Electronic commerce promotion method and system based on cloud computing

Abstract

The invention discloses an e-commerce promotion method and system based on cloud computing, and relates to the technical field of cloud computing e-commerce promotion, and the method comprises the steps: collecting operation event data, and carrying out the event normalization and time window aggregation, and obtaining behavior feature data, channel cost data, commitable supply data, performance time efficiency distribution data and risk data; casting and supplying promotion quota tokens for commodities in each time window according to the promotion quota tokens, and generating a token pool; generating a candidate commodity list and a substitution sequence, and calculating a touch increment net income to form a touch candidate list; solving a scarcity price and determining a promotion action by combining a token pool, and updating according to a substitution sequence when the tokens are insufficient; a two-stage token mechanism is adopted to complete pre-occupation, confirmation and cancel-after-verification / release, audit logs are generated, scarce prices and performance punishment are returned and updated, scarce tokens are preferentially used for high net income reaching opportunities and can be recycled in time, the refund canceling probability is remarkably reduced, and the overall user experience level is improved.

Description

Technical Field

[0001] This invention relates to the field of cloud computing e-commerce promotion technology, and in particular to a cloud computing-based e-commerce promotion method and system. Background Technology

[0002] In a cloud computing environment, e-commerce promotion typically relies on distributed computing and storage capabilities to process multi-source operational data, including user reach, transactions, fulfillment, and channel placement, and to create user preference profiles, channel cost metrics, and product operation profiles within a unified time scale. Based on this, the platform can evaluate user, product, and channel combinations, generate candidate product sets, and select placement channels. This enables coordinated configuration of advertising budgets, display frequency, and reach pace, thereby supporting real-time promotion decisions and performance feedback in large-scale concurrent scenarios.

[0003] When high-frequency promotions coexist with supply fluctuations, conventional practices often rely on demand-side signals and channel costs to allocate opportunities to reach customers. The ability to deliver on supply and the timeliness of fulfillment are difficult to quantify into executable constraints, and the matching degree between promotion quotas and fulfillment capabilities is prone to decline. At the same time, the lack of transactional linkage between key events such as placement, add-to-cart, and payment makes it difficult to ensure consistency in the usage, recovery, and auditing links under concurrency, thereby affecting the stability and traceability of strategy iteration. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a cloud computing-based e-commerce promotion method to solve the problems of low matching degree between promotion quota and cashable supply, and lack of transactional consistency and traceability in the execution of campaigns and key events in existing technologies.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a cloud computing-based e-commerce promotion method, which includes, Collect operational event data, including behavioral characteristic data, channel cost data, committable supply data, fulfillment timeliness distribution data, and risk data. Based on the committable supply data, fulfillment timeliness distribution data, and risk data, generate and supply promotion quota tokens for each product within each time window, and create a token pool. Generate a candidate product list based on behavioral characteristic data and channel cost data, construct alternative sequences within the candidate product list, calculate the net revenue from incremental reach, and obtain a reach candidate list. Based on the token pool and the reach candidate list, calculate the scarcity price, determine promotion actions based on the scarcity price, and update the promotion actions. Use a two-stage token mechanism to bind the deployment and redeemable supply for promotion actions, and update the scarcity price and fulfillment penalty.

[0007] As a preferred embodiment of the cloud computing-based e-commerce promotion method of the present invention, the steps of collecting business event data and performing unified event normalization and time window aggregation on the business event data are as follows: collecting user behavior, order status, inventory changes, warehousing and distribution capacity, channel consumption, and fulfillment feedback data to obtain business event data; unifying the data according to the event mapping table and determining duplicates using idempotent keys; dividing the time window according to the event time and using the delay tolerance threshold as the water level to filter real-time aggregated sub-streams.

[0008] As a preferred embodiment of the cloud computing-based e-commerce promotion method of the present invention, the specific steps for acquiring behavioral feature data, channel cost data, committable supply data, fulfillment timeliness distribution data, and risk data are as follows: generating browsing intensity features and add-to-cart intensity features according to the time decay feature expression to form behavioral feature data; generating unit reach cost and channel congestion from channel consumption amount and click volume to form channel cost data; calculating the committable supply volume to form committable supply data; constructing a segmented histogram using a fixed bin width, and obtaining the fulfillment timeliness probability density from the histogram normalization to obtain fulfillment timeliness distribution data; and forming risk data by statistically analyzing the risk ratios of successful payment, cancellation, and return.

[0009] As a preferred embodiment of the cloud computing-based e-commerce promotion method of the present invention, the generation of the token pool includes: calculating the redeemability probability and the expected tail loss of overdue orders based on the fulfillment timeliness distribution data; using a binned histogram to represent the distribution, determining the median value of each bin, and summing the counts of bins whose median value falls within the range of zero to the promised timeliness to obtain the total number within a threshold; using the ratio of the total number within the threshold to the total count as the redeemability probability; calculating the number of tokens for each product within a time window; when the number of promotion quota tokens is equal to 0, no tokens are generated, and a corresponding number of token records are generated according to the number of promotion quota tokens to obtain the token pool.

