-commerce sales platform. The method comprises the following steps:

[0009] S1: Collects product dynamic data streams, user behavior event streams, and promotion strategy streams in real time, and generates standardized data streams through time series alignment;

[0010] S2: Constructing a dynamic coupled data cube:

[0011] Generate a vector containing inventory status and sales trends in the product dimension;

[0012] Calculate the user urgency coefficient in real time at the user level, which correlates the frequency of user behavior with the inventory changes of added-to-cart items.

[0013] Calculate the regional service pressure value in the resource dimension, which is related to the user request volume and network resource status;

[0014] Construct a product-user coupling matrix, whose element values ​​are dynamically calculated based on the user's urgency coefficient, product inventory status, and promotion timeliness;

[0015] Construct a resource-commodity coupling matrix whose element values ​​relate to commodity demand forecasts, regional service pressure, and logistics costs;

[0016] S3: Execute real-time decisions:

[0017] In response to payment requests, select inventory allocation or alternative product push strategies based on the product-user coupling matrix value;

[0018] Respond to resource overload status and trigger resource scheduling strategies;

[0019] S4: Dynamically adjust the calculation parameters of the coupling matrix based on the decision execution feedback.

[0020] Correspondingly, an embodiment of the present invention also provides an e-commerce sales platform backend data management system, including a memory configured to store instructions, a processor configured to call the instructions from the memory and to implement an e-commerce sales platform backend data management method described in any one of the embodiments of the present invention when executing the instructions.

[0021] Beneficial effects of the present invention:

[0022] 1. The present invention collects commodity dynamic data streams, user behavior event streams, and promotion strategy streams in real time, and utilizes a timing alignment mechanism to uniformly convert multi-source heterogeneous data from relational databases, NoSQL systems, and in-memory databases into standardized data streams. This mechanism solves the problem of inefficient cross-database association queries due to different data sources. A sliding window of configurable length is used to reorder disordered data, and a watermark mechanism is used to process late data, minimizing response timeouts caused by data delays and ensuring real-time performance in a high-concurrency environment. A distributed logical clock algorithm (vector clock) is used to synchronize the timing identifiers of each node, further improving data consistency and collaborative processing efficiency.

[0023] 2. This invention dynamically reflects the relationship between user behavior and inventory changes by calculating the user urgency coefficient in real time. The calculation of the user urgency coefficient comprehensively considers the weighted value of behavior frequency, the inventory change rate of purchased items, and the user level correction factor, accurately capturing the coupling relationship between changes in user behavior and inventory status. This mechanism solves the problem of traditional solutions relying on offline user profiling and long update cycles, ensuring that the system can perceive sudden traffic events in real time.

[0024] 3. This invention dynamically calculates the relationship between products and users using a product-user coupling matrix. Taking into account factors such as user urgency, inventory status, and promotional timeliness, it recommends products with sufficient inventory that meet user needs in real time. During promotional activities, a dynamic decay mechanism for promotional timeliness ensures the timeliness of recommended products and reduces inaccurate recommendations caused by rapid inventory decay.

[0025] 4. In response to the dynamic changes in user behavior and product status, the system generates a real-time strategy for pushing alternative products based on user urgency, product inventory, and promotional timeliness. This strategy, through a comprehensive analysis of user behavior and product attributes, minimizes conversion losses caused by insufficient inventory or unsold products. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0027] Figure 1 is a flow chart of the steps of the method of the present invention;

[0028] Figure 2 Flowchart of the steps for calculating the element values ​​of the commodity-user coupling matrix in S2 according to the method of the present invention;

[0029] Figure 3 This is a flow chart of the steps for adjusting the coupling matrix parameters in method S4 of the present invention. DETAILED DESCRIPTION

[0030] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments. It is also noted that, to provide a more detailed description, the following embodiments are best and preferred embodiments, and those skilled in the art may employ alternative methods for implementing certain known technologies. Furthermore, the accompanying drawings are intended only to provide a more detailed description of the embodiments and are not intended to limit the present invention.

[0031] See Figure 1-Figure 3An embodiment of the present invention provides a backend data management method for an e-commerce sales platform. This method integrates multiple data sources (product inventory data, user behavior data, and promotional strategy data) and uses an efficient data collection engine to collect these data streams in real time. Product inventory data is typically stored in a relational database, user behavior data is stored in a NoSQL system (such as MongoDB or HBase), and promotional strategies are typically cached in an in-memory database (such as Redis).

[0032] To ensure synchronous processing of data from different data sources, a timestamp alignment mechanism is required. Real-time data streams are integrated using event-based time windows to ensure that asynchronous data streams are accurately aligned to the corresponding time points during data stream processing. This step uses a time synchronization algorithm to achieve data consistency and avoid query errors caused by data delays.

[0033] By collecting and aligning data streams in real time, the system can precisely and synchronously process data from different sources, significantly reducing latency in cross-database queries and ensuring data consistency and accuracy in subsequent steps. This approach provides fundamental support for flash sales and promotions in high-concurrency scenarios, ensuring the system's real-time responsiveness.

[0034] In the product dimension, the system constructs vectors based on the product's inventory status and sales trends. Inventory status includes the product's current inventory level, while sales trends are derived through time series analysis of historical sales data, dynamically reflecting any acceleration or deceleration in product sales. This data is stored in a dynamically updated product status data table, ensuring real-time reflection of the product's market performance.

[0035] At the user level, the user urgency coefficient is calculated in real time. This coefficient considers the relationship between the frequency of user behavior (such as the number of times a product is added to a cart or viewed) and the inventory status of the product. For example, if a user frequently browses a product and the inventory is rapidly decreasing, the user's urgency coefficient will increase, indicating a greater urgency to purchase. This coefficient can also be weighted to incorporate variables such as time and user level, making the calculation more personalized and dynamic.