[0010] As a preferred embodiment of the cloud-based e-commerce promotion method of the present invention, the steps of generating a candidate product list based on behavioral feature data and channel cost data, constructing an alternative sequence within the candidate product list, calculating the net incremental benefit of reaching customers, and obtaining a candidate list for reaching customers are as follows: generating a candidate product recall set based on behavioral feature data according to a fixed recall rule; constructing an alternative sequence within the same candidate product list; comparing the purchase occurrence ratio of reached and unreached customers to obtain the incremental purchase probability; calculating the net incremental benefit of reaching customers based on the incremental purchase probability, expected reaching cost, and fulfillment loss; and obtaining a candidate list for reaching customers. The fixed recall rule includes... Products with a higher add-to-cart intensity than browsing intensity are used as a seed set. If the seed set is empty, the product with the highest browsing intensity is used as the seed. The seed products are queried using the product similarity retrieval index to obtain highly similar products, and these are combined with the seed products. Invalid products are removed, and the candidate product list is obtained by sorting the products according to the rule of prioritizing add-to-cart intensity and then browsing intensity. Within the candidate product list, products in the same category as the target product and whose price range falls within the same price range are prioritized to form a price range product set. These products are then selected in descending order of similarity according to the product similarity retrieval index to form a replacement sequence. When the number of replacement sequences is insufficient, they are supplemented according to the rule of adjacent categories and the highest degree of attribute overlap.

[0011] As a preferred embodiment of the cloud computing-based e-commerce promotion method of the present invention, the following steps are taken: Based on a token pool and a candidate list for reaching the target product, a scarcity price is calculated, and a promotional action is determined using the scarcity price. When there are insufficient tokens for the target product or the executable conditions are not met, a replacement sequence is determined as a replacement product, and the promotional action is updated. Specifically, the records in the candidate list are merged by product to obtain a net revenue sequence; the scarcity price is calculated on the net revenue sequence; the remaining amount after deducting the scarcity price from the net revenue for each record in the candidate list is calculated and used as an executable decision quantity; and a promotional action is generated based on the executable decision quantity.

[0012] As a preferred embodiment of the cloud computing-based e-commerce promotion method of the present invention, the specific steps of generating the promotion action are as follows: selecting the product and channel combination with the largest executable decision quantity from the reach candidate list as the preferred promotion action; generating a promotion action candidate and deducting the executable quota count when the number of available tokens for the product is greater than zero and the executable decision quantity is positive; when the preferred action token is insufficient or cannot be executed, replacing the target product according to the substitution sequence in the user candidate list and simultaneously verifying the executable decision quantity and the number of available tokens to generate the promotion action.

[0013] As a preferred embodiment of the cloud computing-based e-commerce promotion method of the present invention, the promotion action adopts a two-stage token mechanism to bind the delivery and redeemable supply. The specific steps are as follows: retrieve available tokens from the token pool according to the earliest expiration time and set the tokens to a pre-reserved state through atomic state transition; generate a delivery instruction after successful pre-reservation; subscribe to user behavior feedback and order feedback event streams during the pre-reservation validity period; match add-to-cart events or payment initiation events to transition the tokens to a confirmed state; match successful payment events to transition the tokens to a used state; match payment failure, order cancellation or refund events to revert the tokens to an available state.

[0014] As a preferred embodiment of the cloud computing-based e-commerce promotion method of the present invention, the specific steps of updating the scarcity price and performance penalty are as follows: generating a scarcity price smoothing coefficient based on the difference between token utilization rate and target utilization rate and linearly fusing it to obtain a smoothed scarcity price; updating the performance penalty parameters based on the difference between the actual average timeout and the expected timeout.

[0015] Secondly, this invention provides a cloud-based e-commerce promotion system, comprising: an event aggregation module, which collects operational event data and obtains behavioral characteristic data, channel cost data, committable supply data, fulfillment timeliness distribution data, and risk data; a token minting module, which, based on committable supply data, fulfillment timeliness distribution data, and risk data, mints and supplies promotion quota tokens for each product within each time window and generates a token pool; a candidate evaluation module, which generates a candidate product list based on behavioral characteristic data and channel cost data, constructs an alternative sequence within the candidate product list, calculates the net benefit of incremental reach, and obtains a reach candidate list; an action decision module, which, based on the token pool and the reach candidate list, calculates the scarcity price, determines the promotion action through the scarcity price, and updates the promotion action; and a transaction closed-loop module, which uses a two-stage token mechanism to bind the delivery and redeemable supply for the promotion action and sends back updated scarcity prices and fulfillment penalties.

[0016] The beneficial effects of this invention are as follows: by constructing an alternative sequence within the candidate product list and forming an access candidate list based on the incremental net revenue, it is possible to quickly switch alternative products under token constraints; by calculating the scarcity price based on the token pool and the access candidate list and determining the promotion action accordingly, it is possible to prioritize the allocation of scarce tokens to high-net-return access opportunities; through a two-stage token mechanism, it is possible to achieve traceable closed-loop delivery and transaction, prevent tokens from being repeatedly occupied, and ensure timely recovery. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a cloud computing-based e-commerce promotion method.

[0019] Figure 2 This is a schematic diagram of a cloud-based e-commerce promotion system.

[0020] Figure 3 A flowchart for reaching the candidate list.

[0021] Figure 4 This is a flowchart for two-phase token transaction processing and postback updates.

[0022] Figure 5 This is a comparative data chart of average substitution depth. Detailed Implementation

[0023] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0024] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0025] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0026] Reference Figures 1-5 This is one embodiment of the present invention, which provides a cloud computing-based e-commerce promotion method, including the following steps: S1. Collect operational event data, and obtain behavioral characteristic data, channel cost data, promised supply data, performance time distribution data, and risk data.

[0027] Collect user behavior, order status, inventory changes, warehousing and distribution capacity, channel consumption, and fulfillment feedback data to obtain operational event data; unify event types into an enumeration set according to the event mapping table (e.g., browsing, clicking, adding to cart, payment, cancellation, return, shipment, and receipt), and unify them into a standard unit; use idempotent keys to determine duplicates, and events determined to be duplicates are not aggregated, only the first valid event is retained; divide time windows by event time, and set a delay tolerance threshold as a waterline; when an event arrives later than the delay tolerance threshold, it is not aggregated in real time, and events that meet the waterline are aggregated into substreams in real time.

[0028] It should be noted that the event mapping table is pre-configured and generated by the data governance side based on the original field dictionary of each data source and the business definition specifications; the latency tolerance threshold is obtained by long-term sampling of the difference between the event generation time and the arrival time in the cloud of each data source, and selecting the quantile corresponding to the target coverage as the latency tolerance threshold, such as the 95th latency quantile.

[0029] For each user and product within a time window, browsing intensity features are generated based on events, and add-to-cart intensity features are generated based on add-to-cart events. These browsing intensity features and add-to-cart intensity features are combined into behavioral feature data, and a time decay feature expression is defined as follows: ; in, Indicates user In the time window For goods behavioral intensity characteristics Indicates time window At the end of the day, Indicates the length of the time window. Represents the integral variable. Indicates the time decay coefficient. Indicates at time User For goods The value is 1 if the target event occurs, otherwise it is 0.

[0030] It should be noted that, The target event refers to an add-to-cart event or a browsing event; It is based on the statistical fitting of the conversion probability decay curve at different time intervals after the occurrence of the target event based on historical logs (for example, calculating the conversion rate by bucketing according to the time difference between the occurrence of the event and the payment), and selects the decay rate that minimizes the fitting error as the time decay coefficient.

[0031] For each channel within a time window, aggregate channel spending and effective reach (clicks). Use the ratio of spending to effective reach within the time window as the unit reach cost for that time window. Simultaneously, calculate the bidding success rate to indicate channel congestion. Combine the unit reach cost and channel congestion into channel cost data.

[0032] For each product and fulfillment warehouse, the current inventory is accumulated based on inventory change events within the time window; the committed but not yet shipped quantity is aggregated based on orders that have been successfully paid for but not yet shipped; the difference between the current inventory and the committed but not yet shipped quantity is used as the median value, and the difference between the median value and the safety stock threshold is used as the committable supply quantity. If the committable supply quantity is negative, it is truncated to 0 to obtain the committable supply data.

[0033] It should be noted that the safety stock threshold is calculated by taking the highest percentile value of the demand during the replenishment lead time and using it as the safety stock threshold, such as the 90th percentile value, with a value range of not less than 0.

[0034] For each product and fulfillment warehouse within a time window, extract the fulfillment duration sample for each order from fulfillment feedback events (e.g., from payment to shipment). Construct a segmented histogram using a fixed bin width, and obtain the fulfillment time probability density by histogram normalization. Obtain the fulfillment time distribution data, expressed as: ; in, Indicates goods In the fulfillment warehouse And within the time window The probability density of the fulfillment time at that time Indicates the width of the compartment. Indicates the maximum index of the bins. Indicates time window Domestic goods From fulfillment warehouse The fulfillment time of the fulfilled orders falls into the first category. Number of samples per bin.

[0035] For each product within a time window, the number of successfully paid orders, canceled orders, and returned orders are counted. The sum of the canceled and returned orders is used as the median value, and the ratio of the median value to the number of successfully paid orders is used as the risk ratio to obtain risk data. When the number of successfully paid orders is 0, the risk ratio is set to 1 to indicate that there are no valid transaction samples within the time window and the risk cannot be estimated with a lower bound.