[0036] The resource dimension calculates regional service stress, which is closely related to user request volume and network resource status. Specifically, the system calculates the service stress value for each region in real time based on the user request volume (such as page requests and order requests) and the region's network bandwidth and server load. When the stress value is too high, the system triggers resource scheduling strategies to prevent service unavailability due to overload.

[0037] The product-user coupling matrix describes the interactive relationship between products and users. Its elements are dynamically calculated based on the user's urgency coefficient, product inventory status, and promotional timeliness. The matrix's primary function is to optimize inventory allocation and product recommendations by analyzing user needs, combining real-time inventory changes with the impact of promotional activities.

[0038] The resource-product coupling matrix integrates factors such as product demand forecasts, regional service pressures, and logistics costs to dynamically adjust product supply strategies. Using this matrix, the system can identify which products face logistics pressures in certain regions and implement appropriate resource scheduling measures.

[0039] By constructing a dynamically coupled data cube, this paper achieves deep data integration between products, users, and resources, more accurately reflecting actual product demand, user purchasing urgency, and system resource load. This dynamically coupled cube provides a flexible, real-time data framework, enabling e-commerce platforms to make rapid decisions based on actual conditions and avoid overstocking or stockouts.

[0040] The system responds to payment requests based on the value of the product-user coupling matrix. Using this matrix, the system can quickly determine whether the product can be allocated inventory on demand or whether an alternative product should be recommended. In high-concurrency flash sales, the system optimizes inventory allocation strategies to ensure timely delivery based on user demand and reduce order cancellations due to insufficient inventory.

[0041] When the system detects resource overload in a certain area (such as network bandwidth or server load), it triggers a resource scheduling policy. Using the resource-product coupling matrix, the system can identify areas where product supply needs to be prioritized, dispatching more server resources or enabling more backup bandwidth to ensure a normal user experience.

[0042] Through a real-time decision-making mechanism, the present invention can flexibly adjust inventory allocation and resource scheduling strategies according to product inventory, user demand, and system resource status, ensuring that the system can operate stably under high concurrency and avoiding common problems such as system crashes and oversold inventory.

[0043] The system provides feedback based on the results of each decision (such as inventory allocation and resource scheduling), dynamically adjusting the calculation parameters in the coupling matrix. For example, if a product sells too quickly in a certain area, the system can automatically increase the inventory forecast weight for that product and adjust the element values ​​in the product-user coupling matrix to ensure that subsequent decisions are more in line with market demand.

[0044] Through dynamic adjustment of the feedback mechanism, the system can continuously optimize the decision-making model according to actual operating conditions, so that the platform can continuously improve the accuracy of decision-making and real-time response capabilities in long-term operations.

[0045] This invention leverages multi-dimensional technological innovations, integrating real-time data stream acquisition, dynamically coupled data cube construction, and real-time decision-making and feedback mechanisms, to achieve efficient data management for e-commerce sales platforms in high-concurrency, highly dynamic demand scenarios. By deeply coupling products, users, and resources, the platform can accurately grasp market changes, optimize inventory allocation, enhance user experience, and ultimately effectively reduce conversion losses and improve operational efficiency.

[0046] In one possible implementation, during the time alignment process, for data streams from multiple data sources, the order of the data may be disordered due to network delays or other factors in data collection. Therefore, the system designs a sliding window with a configurable length to reorder the disordered data. Specifically, the sliding window dynamically sets the window length based on historical statistical values ​​(such as the average or standard deviation of the data source delay). This ensures that during data collection, the platform can adjust the window length according to the real-time delay situation, thereby reducing disorder problems caused by network fluctuations.

[0047] To adapt to different scenarios and data sources, the sliding window length is dynamically configurable. The system calculates the appropriate window size based on each data source's historical latency statistics (such as average latency, maximum latency, and latency standard deviation). As data source latency changes, the window length adjusts accordingly, ensuring that the reordering operation maintains real-time performance while maximizing accuracy.

[0048] The sliding window mechanism effectively resolves data out-of-order issues in data streams, ensuring the correct order of data in real-time stream processing, particularly in environments with multiple data sources. Dynamically adjusting the window length optimizes data stream processing based on actual network conditions, improving system adaptability and stability and avoiding data omissions caused by overly small windows or delays and backlogs caused by overly large windows.

[0049] To address the issue of late data, this invention employs a watermarking mechanism. A watermark is a special marker placed in a data stream, typically indicating that all data prior to a certain point in time has arrived. By adding a watermark to each piece of data in the data stream, the system can identify and process data that arrives late due to factors such as network latency. In real-time stream processing, if data arrives late, the system will use the watermark information to process and compensate for the delay.

[0050] The system sets a maximum latency tolerance threshold, which is determined by the response timeout requirements defined in the platform's Service Level Agreement (SLA). The response timeout threshold typically refers to the maximum latency the system can tolerate. For example, the platform may stipulate a maximum latency of 500 milliseconds for user payment requests, exceeding which the platform must take additional remedial measures. Based on this timeout threshold, the system dynamically adjusts its data processing strategy to accommodate late data to the greatest extent possible and ensure the timeliness of the data stream.

[0051] The watermark mechanism ensures that even if data arrives late, the system still maintains data integrity and consistency, avoiding omissions or inconsistencies caused by data delays. By setting a maximum delay tolerance based on the platform's service-level agreement, the system ensures that it meets efficient service quality requirements and triggers appropriate remedial measures when delays are excessive, thereby optimizing the user experience.

[0052] In data stream processing, the system attaches a unified time sequence identifier to each piece of data. This identifier is generated by combining a physical clock timestamp and a logical clock sequence number. The physical clock timestamp typically refers to the actual time when the data was collected, while the logical clock sequence number is calculated based on the state of each node and the order of events in the distributed system.

[0053] Because the platform may consist of multiple distributed nodes, each with its own clock, clock drift and synchronization issues may occur. To address this, the system uses a vector clock algorithm to synchronize the logical clock sequences between nodes. The vector clock algorithm ensures consistency in the order of events across distributed nodes, ensuring that data is processed in accurate chronological order when transferred between multiple nodes.