[0036] S2. Based on the data on commensurable supply, the distribution data on fulfillment timeliness, and the risk data, tokens for promotional quotas are minted and supplied for each commodity within each time window, and a token pool is generated.

[0037] Based on the performance time distribution data, the overtime amount is calculated and averaged for each sample whose performance time exceeds the promised time limit to obtain the expected tail loss of overtime. A bin histogram is used to represent the distribution, the median value of each bin is determined, and the counts of bins whose median value falls between zero and the promised time limit are summed to obtain the total number within the threshold. The ratio of the total number within the threshold to the total count is used as the redeemability probability. For each product within the time window, the number of tokens is calculated, expressed as: ; in, Indicates goods In the time window The number of promotional tokens internally minted. Indicates goods In the main fulfillment warehouse And time window The available supply within the country Indicates the main fulfillment warehouse In the time window The maximum number of orders that can be processed within the system. Indicates goods The proportion of risk within the time window θ Indicates goods The promised time limit, This indicates the upper bound of the integral of the time taken to fulfill the contract.

[0038] It should be noted that the promised delivery time is the maximum shipping and delivery time promised to external parties when the product is published. Within each time window, the system reads the number of picked items, packed items, and shipped items from the fulfillment warehouse, and takes the minimum number of completed items as the upper limit of the number of orders that can be processed in the current time window. It involves statistically analyzing the distribution of time taken from payment to fulfillment nodes using historical fulfillment feedback data, and selecting the corresponding high quantile time as the upper bound of the fulfillment time integral, such as covering the time taken for 99% of the samples.

[0039] When the number of promotion quota tokens is equal to 0, it means that the product does not have a promotion quota that can be promised and redeemed within the current time window, and no token is generated; generate a corresponding number of token records according to the number of promotion quota tokens, obtain the token pool, and each token record must contain at least the unique token identifier, the corresponding product identifier, the corresponding fulfillment warehouse identifier, the corresponding time window identifier, the token expiration time, and the token status.

[0040] It should be noted that the token status is determined by writing an initial value of "available" when the token is generated, setting it to "pre-reserved" before distribution, setting it to "confirmed" when high-intent events such as adding to purchase or payment occur, setting it to "used" when payment is successful, and reverting to "available" if canceled or timed out. The token expiration time is the end time of the time window to which the token belongs.

[0041] S3. Generate a candidate product list based on behavioral feature data and channel cost data, construct an alternative sequence within the candidate product list, calculate the net revenue from incremental reach, and obtain the candidate reach list.

[0042] For each user, a candidate product recall set is generated based on behavioral feature data according to fixed recall rules. The fixed recall rules include: using products whose add-to-cart intensity is higher than their browsing intensity within the current time window as the seed set; if the seed set is empty, using a number of products with the highest browsing intensity as the seed set; for each seed product, querying the product similarity retrieval index and selecting a fixed number of similar products with the highest similarity; after merging all similar products and seed products, invalid products (e.g., products that are delisted, unsaleable, or whose promised supply is zero within the time window) are removed, and the top fixed number are selected according to the sorting rule of adding-to-cart intensity first, followed by browsing intensity, to obtain the user's candidate product list.

[0043] It should be noted that the product similarity search index is formed by extracting the title, category, key attributes (e.g., specifications, brand, price range) of each product from the product database, as well as high-frequency keywords extracted from the review text, concatenating them into standardized text and quantifying them. Then, a list of similar products and a similarity score are pre-calculated and stored for each product using a nearest neighbor search method. The mapping from product identifiers to the list of similar product identifiers serves as the product similarity search index.

[0044] For each product in the candidate product list, an alternative sequence is constructed within the same candidate product list. The generation rules for the alternative sequence are as follows: products in the same category as the product and whose price bands fall within the same price band range are selected as the price band product set; in the price band product set, products are sorted from high to low according to the similarity returned by the product similarity retrieval index; the first fixed number of alternative products are taken as the alternative sequence. If the number of alternative products is less than the fixed number, the sequence is supplemented to the fixed alternative length according to the supplementation rule of adjacent categories and the highest degree of attribute overlap.

[0045] It should be noted that the same price band range is determined by summarizing the median price distribution of payment orders in the past 30 days under the same third-level category, calculating the lower quartile and upper quartile prices of the median price distribution, and using the quartile segment where the median price of the target product falls as the price band range. Products among the candidate products whose median price falls within the price band range are identified as products of the same price band. The fixed substitution length is determined by measuring the average additional calculation time brought about by each additional substitution product in online statistics, using the difference between the maximum processing latency allowed for real-time delivery and the basic processing time as the available time, and using the ratio of available time to additional time as the upper limit of quantity determined by latency. The upper limit of the number of interfaces is obtained by reading the maximum number of candidate products allowed to be carried in a single delivery instruction, and the minimum value between the upper limit of quantity determined by latency and the upper limit of the number of interfaces is taken as the fixed substitution length.