[0054] By attaching a unified time sequence identifier to each piece of data, we can ensure that the time sequence of data flows is clear and consistent across the entire system. This approach effectively solves the problem of clock asynchrony in a distributed environment, ensuring that each node operates in the correct order when processing data, and avoiding data processing errors caused by time sequence disruptions.

[0055] This invention further optimizes the timing alignment process and innovates in the sliding window mechanism, watermark mechanism, maximum delay tolerance time, and unified timing identifiers, ensuring efficient and accurate backend data management for e-commerce platforms. The dynamic adjustment mechanism of the sliding window and the watermark mechanism effectively address data disorder and lateness, while the generation of a unified timing identifier and synchronization of the vector clock algorithm ensure data timing consistency in a distributed environment. These technical features collectively enhance the platform's real-time data processing capabilities, improve system stability and responsiveness, thereby optimizing the user experience and ensuring platform service quality.

[0056] In one possible implementation, when a user is shopping, the system first captures a sequence of user behavior events, including but not limited to keyword searches, product browsing, and add-to-cart actions. Each behavioral event is part of the user's decision-making process and reflects their interests and purchasing intent. By capturing these behaviors, the system can analyze user behavior patterns in real time and predict their immediate needs.

[0057] Different behavioral chain events reflect the user's purchase decision stage. Search keywords reflect the user's interest exploration stage, product browsing indicates the user's interest in a specific product, and the add-to-cart action signals the user's initial purchase intention. By accurately capturing and analyzing these behavioral chains, the platform can identify which user behaviors are more urgent, providing basic data for subsequent urgency coefficient calculations.

[0058] By capturing users' behavioral chains, the platform can fully understand their shopping needs and interests, accurately identifying their purchasing intentions. This helps the platform adjust marketing strategies in real time and improve user purchase conversion rates.

[0059] For user search events, the platform applies a decaying weight to each event based on the time of occurrence. This weight decay uses a negative exponential function, meaning that more recent search events receive a higher weight, more reflective of the user's current interests and needs. This negative exponential decay function reflects the gradual waning of user interest over time, thereby assigning a higher weight to recent searches.

[0060] Furthermore, the decay rate isn't fixed but dynamically adjusted based on the average duration of a user's historical sessions. This means that for some users, shorter session durations may indicate they make decisions more quickly, thus requiring a higher decay rate. Meanwhile, for users with longer session durations, the decay rate can be appropriately slowed.

[0061] The negative exponential decay model accurately reflects the urgency of users' current needs, especially for users with temporary interests, allowing for rapid response and appropriate weighting. Furthermore, dynamically adjusting the decay rate based on historical session duration helps provide more personalized predictions for different types of user behavior, improving the accuracy of the urgency coefficient.

[0062] When analyzing inventory changes for purchased items, the system uses a linear regression algorithm to fit an inventory change curve within a specific time window. By calculating the rate of change of inventory over time (i.e., the slope of the curve), the instantaneous rate of change of the item, i.e., the rate at which inventory is declining, can be determined.

[0063] The inventory decline rate is calculated by linear regression of inventory data over a recent period. This helps the platform determine the supply status of products and whether they are in a tight inventory state, thereby making a more accurate assessment of user urgency.

[0064] By capturing real-time inventory change rates, platforms can promptly identify products experiencing inventory shortages, accelerating sales and inventory management. Furthermore, combining this with a product's inventory change rate can improve the accuracy of the urgency factor, helping platforms deliver more compelling promotional messages to potential customers when inventory is limited.

[0065] The platform assigns a user level to each user based on their historical purchasing power and return rate. This level reflects their purchasing propensity and reliability. For example, a user with high purchasing power may be more inclined to make urgent purchases, while a user with a high return rate may be less interested in the product and therefore require a lower urgency factor. This correction factor, derived from historical data analysis, influences the final urgency factor.

[0066] The final user urgency coefficient is calculated by multiplying the weighted frequency of behavior, the inventory decline rate, and the user level correction factor. By integrating these factors, the platform can calculate a coefficient that reflects the user's current purchase urgency. This coefficient is used in further marketing and operational decisions, such as whether to make product recommendations to the user or prioritize specific products in promotions.

[0067] Incorporating a user level correction factor allows for more accurate predictions of which users are most likely to make a purchase at a given moment, avoiding blindly pushing products they aren't interested in. This personalized urgency factor calculation also improves the platform's marketing efficiency, reduces ineffective push notifications, and enhances user experience and conversion rates.

[0068] By comprehensively considering user behavior chains, weighted search events, inventory change rates, and user level correction factors, this paper provides a multi-dimensional method for calculating user urgency coefficients. This method not only accurately reflects users' purchasing urgency but also dynamically adjusts based on real-time inventory levels and user behavior patterns, thereby providing e-commerce platforms with more accurate sales forecasts and personalized recommendation strategies. This will help platforms improve marketing efficiency, optimize inventory management, and enhance user experience, ultimately boosting sales and user loyalty.

[0069] In one possible implementation, in this method, the user urgency coefficient reflects the user's current level of urgency to purchase a product. The calculation method for this coefficient has been described in detail above. A higher value indicates a more urgent desire to purchase a product and a stronger purchase intention.

[0070] A product's inventory balance refers to the real-time inventory level of that product within the user's geographic area. Fluctuations in inventory levels are a key factor in optimizing inventory management and improving the alignment of supply and demand on e-commerce platforms. A product with a high inventory balance indicates ample supply, reducing the user's purchasing urgency. A product with a low inventory balance indicates insufficient supply, leading to a more urgent purchasing demand.

[0071] Using the user urgency coefficient as the numerator and the product's inventory level as the denominator, this ratio calculation can, to a certain extent, quantify the relationship between the user's urgency to purchase a product and the product's supply. When inventory is tight, the user's urgency is more prominent; when inventory is sufficient, the urgency is relatively low.

[0072] The calculation of this ratio fully considers the urgency of user demand and the matching degree of inventory supply, helping the platform accurately predict user purchasing behavior and providing a scientific basis for subsequent marketing strategies. For example, when inventory is limited, the platform can prioritize recommending the product to users with high urgency, thereby increasing sales conversion rate.

[0073] The promotional time-to-decrease factor is dynamically adjusted based on the remaining time of the promotion and the historical order delay rate for the product. The historical order delay rate reflects the percentage of users who fail to complete their purchases within the scheduled timeframe during a promotion. This data can be derived from an analysis of historical order behavior, typically calculated using big data algorithms. A higher base decay coefficient indicates a tighter promotion deadline, which increases the likelihood of users feeling time pressure and a greater sense of urgency.

[0074] When the remaining time on a promotion falls below a certain threshold, the platform activates an accelerated decay mechanism, significantly adjusting the promotion's time-to-decrease factor. This threshold isn't fixed but is dynamically calculated based on the average decision time by product category. Decision times vary across product categories. For example, electronic products may require a longer decision time, while daily necessities may require a shorter one. By dynamically adjusting the promotion's time-to-decrease factor, the platform can more precisely control promotion strategies and increase users' sense of urgency.

[0075] This decay mechanism ensures that promotions can be flexibly adjusted based on actual conditions, improving promotion effectiveness. For certain products, particularly high-value items or those requiring a longer decision-making period, a longer promotion period results in slower decay. For products with shorter decision-making periods, a shorter promotion period triggers a faster decay mechanism, maximizing user urgency.

[0076] Regional correction weights adjust inventory balances, taking into account the supply and demand conditions within different geographic regions. Specifically, the regional supply-demand imbalance coefficient is determined by comparing the total inventory of a commodity within a specific geographic region to the 72-hour demand forecast. If inventory exceeds the demand forecast, it indicates sufficient supply and minimal inventory pressure in the region. If inventory is less than the demand forecast, it indicates that demand for the commodity in the region may exceed supply, resulting in tight inventory and urgent demand.

[0077] The regional correction weight adjusts the inventory balance of the product based on the supply-demand imbalance coefficient. If the inventory in a certain region is tight, the user urgency in that region will increase significantly, and the corresponding coupling matrix value will increase; otherwise, it will decrease.

[0078] By introducing regional adjustment weights, the platform can dynamically adjust product sales strategies based on supply and demand conditions in different regions. This approach not only helps optimize inventory management but also improves product supply chain efficiency, ensuring timely access to products during peak demand and avoiding sales losses due to insufficient inventory. Furthermore, regional adjustment weights help the platform identify regions with the most pressing user needs, thereby optimizing regional promotional activities.

[0079] The final product-user coupling matrix element values ​​are obtained based on the following calculation results:

[0080] Numerator: User urgency coefficient

[0081] Denominator: The inventory balance of the product in the user's geographic area

[0082] Promotional time decay factor: Dynamic adjustment based on historical timeout rate and remaining time

[0083] Regional correction weight: determined by the supply and demand imbalance coefficient

[0084] The final coupling matrix value is a comprehensive indicator used to measure the urgency between users and products, the matching degree between demand and supply, and the impact of promotional activities.

[0085] This calculation method comprehensively considers user urgency, product inventory status, promotional timeliness, and regional demand differences, providing more accurate product recommendations and sales forecasts. This not only optimizes marketing strategies but also improves inventory utilization, ensuring the platform can promptly meet user needs during peak demand, thereby increasing sales conversion rates and user satisfaction.

[0086] By comprehensively considering user urgency, inventory balance, promotional time decay factor, and regional correction weights, this method provides a precise way to calculate the product-user coupling matrix. This method can dynamically adjust inventory management, promotion strategies, and regional supply and demand matching, helping e-commerce platforms to more finely manage product sales, improve user experience, and increase sales.

[0087] In one possible implementation, in step S3, the products are first screened to ensure that the candidate products are from the same product category. This approach ensures that the type of the replacement product is consistent with the original product, avoiding the situation where irrelevant products are recommended due to category mismatch.

[0088] Items with inventory remaining above a safety threshold are further screened. The safety threshold is dynamically calculated based on the standard deviation of the item's sales cycle. The standard deviation reflects the fluctuation in product sales. A higher standard deviation indicates greater sales volatility and requires more inventory to accommodate demand fluctuations. Setting a safety threshold based on this standard deviation ensures sufficient inventory for recommended alternative items.

[0089] By screening inventory balances, we avoid recommending out-of-stock products to users as substitutes, ensuring the availability and supply stability of substitute products and improving the user's purchasing experience.

[0090] When calculating attribute matching, the system extracts multiple attributes from both the original and candidate products, including price range, brand level, and feature tag sets. These attributes reflect the core characteristics of the product and key factors that users care about.

[0091] The price segment matching degree is calculated by comparing the overlap between the original product's price range and the candidate product's price range. If the original product's price is in range A and the candidate product's price is in range B, and there is overlap between A and B, the matching degree is high; otherwise, it is low.

[0092] The function tag matching degree is calculated using the Jaccard similarity algorithm. The Jaccard similarity algorithm measures the matching degree by calculating the ratio of the intersection and union of the function tag sets of the original product and the candidate product. Specifically, assuming that the original product has tag set A and the candidate product has tag set B, the Jaccard similarity is:

[0093]

[0094] This algorithm can effectively measure the similarity of product functions and help screen out products that best meet user needs.

[0095] By calculating the matching degree of price segments and function tags, the system can screen out products that better meet user needs, ensuring that the recommended alternative products are consistent with the original products on multiple levels, thereby improving the accuracy of product recommendations and user satisfaction.

[0096] To generate the comprehensive ranking weight, first multiply the item-user coupling matrix value calculated in the previous step by the attribute matching degree. The item-user coupling matrix value reflects the urgency and demand matching between the user and the item, and can dynamically adjust the recommendation priority.