[0046] For each product and channel combination in the candidate product list, an outreach sample set and an inreach sample set are constructed based on historical logs. The purchase occurrence ratio under the same behavioral characteristics is statistically analyzed to obtain the purchase probability under outreach conditions and the purchase probability under inreach conditions. The difference between the purchase probability under outreach conditions and the purchase probability under inreach conditions is taken as the incremental purchase probability of the product reaching users through the current channel. Based on the unit outreach cost of the channel within the time window and the historical cost deviation coefficient of the product in the current channel, the expected outreach cost of the current combination is calculated. The fulfillment time distribution of the product in the main fulfillment warehouse is obtained from the fulfillment feedback, and the average overtime amount exceeding the promised time limit is calculated and converted into fulfillment loss. The product of the incremental purchase probability and the unit gross profit is taken as the incremental median value. The difference between the incremental median value, the expected outreach cost, and the fulfillment loss is taken as the incremental net benefit of the product under the channel. A candidate list with incremental net benefit is obtained as the outreach candidate list.

[0047] The behavioral characteristics include user browsing intensity and add-to-cart behavior intensity.

[0048] In this embodiment, to verify the impact of substitution sequence length on average substitution depth, under the same token constraint conditions, the substitution sequence length was set to 0, 3, and 6 respectively, and statistics were performed. Figure 5 As shown, when the substitution sequence length is used as a variable, the average substitution depth changes with the time window identifier. The substitution depth reflects which substitution item must be switched from the preferred option in the candidate item list to obtain an executable access opportunity under the token constraint. The lower the average substitution depth, the easier it is to complete the action with a shallower substitution in the access candidate list, thus switching faster and saving computation. The multiple curves in the figure correspond to different substitution sequence lengths (such as 0, 3, and 6). When the length is 0, the average substitution depth is close to zero, indicating a lack of substitution mechanism. After the length increases, the average substitution depth shows distinguishable changes in several time windows. The magnified part is used to highlight the subtle differences and prove that the substitution sequence length affects the range and magnitude of the rapid switching capability.

[0049] It should be noted that the historical cost deviation coefficient is calculated by summarizing the actual unit reach cost over a recent period (e.g., one month) for the same product and the same channel, and then comparing it with the overall unit reach cost of the channel during the same period. This ratio is used as the cost deviation coefficient of the product in the channel.

[0050] S4. Based on the token pool and the reach candidate list, calculate the scarcity price, determine the promotion action based on the scarcity price, and update the promotion action.

[0051] Within each time window, the number of tokens available for each product within the time window is read from the token pool, and an executable quota count is generated for each product. The executable quota count decreases only as the selected action consumes tokens within the current time window and is refreshed in the next time window. The reach candidate list is read, and the records in the reach candidate list are merged by product. For each product, a net revenue sequence is obtained, where each sequence element corresponds to the incremental net revenue value of a user reaching the product through a certain channel.

[0052] For each commodity, using the number of available tokens as a constraint, solve for the scarcity price on the net revenue sequence, ensuring that the net revenue is not lower than the number of reach opportunities and available tokens that match the scarcity price. The expression is: ; in, Indicates goods In the time window The scarcity price within, Indicates the candidate threshold value. Indicates goods In the time window The collection of reach opportunities within, Indicates user In the time window Through channels Reaching products The incremental net revenue from reaching [the target audience] Indicates goods In the time window The number of available tokens, This indicates an indicator function; the value is 1 if the condition within the parentheses is true, and 0 if it is false.

[0053] It should be noted that, It extracts product identifier, user identifier, and channel identifier from each record in the candidate list of reach, and groups them by product identifier. For records in the same product identifier group, the combination of user identifier and channel identifier is used as a reach opportunity. After deduplication of duplicate user identifier and channel identifier combinations, a set of reach opportunities for the product within the time window is formed. This is achieved by aggregating samples of incremental net revenue from reach opportunities for the same product across various channels within a statistical period (e.g., one month), and using the net revenue value corresponding to the highest percentile as a candidate threshold, such as the 90th percentile, so that the reach opportunities included in the scarcity price calculation meet the minimum revenue requirement.

[0054] For each record in the reach candidate list, calculate the net profit minus the scarcity price and use the remaining amount as the executable decision quantity. When the remaining amount is positive, it means that the reach opportunity still has positive profit potential under the scarcity constraint. When the remaining amount is negative, it means that it should give way to higher net profit opportunities, thereby avoiding the consumption of scarce tokens by low-value reach.

[0055] Promotional actions are generated at the user level. Specifically, for the same user, the target product and channel combination with the highest executable decision value is selected as the primary action from the reach candidate list. If the available tokens for the product corresponding to the primary action are greater than zero and the executable decision value is positive, the action is recorded as a promotional action candidate for the user within the time window, and the executable quota count for the product is decremented by one. If the available tokens for the product corresponding to the primary action are zero, or the executable decision value is not positive, the alternative sequence of the primary target product in the user's candidate list is read, and the target product is replaced one by one according to the order of the alternative sequence. After each replacement, the executable decision value and available token quantity conditions of the corresponding record are recalculated. When there is an alternative product that satisfies the condition of having an available token value greater than zero and an executable decision value positive, the product and channel corresponding to the alternative product are recorded as promotional action candidates for the user, and the executable quota count for the alternative product is decremented by one. If no product satisfies the conditions after traversing the alternative sequence, no promotional action is generated for the user within the time window.

[0056] S5. A two-stage token mechanism is used for promotional activities to bind the release and redeemable supply, and the scarcity price and performance penalty are updated back.

[0057] For each action, an action transaction record is generated. The action transaction record includes user identifier, product identifier, channel identifier, time window identifier, action identifier, the incremental net revenue value corresponding to the action, and the scarcity price value of the product within the time window. The action transaction record is written to the pre-occupancy state.

[0058] The system retrieves target product tokens from the token pool in the order of earliest expiration time, prioritizing those in an available state. It then performs atomic state transitions on the tokens. Specifically, the token is updated to a pre-reserved state and an action identifier and pre-reservation start time are written only if the token's current state is available. If the atomic update fails, the system continues to retrieve the next available token until it succeeds or the tokens are exhausted. If the tokens are exhausted, the action transaction record is set to "no token failed," thus ensuring that the same token is not pre-reserved concurrently.

[0059] Once the pre-allocation is successful, a delivery instruction is generated. The delivery instruction includes the user identifier, product identifier, channel identifier, action identifier, and unique token identifier. The exposure identifier returned by the channel is written into the action transaction record for subsequent event association, so that the delivery behavior and the pre-allocated redeemable supply tokens form a traceable binding relationship.

[0060] Based on a sample set selected from historical logs that meets the criteria of having generated exposure receipts and corresponding orders that have been successfully paid, the time difference from the exposure receipt time to the add-to-cart event time and the time difference from the exposure receipt time to the payment success event time are calculated for each sample. The time differences are merged to form a time difference sequence, and the time difference sequence is sorted from smallest to largest. The time difference at the corresponding percentile position (e.g., the 95th percentile position) after sorting is taken as the pre-occupancy validity period. The pre-occupancy validity period is written into the pre-occupancy record, and a timed scanning task is started to release tokens that are still in the pre-occupancy state after the pre-occupancy validity period and have not been confirmed, thereby avoiding the tokens being suspended and occupied for a long time.

[0061] Continuously subscribe to user behavior feedback and order feedback event streams. When a add-to-cart event or payment initiation event matching the current action transaction record is detected (the matching conditions include user ID, product ID, exposure ID, receipt ID, and action ID), the corresponding token is transitioned from the pre-reserved state to the confirmed state, and the confirmation time and trigger event type are recorded. If a cancellation event is received before confirmation or the pre-reservation validity period expires, confirmation is not initiated.

[0062] When a payment success event matching the current action transaction record is detected, the corresponding token is moved from the confirmed state to the used state and written to the order identifier; when a payment failure, order cancellation, or refund event is detected, the token is moved to the available state.

[0063] For each token state migration and delivery call, an append-only audit log is generated. The audit log includes the token's unique identifier, action identifier, old state, new state, trigger event type, trigger event time, channel receipt identifier, and the incremental net revenue and scarcity price corresponding to the action.

[0064] For each product, the number of tokens minted and redeemed within the current time window is recorded, and the token utilization rate is calculated. The target utilization rate of the product is read, and the difference between the utilization rate of the current time window and the target utilization rate is calculated. Based on the difference, a scarcity price smoothing coefficient is determined linearly. Specifically, when the difference is 0, the scarcity price smoothing coefficient is 1; when the difference is 1, the scarcity price smoothing coefficient is 0; when the difference is between 0 and 1, the scarcity price smoothing coefficient is "1 minus the difference", so that the smaller the difference, the larger the scarcity price smoothing coefficient, and the larger the difference, the smaller the scarcity price smoothing coefficient. The scarcity price smoothing coefficient is used to linearly merge the smoothed scarcity price of the previous time window and the original scarcity price of the current time window to obtain a smoothed scarcity price for direct use in determining promotion actions in the next time window.

[0065] It should be noted that within the statistical period, the token redemption quantity and token minting quantity of the same product in each time window are summarized, and the total redemption quantity and total minting quantity within the statistical period are obtained by summing them. The ratio of the total redemption quantity to the total minting quantity is used as the target utilization rate of the product in the next time window. When the total minting quantity is 0, the target utilization rate is set to 0, and the value range is from 0 to 1.