[0097] In addition to attribute matching, a real-time click heat correction value is also added. This value is calculated by counting the rate of change in visits to the product detail page within a unit of time. Click heat reflects the level of user attention to the product. A greater rate of change in clicks indicates that the product has received more attention during the current period and has higher promotion potential.

[0098] The final comprehensive ranking weight is calculated by multiplying the product-user coupling matrix value by the attribute matching degree, and then adding the click popularity correction value. This comprehensive ranking weight represents the priority of product recommendations. Products with higher weights are more closely aligned with user needs in terms of attributes and are receiving more attention during the current period, thus being recommended first.

[0099] By comprehensively considering the product-user coupling matrix, attribute matching, and real-time click heat correction, the system can finely sort alternative products, ensuring that recommended products not only meet user needs but also align with current market trends and user behavior. This approach helps improve the accuracy and effectiveness of product recommendations, increasing conversion rates and user purchasing experience.

[0100] Combining these steps, the system automatically generates a set of high-quality alternative product push strategies and recommends products that meet user needs. These alternative products are not only highly consistent with the original products in terms of category consistency and attribute matching, but also dynamically adjust based on inventory levels, sales cycles, and real-time clickthrough rate to ensure product availability and user purchasing needs.

[0101] Optimizing this strategy can effectively improve product recommendations on e-commerce platforms, enhance user experience and satisfaction, and reduce the likelihood of users abandoning purchases due to out-of-stock or unsatisfactory products. Furthermore, real-time clickthrough rate adjustments can keep pace with evolving user needs, improve product sales conversion rates, and boost overall platform sales growth.

[0102] By comprehensively considering factors such as product category, inventory level, attribute matching, and click-through rate correction, the S3 alternative product push strategy enables precise product recommendations, maximizing user satisfaction, improving sales conversion rates, and providing more optimized product recommendation solutions within e-commerce platforms. This strategy effectively enhances the intelligence of product recommendations and increases the market competitiveness of e-commerce platforms.

[0103] In one possible implementation, the system first records the absolute deviation between the actual inventory consumption rate and the predicted value. The predicted value is an estimated inventory consumption rate based on historical sales data and product characteristics, while the actual consumption rate is calculated based on real-time sales data. When the actual sales rate deviates from the predicted value, the system records the deviation and indicates its magnitude and direction.

[0104] When inventory consumption rates deviate in the same direction over consecutive time periods (i.e., consistently above or below the forecast), the system analyzes the trend of these deviations and, based on the proportion of the deviation, adjusts the demand forecast weights in the resource-product coupling matrix in the opposite direction. For example, if actual sales fall below the forecast for several consecutive days, the system will deem the forecast demand excessive and will dynamically adjust the demand forecast weight for that product in the coupling matrix accordingly.

[0105] By dynamically adjusting inventory forecast deviations, the system can more accurately reflect actual product demand and sales trends, thereby avoiding overproduction or overpurchasing, improving inventory management precision, and reducing the risk of overstocking and stockouts. The reverse-corrected coupling matrix more accurately reflects market demand, improving the accuracy and efficiency of product recommendations.

[0106] For each recommended alternative product, the system calculates the conversion rate by counting the percentage of click-through orders compared to the total number of recommended products. This conversion rate reflects users' actual interest in and purchase behavior for the recommended alternative product, effectively measuring the effectiveness of recommended products.

[0107] When the conversion rate of certain alternative products falls below a certain percentage of the historical average, the system adjusts the recommendation weight of the product based on the deviation in conversion rate. Specifically, the system reduces the strength of the product's category association, that is, reduces the coupling between the product and other products or users, to avoid continuing to recommend poorly performing products.

[0108] The magnitude of the reduction in product category association strength is piecewise linearly related to the conversion rate deviation. For example, if the conversion rate deviation is less than a certain critical value, the adjustment is smaller; when the deviation exceeds a certain threshold, the adjustment is larger. The critical ratio is determined by optimizing historical A / B testing data to ensure a more accurate and reasonable conversion rate adjustment process.

[0109] By providing feedback and adjusting the conversion rates of alternative products, the system can promptly identify inefficiently recommended products, reduce the impact of these products on users, and improve the accuracy of product recommendations. By adjusting the strength of product category associations, we can ensure that recommended products are more aligned with user needs and purchasing preferences, preventing low-conversion products from occupying recommendation resources and further improving the platform's sales conversion rate.

[0110] By combining feedback from inventory forecast deviations and conversion rates of alternative products, the system dynamically optimizes its product recommendation strategy. When a product's inventory forecast deviation is large or its conversion rate is low, the system automatically adjusts the weights in the resource-product coupling matrix to ensure that recommended products meet users' immediate needs while avoiding excessive recommendations of products with low conversion rates or that don't meet market demand.

[0111] By continuously optimizing critical ratio values ​​and adjustment rules through A / B testing, the system can make real-time strategy adjustments based on the changing trends of historical data, ensuring that each adjustment can be verified in actual operations and further refining the algorithm model.

[0112] This joint adjustment mechanism, based on real-time data feedback, ensures more intelligent and personalized product recommendations. By leveraging inventory and conversion rate feedback, the platform can adjust product recommendation weights in real time, optimizing the user shopping experience and improving sales efficiency. Furthermore, optimization based on A / B testing allows for continuous adjustment and refinement of recommendation strategies, enhancing the platform's long-term operational efficiency.

[0113] By introducing two adjustment mechanisms—inventory forecast deviation feedback and alternative product conversion feedback—the coupling matrix parameter adjustment in step S4 dynamically optimizes the product recommendation system and improves the platform's operational efficiency. Inventory deviation feedback accurately adjusts product demand forecasts to avoid inventory overstocks or stockouts. Alternative product conversion feedback monitors conversion rates, optimizes the push of inefficient products, and increases user purchase conversion rates. These optimization measures help improve the accuracy and precision of product recommendations, thereby boosting overall sales and user satisfaction on the e-commerce platform.

[0114] In one possible implementation, the resource scheduling strategy first calculates and obtains a regional service pressure value through real-time monitoring of service nodes. This value represents the load on a specific service node on the platform (e.g., a server or data center in a specific region), typically derived from a comprehensive assessment of metrics such as request volume, response time, and bandwidth usage.

[0115] To cope with fluctuations in platform service pressure, the system calculates a dynamic threshold to determine whether the current service load is too high. This dynamic threshold is calculated using the following method:

[0116] First, obtain the historical peak pressure value of the node, that is, the maximum load borne by the node within a period of history.

[0117] Then, a moving average of the historical peak pressure value is calculated, which can smooth the historical data and reduce the impact of volatility.

[0118] Finally, this moving average is multiplied by the traffic fluctuation coefficient for the current period. This coefficient is dynamically adjusted based on the actual traffic fluctuations. For example, if the platform traffic fluctuates drastically, the traffic fluctuation coefficient will increase, making the dynamic threshold more sensitive.

[0119] This dynamic threshold calculation method can timely adjust the tolerance of service pressure, making resource scheduling strategies more flexible and adaptable.

[0120] When the regional service pressure value continues to exceed the dynamic threshold, the system will start the hierarchical load reduction mechanism to effectively allocate and optimize resources to prevent service node overload and platform performance degradation. Hierarchical load reduction includes two stages:

[0121] When the service pressure value exceeds the threshold, the system will first convert static resources (such as pictures, videos, etc.) into low-resolution versions.

[0122] The resolution degradation ratio will gradually increase based on the magnitude of the pressure value exceeded. For example, the greater the pressure value exceeds the threshold, the higher the resolution degradation ratio will be, reducing the loading burden of static resources and optimizing system performance.

[0123] This downgrade strategy can effectively reduce the load on static resources, especially in high-concurrency scenarios, reducing bandwidth usage and alleviating server pressure. At the same time, this downgrade operation has a minimal impact on the user experience, as pages with dynamic content and high-value products are not immediately affected.

[0124] When static resource degradation cannot meet demand, the system enters the second stage, which is to expand the container replicas of high-value product pages in the product-user coupling matrix.

[0125] The system will sort products according to the coupling matrix value between products and users, give priority to high-value products (i.e. products with higher conversion rates and larger sales volumes), and increase the number of container copies of these product pages.

[0126] The number of replica expansions is positively correlated with the pressure growth rate. That is, when service pressure increases, the system will increase the number of replicas according to the speed and magnitude of the pressure growth, thereby sharing more traffic requests and ensuring the response speed and user experience of high-value products.

[0127] This container replica expansion mechanism ensures that high-value products maintain a high response speed and a good user experience even when the platform is under high pressure, maximizing platform sales performance. By dynamically expanding capacity based on the pressure value growth rate, it effectively avoids resource waste and enables rapid expansion operations when the system is under high pressure.

[0128] By combining a two-phase strategy of static resource degradation and replica expansion of high-value commodity containers, the system dynamically adjusts resource allocation when facing excessive regional service pressure, ensuring high platform availability and a positive user experience. Furthermore, based on dynamic threshold calculations, the scheduling strategy automatically adjusts to traffic fluctuations, avoiding over-reliance on fixed thresholds and improving the system's adaptability and flexibility.

[0129] This hierarchical load reduction and dynamic scheduling mechanism enables e-commerce platforms to flexibly respond to traffic fluctuations, improve system performance, ensure the continuous and stable operation of the platform, and maintain a good user experience even under high-pressure conditions.

[0130] In one possible implementation, the demand forecast is generated using a time series decomposition model, including the following specific steps:

[0131] First, use a time series decomposition model (such as the STL method) to decompose historical sales data into three parts: trend term, seasonal term, and random residual term.

[0132] Trend items reflect the long-term trend of sales over time, such as the gradual increase in product popularity.

[0133] The seasonal item reflects cyclical fluctuations and is usually related to factors such as months, seasons, and holidays.

[0134] The random residual term is other unexplained random fluctuations.

[0135] Based on the existing trend term, the slope of the trend term is modified by the change rate of coupling between each product and the target user group in the product-user coupling matrix. These change rates reflect fluctuations in product demand and changes in user interest. For example, if the purchase frequency of a particular product or user activity increases, the trend prediction will be improved accordingly.

[0136] The system converts the discount strength in promotional strategies into a seasonal fluctuation adjustment factor. The discount strength and seasonal fluctuation adjustment factor exhibit a nonlinear positive correlation: greater discount strength leads to greater seasonal fluctuations in demand. For example, increasing discount strength may lead to more dramatic fluctuations in product sales, necessitating adjustments to the seasonal fluctuation.

[0137] Quantification of logistics costs involves the following steps:

[0138] The system uses a Geographic Information System (GIS) routing interface to calculate the shortest path distance from the warehouse to the target user area in real time. This data is dynamically updated based on the actual geographic location and routing algorithm, ensuring that the calculated path distance accurately reflects the actual delivery route.

[0139] To more accurately calculate logistics costs, the system introduces a traffic correction factor. This factor is based on the traffic congestion levels (e.g., smooth, lightly congested, heavily congested, etc.) returned by the real-time traffic API and maps these congestion levels to numerical weights. For example, a smooth flow corresponds to a factor of 1, while severe congestion corresponds to a factor of 2.5, indicating increased transportation costs and time. By weighting the path distance, the system ensures more accurate quantification of logistics costs, reflecting the timeliness and cost fluctuations in actual delivery.

[0140] By combining time series analysis, user behavior analysis, and real-time logistics data, we have achieved efficient resource scheduling, accurate demand forecasting, and optimized logistics management in the backend data management of the e-commerce platform, providing strong data support for the platform's operational decisions, thereby improving overall operational efficiency and user satisfaction.