[0066] Extract the fulfillment time of paid orders for goods under the main fulfillment warehouse from the fulfillment feedback records, and count the orders whose fulfillment time exceeds the promised time limit of the goods; calculate the time difference of exceeding the promised time limit for each order exceeding the promised time limit and average it to obtain the actual average timeout amount for this time window; read the fulfillment time limit distribution data corresponding to the goods in this time window, and calculate the expected timeout amount under the same promised time limit; use the difference between the actual average timeout amount and the expected timeout amount as the update amount. When the actual average timeout amount is greater than the expected timeout amount, increase the fulfillment penalty parameter; when the actual average timeout amount is less than the expected timeout amount, decrease the fulfillment penalty parameter; obtain the updated fulfillment penalty parameter, and use it for supply reduction when token minting in the next time window and for fulfillment loss deduction of incremental net revenue when promotion actions are determined, so that the promotion amount can be adaptively adjusted according to the change of fulfillment risk.

[0067] This embodiment also provides a cloud computing-based e-commerce promotion system, including: an event aggregation module, which collects business event data, performs unified event normalization and time window aggregation on the business event data, and obtains behavioral characteristic data, channel cost data, commensurable supply data, fulfillment timeliness distribution data, and risk data; The token minting module, based on the committable supply data, fulfillment time distribution data, and risk data, mints promotional tokens corresponding to the supply redemption capacity of each commodity within each time window, and generates a token pool. The candidate evaluation module generates a candidate product list based on behavioral characteristic data and channel cost data, constructs an alternative sequence within the candidate product list, calculates the net revenue from incremental reach, and obtains a candidate reach list. The action decision module calculates the scarcity price based on the token pool and the reach candidate list, determines the promotion action based on the scarcity price, and determines the alternative sequence as the alternative product and updates the promotion action when there are insufficient tokens for the target product or the execution conditions are not met. The transaction closed-loop module uses a two-stage token mechanism to bind the distribution and redeemable supply for promotion actions, and sends back updates on scarcity prices and performance penalties.

[0068] In summary, this invention enables rapid switching of alternative products under token constraints by constructing an alternative sequence within the candidate product list and forming an access candidate list based on the incremental net revenue from reaching the target; it enables priority allocation of scarce tokens to high-net-return access opportunities by calculating the scarcity price based on the token pool and the access candidate list and determining the promotion action accordingly; and it achieves traceability of the delivery and transaction closed loop through a two-stage token mechanism, ensuring that tokens are not repeatedly used and can be recovered in a timely manner.

[0069] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A cloud computing-based e-commerce promotion method, characterized in that: include, Collect operational event data, and obtain behavioral characteristic data, channel cost data, available supply data, performance time distribution data, and risk data; Based on commensurable supply data, fulfillment timeliness distribution data, and risk data, promotion quota tokens are minted and supplied for each commodity within each time window, and a token pool is generated. A candidate product list is generated based on behavioral feature data and channel cost data. An alternative sequence is constructed within the candidate product list, the net revenue from incremental reach is calculated, and a candidate reach list is obtained. Based on the token pool and the reach candidate list, calculate the scarcity price, determine the promotion action based on the scarcity price, and update the promotion action; A two-stage token mechanism is used to link the distribution and redeemable supply for promotional activities, and the scarcity price and performance penalty are updated back.

2. The cloud computing-based e-commerce promotion method as described in claim 1, characterized in that: The process of collecting operational event data, performing unified event normalization and time window aggregation on the operational event data, involves the following steps: Collect operational event data from user behavior, order status, inventory changes, warehousing and distribution capacity, channel consumption, and fulfillment feedback data, unify them according to the event mapping table, and determine duplicates using idempotent keys; Time windows are divided according to event time, and the delay tolerance threshold is used as the water level to filter real-time aggregated substreams.

3. The cloud computing-based e-commerce promotion method as described in claim 2, characterized in that: The specific steps for acquiring behavioral characteristic data, channel cost data, committable supply data, fulfillment timeliness distribution data, and risk data are as follows: Generate browsing intensity features and add-to-cart intensity features based on the time decay feature expression, and form behavioral feature data; The unit cost of reaching and channel congestion are generated from channel consumption and clicks to form channel cost data, and the promised supply is calculated to form promised supply data. A segmented histogram is constructed using a fixed bin width, and the delivery time probability density is obtained by normalizing the histogram to acquire delivery time distribution data. Risk data is generated by statistically analyzing the risk ratios of successful payments, cancellations, and returns.

4. The cloud computing-based e-commerce promotion method as described in claim 3, characterized in that: The token generation pool includes, Calculate the probability of fulfillment and the expected tail loss of overdue payments based on the distribution data of performance timeliness; The distribution is represented by a bin histogram. The median value of each bin is determined. The total number of bins whose median value falls between zero and the promised time limit is summed to obtain the total number within the threshold. The ratio of the total number within the threshold to the total count is used as the redeemability probability. For each product, the number of tokens is calculated within the time window. When the number of promotion quota tokens is 0, no tokens are generated. Instead, a corresponding number of token records are generated according to the number of promotion quota tokens, and the token pool is obtained.