[0141] In one possible implementation, the inventory allocation strategy includes a dynamic threshold setting mechanism and cross-warehouse transfer routing. Implementing these strategies can optimize inventory management, improve supply chain efficiency, reduce inventory backlogs, and ensure timely order delivery.

[0142] Specifically, to accurately predict product inventory demand, the first threshold for inventory allocation is set based on the product's historical sell-out rate. The sell-out rate refers to the percentage of a product sold out within a specified time period, reflecting the intensity of demand for the product.

[0143] A higher sell-out rate indicates greater demand for the product, necessitating more inventory. To avoid stockouts, the first threshold is set low, indicating a higher sensitivity to inventory replenishment and prompt stocking. Conversely, if the sell-out rate is low, the inventory threshold can be set higher to reduce unnecessary inventory reserves.

[0144] The second threshold is negatively correlated with the regional inventory turnover rate. Turnover rate refers to the frequency with which inventory is updated within a specific area. By counting the frequency of inventory updates per unit time, we can reflect the speed of commodity turnover in that area.

[0145] When the turnover rate in a certain area is low, it indicates that the goods in that area are selling slowly, so the inventory threshold needs to be increased to avoid excessive goods being accumulated in that area. When the turnover rate is high, the inventory threshold can be appropriately lowered to ensure the liquidity of goods and avoid over-replenishment.

[0146] The inventory adequacy of each candidate warehouse is a key indicator for evaluating the priority of stock transfers. Warehouses with high inventory adequacy are prioritized for stock transfers to ensure that order demand can be met.

[0147] The time cost of transferring goods also plays a role in the calculation. By calculating the transportation time and transportation cost from the alternative warehouse to the target warehouse, the transfer can be completed as quickly as possible and the additional transportation costs can be reduced.

[0148] A warehouse's service pressure value reflects its order processing capacity and current load. When a warehouse is heavily loaded, its service pressure value is high. The system will select warehouses with more ample inventory and less pressure to transfer goods, avoiding over-exerting a single warehouse's resources.

[0149] The weighting coefficient is dynamically adjusted based on the historical transfer success rate. If a warehouse has a high historical transfer success rate, it means that its transfer routing and logistics plan are more stable and reliable. In this case, the weighting coefficient of this warehouse is larger, and it is preferred for transfers.

[0150] The system monitors the success and failure rates of transfers in real time and makes dynamic adjustments based on historical data. If a warehouse's transfer success rate decreases, its weighting factor will be appropriately reduced; if it decreases, its weighting factor will be increased, thereby optimizing the selection of cross-warehouse transfer routes.

[0151] The dynamic threshold setting mechanism automatically adjusts inventory thresholds based on historical product sell-out rates and regional inventory turnover rates, ensuring optimal inventory allocation for each product and region. This reduces the risk of overstocking and prevents stockouts of certain products due to insufficient inventory, improving inventory management accuracy.

[0152] By calculating the transfer priority of alternative warehouses and comprehensively considering multiple factors such as inventory adequacy, arrival time cost, and service pressure value, the system can select the best transfer path among multiple warehouses, avoid excessive concentration of resources between warehouses, and optimize collaborative operations between warehouses.

[0153] Optimizing cross-warehouse transfer routes not only improves transfer efficiency but also reduces logistics costs. By selecting warehouses with sufficient inventory and a high transfer success rate, the platform ensures an efficient and smooth transfer process, minimizing logistics delays and ensuring timely order delivery.

[0154] The system dynamically adjusts weighting coefficients based on historical inventory data and inventory changes, enabling the platform to quickly respond to changes in market demand and supply. This dynamic adjustment mechanism allows the platform to flexibly respond to demand fluctuations and emergencies, ensuring that goods are replenished to the appropriate warehouse in a timely manner, thereby improving service levels and user satisfaction.

[0155] The inventory allocation strategy proposed in this invention can effectively improve the inventory management efficiency of e-commerce platforms, optimize the scheduling process between warehouses, and reduce logistics costs, ultimately achieving faster and more accurate services for consumers.

[0156] Correspondingly, an embodiment of the present invention also provides an e-commerce sales platform backend data management system, including a memory configured to store instructions, a processor configured to call the instructions from the memory and to implement an e-commerce sales platform backend data management method described in any one of the embodiments of the present invention when executing the instructions.

[0157] The present invention encompasses any alternatives, modifications, equivalents, and solutions that fall within the spirit and scope of the present invention. To provide a thorough understanding of the present invention, specific details are described in detail below in connection with the preferred embodiments of the present invention, but those skilled in the art will be able to fully understand the present invention without these detailed descriptions. Furthermore, to avoid unnecessary confusion regarding the essence of the present invention, well-known methods, processes, procedures, components, and circuits have not been described in detail.

[0158] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A backend data management method for an e-commerce sales platform, characterized in that: The following steps are involved: S1: Collects product dynamic data streams, user behavior event streams, and promotion strategy streams in real time, and generates standardized data streams through time series alignment; S2: Constructing a dynamic coupled data cube: Generate a vector containing inventory status and sales trends in the product dimension; Calculate the user urgency coefficient in real time at the user level, which correlates the frequency of user behavior with the inventory changes of added-to-cart items. Calculate the regional service pressure value in the resource dimension, which is related to the user request volume and network resource status; Construct a product-user coupling matrix, whose element values ​​are dynamically calculated based on the user's urgency coefficient, product inventory status, and promotion timeliness; Construct a resource-commodity coupling matrix whose element values ​​relate to commodity demand forecasts, regional service pressure, and logistics costs; S3: Execute real-time decisions: In response to payment requests, select inventory allocation or alternative product push strategies based on the product-user coupling matrix value; Respond to resource overload status and trigger resource scheduling strategies; S4: Dynamically adjust the calculation parameters of the coupling matrix based on the decision execution feedback.

2. The backend data management method of an e-commerce sales platform according to claim 1, characterized in that: Timing alignment in S1 includes: Set up a sliding window of configurable length to reorder out-of-order data. The window length is dynamically set based on the historical statistics of data source delays. A watermark mechanism is used to process late data, and the maximum delay tolerance time is determined based on the response timeout threshold defined in the platform service level agreement; A unified timing identifier is attached to each piece of data. The identifier is generated by combining the physical clock timestamp and the logical clock sequence number. The logical clock sequence number is synchronized between distributed nodes using the vector clock algorithm.

3. The backend data management method of an e-commerce sales platform according to claim 1, characterized in that: The calculation of user urgency coefficient in S2 includes: Capture the user's current session's behavioral chain event sequence, including keyword search, product browsing, and add-to-cart operations; Calculate the behavior frequency weight: Apply decaying weights to search events with the same keyword based on the time of occurrence. The closer the time, the higher the weight. The decay function uses a negative exponential form, and the decay rate is adjusted based on the average duration of the user's historical sessions. Get the real-time inventory decline rate of added-to-cart items: Use linear regression to fit the inventory change curve within the recent time window, and use the slope value as the instantaneous rate of change; The final coefficient is obtained by multiplying the weighted behavior frequency with the inventory decline rate and then multiplying it by the user level correction factor. The user level is comprehensively determined by historical purchasing power and return rate.

4. The backend data management method of an e-commerce sales platform according to claim 1, characterized in that: The calculation of the element values ​​of the product-user coupling matrix in S2 includes: The user's urgency coefficient is used as the numerator, and the real-time inventory balance of the product in the user's geographical area is used as the denominator; Multiply by the promotion time decay factor: The basic decay factor is set based on the historical order timeout rate. When the remaining time of the promotion is less than the critical value, the accelerated decay mechanism is activated. The critical value is dynamically calculated based on the average decision time of the product category. Superimposed regional correction weight: A regional supply and demand imbalance coefficient is introduced to the inventory balance. This coefficient is determined by comparing the total inventory in the region with the 72-hour demand forecast value.

5. The backend data management method of an e-commerce sales platform according to claim 1, characterized in that: The strategies for pushing alternative products in S3 include: Pre-screen candidate products within the same product category whose inventory balance is greater than a safety threshold. The safety threshold is set based on the standard deviation within the product's sales cycle. Calculate attribute matching: Extract multi-dimensional attributes of original products and candidate products, including price segment, brand level, and function tag set; The price segment matching is calculated by comparing the overlapping ratio of the price ranges of the two; The matching degree of functional labels is determined by the Jaccard similarity algorithm based on the proportion of shared labels; Generate comprehensive ranking weights: multiply the product-user coupling matrix value by the attribute matching degree, and then add the real-time click heat correction value. The click heat is obtained by counting the change rate of detail page visits per unit time.

6. The backend data management method of an e-commerce sales platform according to claim 1, characterized in that: The coupling matrix parameter adjustments in S4 include: Inventory forecast deviation feedback: Record the absolute deviation of the actual inventory consumption rate from the forecast value; When the deviation direction is consistent within consecutive time periods, the demand forecast weight in the resource-commodity coupling matrix is ​​reversely revised according to the deviation ratio; Alternative product conversion feedback: Count the conversion rate of users’ clicks on recommended alternative products. The conversion rate is calculated by the ratio of the number of paid orders after clicks to the number of recommended products. When the conversion rate is lower than the historical average by a certain percentage, the correlation strength of product categories is reduced. The reduction extent is in a piecewise linear relationship with the conversion rate deviation value. The critical ratio value is determined by optimizing historical A / B test data.

7. The backend data management method of an e-commerce sales platform according to claim 1, characterized in that: Resource scheduling strategies in S3 include: When the regional service pressure value continuously exceeds the dynamic threshold, graded load reduction is performed: Phase 1: Convert static resources to low-resolution versions, with the resolution degradation ratio increasing step by step based on the pressure value exceeding the limit; Phase 2: Sort by product-user coupling matrix value and increase the number of container replicas for high-value product pages. The number of replicas added is positively correlated with the growth rate of pressure value. The calculation basis of the dynamic threshold is: the moving average of the historical peak pressure value of the node multiplied by the flow fluctuation coefficient of the current period.

8. The backend data management method of an e-commerce sales platform according to claim 1, characterized in that: The calculation of the element values ​​of the resource-commodity coupling matrix in S2 includes: Demand forecasts are generated using a time series decomposition model, which includes the following steps: Decompose historical sales data into trend items, seasonal items, and random residual items; Use the rate of change of the commodity-user coupling matrix value to correct the slope of the trend term; The discount intensity in the promotion strategy is converted into the seasonal amplitude adjustment factor. The discount intensity is nonlinearly positively correlated with the amplitude adjustment factor. Quantification of logistics costs: Obtain the path distance from the warehouse to the user area, and the distance data is updated in real time through the geographic information system path planning interface; A traffic status correction factor is introduced, which is mapped to a numerical weight based on the congestion level returned by the real-time traffic API.

9. The backend data management method of an e-commerce sales platform according to claim 1, characterized in that: Inventory allocation strategies in S3 include: Dynamic threshold setting mechanism: The first threshold is set based on the historical sell-out rate of the product. The higher the sell-out rate, the lower the threshold. The second threshold is negatively correlated with the regional inventory turnover rate, which is obtained by counting the frequency of inventory updates per unit time. Cross-warehouse transfer path selection: Calculate the priority of the alternative warehouse for transfers, which is weighted by the warehouse's inventory adequacy, delivery time cost, and service pressure value; The weighting coefficient is dynamically adjusted based on the historical success rate of goods transfer.

10. An e-commerce sales platform backend data management system, characterized in that: It includes a memory configured to store instructions, a processor configured to call the instructions from the memory and to implement an e-commerce sales platform background data management method as described in any one of claims 1-9 when executing the instructions.

Images