5. The cloud computing-based e-commerce promotion method as described in claim 4, characterized in that: The steps involve generating a candidate product list based on behavioral characteristic data and channel cost data, constructing an alternative sequence within the candidate product list, calculating the net revenue from incremental reach, and obtaining the reach candidate list. A candidate product recall set is generated based on behavioral feature data and according to fixed recall rules; Construct alternative sequences within the same candidate product list; The incremental purchase probability is obtained by comparing the purchase occurrence ratio between those reached and those not reached. The net incremental benefit of reaching is calculated based on the incremental purchase probability, expected reaching cost, and fulfillment loss, and a candidate list of reaching is obtained. The fixed recall rule includes taking products with higher add-to-cart intensity characteristics than browsing intensity characteristics as a seed set. When the seed set is empty, the product with the highest browsing intensity characteristic is taken as the seed. The seed product is queried in the product similarity retrieval index to obtain highly similar products and the seed product is combined. After removing invalid products, the candidate product list is obtained by sorting according to the rule of adding to cart intensity first and browsing intensity second. The process involves prioritizing the selection of products within the candidate product list that are in the same category as the target product and whose price bands fall within the same price band range to form a price band product set. These products are then selected in descending order of similarity according to the product similarity retrieval index to form a replacement sequence. If the number of replacement sequences is insufficient, they are supplemented according to the rule of adjacent categories and the highest degree of attribute overlap.

6. The cloud computing-based e-commerce promotion method as described in claim 5, characterized in that: The process involves calculating a scarcity price based on the token pool and the reach candidate list, determining the promotion action based on the scarcity price, and identifying alternative products when there are insufficient tokens for the target product or when the execution conditions are not met. The specific steps are as follows: The records in the candidate list are grouped by product to obtain the net revenue sequence, and the scarcity price is solved on the net revenue sequence. For each record in the candidate list, calculate the net revenue minus the scarcity price and use the remaining amount as the executable decision quantity; Promotional actions are generated based on executable decision variables.

7. The cloud computing-based e-commerce promotion method as described in claim 6, characterized in that: The specific steps for generating the promotion action are as follows: Select the product and channel combination with the highest executable judgment volume from the candidate reach list as the preferred promotion action; When the number of available tokens for a product is greater than zero and the executable decision quantity is positive, a promotion action candidate is generated and the executable quota count is deducted. When the preferred action tokens are insufficient or cannot be executed, the target product is replaced according to the alternative sequence in the user's candidate list, and the number of executable judgments and available tokens are checked simultaneously to generate the promotion action.

8. The cloud computing-based e-commerce promotion method as described in claim 7, characterized in that: The promotional action employs a two-stage token mechanism to bind the distribution and redeemable supply. The specific steps are as follows: Retrieve available tokens from the token pool in order of earliest expiration time and set the tokens to a pre-reserved state through atomic state transition. After successful pre-reservation, generate a delivery instruction. During the pre-reservation validity period, subscribe to user behavior feedback and order feedback event streams, and migrate the token to the confirmation state when matching add-to-cart or payment initiation events; A successful payment event will move the token to a used state, while a failed payment, order cancellation, or refund event will revert the token to a usable state.

9. The cloud computing-based e-commerce promotion method as described in claim 8, characterized in that: The process of updating the scarcity price and fulfillment penalty, and then transmitting the updated information back, involves the following steps: The scarcity price smoothing coefficient is generated based on the difference between token utilization and target utilization, and then linearly fused to obtain a smoothed scarcity price. The performance penalty parameters are updated based on the difference between the actual average timeout and the expected timeout.

10. A cloud-based e-commerce promotion system, based on the cloud-based e-commerce promotion method according to any one of claims 1 to 9, characterized in that: This includes an event aggregation module that collects operational event data, and obtains behavioral characteristic data, channel cost data, promised supply data, fulfillment timeliness distribution data, and risk data. The token minting module, based on the committable supply data, fulfillment timeliness distribution data, and risk data, mints and supplies promotion quota tokens for each product within each time window and generates a token pool. The candidate evaluation module generates a candidate product list based on behavioral characteristic data and channel cost data, constructs an alternative sequence within the candidate product list, calculates the net revenue from incremental reach, and obtains a candidate reach list. The action decision module calculates the scarcity price based on the token pool and the reach candidate list, determines the promotion action based on the scarcity price, and updates the promotion action. The transaction closed-loop module uses a two-stage token mechanism to bind the distribution and redeemable supply for promotion actions, and sends back updates on scarcity prices and performance penalties.

Images