E-commerce activity intelligent adaptation method based on multi-dimensional income dynamic deduction

Abstract

The invention discloses an e-commerce activity intelligent adaptation method based on multi-dimensional income dynamic deduction, and the method comprises the steps: collecting multi-source original data, carrying out the processing of missing values and abnormal values, and extracting key feature vectors; constructing an activity adaptation degree prediction model containing a feature dynamic coding layer and the like, integrating and generating shop exclusive and activity parameter vectors, and calculating a dynamic weight to obtain a personalized coding feature vector; decoupling to obtain basic income, incremental income and risk cost factors, and outputting a prediction result through weight adjustment and deviation correction; and through a preset decision threshold system and a multi-dimensional weight rule, performing analysis to obtain an activity adaptation degree decision, and outputting a result. According to the method, accurate adaptation of the e-commerce activities and shops is realized, the prediction precision and decision scientificity are improved, the operation risk is reduced, and the method is suitable for marketing activity decision support of various e-commerce merchants.

Description

Technical Field

[0001] This invention relates to the field of e-commerce platform management, specifically to an intelligent adaptation method for e-commerce activities based on multi-dimensional revenue dynamic extrapolation. Background Technology

[0002] With the rapid development of the e-commerce industry, platform marketing activities have become a core means for merchants to increase sales, expand their user base, and optimize inventory structure. Their adaptability directly impacts merchants' operating profits and market competitiveness. Currently, e-commerce platforms offer a variety of marketing activities, such as discounts, spending-based promotions, and limited-time offers. However, the operating characteristics and strategic goals of different stores vary significantly, leading to substantial differences in the effectiveness of the same campaign across different stores. Therefore, precise campaign adaptation decision support is urgently needed.

[0003] Existing e-commerce activity adaptation technologies and methods suffer from the following key shortcomings: Most merchants still select activity plans based on past experience or industry-standard practices, failing to fully integrate their own historical transaction data, user behavior data, and market trend data for quantitative analysis. This results in insufficient activity adaptability, easily leading to problems such as profit dilution, inventory backlog, or lower-than-expected user growth. Existing prediction models often use fixed feature weights or generalized feature encoding methods, failing to dynamically adjust feature importance based on the correlation between store operating characteristics and activity objectives. This limits the model's adaptability to different industries and store sizes, making prediction accuracy difficult to guarantee. Existing technologies primarily focus on predicting explicit indicators such as sales volume and direct revenue, failing to decouple and integrate basic and incremental revenue with implicit risk costs such as inventory backlog costs, profit dilution costs, and user expectation enhancement costs. This leads to one-sided revenue assessment and insufficient scientific decision-making. Existing models lack personalized correction logic for historical activity deviations of individual stores, relying solely on fine-tuning of general model parameters. This cannot offset prediction deviations caused by unique store operating models, making it difficult to reduce the discrepancy between prediction results and actual effects over long-term use.

[0004] Therefore, in order to solve the problems existing in the prior art, this invention proposes an intelligent adaptation method for e-commerce activities based on multi-dimensional revenue dynamic extrapolation. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention aims to provide an intelligent adaptation method for e-commerce activities based on multi-dimensional revenue dynamic extrapolation.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A smart adaptation method for e-commerce activities based on multi-dimensional revenue dynamic extrapolation includes: The data collection and processing steps involve collecting historical transaction data, user behavior event stream data, market trend data, and merchant preference parameters from the target store as raw data. Missing values ​​and outliers are processed on the raw data, and key feature vectors are extracted, including historical price elasticity coefficients, user value stratification indicators, store operation characteristics, and market environment characteristics. The prediction model construction steps include building an activity fit prediction model. The activity fit prediction model architecture includes a feature dynamic encoding layer, a revenue factor decoupling layer, a personalized deviation correction layer, and a multi-dimensional output layer. The key feature vectors are integrated to generate store-specific feature vectors, and the platform activity default parameters and merchant preference parameters are integrated to generate activity parameter vectors. These are input into the activity fit prediction model. Dynamic weights are calculated based on the semantic similarity between the activity target and the features, and the degree of influence of the features on the revenue output. The initial input matrix is ​​weighted and encoded using dynamic weights to obtain a personalized encoded feature vector. The adaptive dynamic deduction step decouples the personalized coding feature vector to obtain the basic revenue, incremental revenue and risk cost factor. The weights are adjusted according to the store's risk tolerance and the priority of the activity objectives, and the total revenue feature vector is regenerated. After correction by combining historical activity deviation data, the activity prediction result is output. The activity prediction result includes expected sales, profit and the number of new high-value users. The activity decision generation step involves analyzing the activity prediction results using a preset adaptation decision strategy to obtain the final activity adaptation decision and outputting the activity adaptation judgment result.

[0007] As a further improvement of the present invention, the data acquisition and processing steps include: identifying abnormal data through box plot analysis and selecting data to delete according to data importance; filling time series data using forward filling and backward filling methods; filling numerical data using mean processing; and filling categorical data using mode processing.

[0008] As a further improvement of the present invention, the calculation of the key feature vector includes: obtaining the historical price elasticity coefficient by fitting the data of past discounts and sales changes of store products using the least squares method; grouping user behavior data by clustering algorithm and calculating user value stratification index by combining average purchase frequency and average order value; calculating average gross profit margin, inventory turnover days, and visitor conversion rate as store operation features based on historical operating data; and extracting industry popularity index, holiday weight coefficient, and competitor activity data as market environment features.

[0009] As a further improvement of the present invention, the calculation of the dynamic weights includes: clarifying the correlation dimensions corresponding to each feature by combining the store's personalized operation characteristics and the activity objectives; quantifying the semantic similarity between each feature and the activity objectives by using cosine similarity; determining the degree of influence of each feature on the revenue output based on the industry's general data pre-trained influence factor library and integrating the actual effect of each feature on the revenue output in the store's historical activities; multiplying the semantic similarity of a single feature with its corresponding degree of influence to obtain the comprehensive contribution value of that feature; and performing linear normalization on the comprehensive contribution values ​​of all features to complete the allocation of dynamic weights for each feature.

[0010] As a further improvement of the present invention, the basic revenue calculation of the adaptation dynamic deduction step includes: using the historical price elasticity coefficient correlation data in the key feature vector, statistically analyzing the store's historical average daily sales in the same period, combining the average price of a single product and the fixed cost rate to calculate the basic revenue, and calculating the fixed cost rate based on the operating costs and warehousing costs in the store's historical transaction data.

[0011] As a further improvement of the present invention, the incremental revenue calculation in the dynamic adaptation step includes: using the average daily sales volume of the same period in history as a benchmark, combining the historical price elasticity coefficient and the activity discount rate to calculate the sales growth to obtain incremental revenue; integrating the activity traffic growth and activity conversion improvement rate obtained by nonlinear mapping of personalized coding feature vectors; obtaining the revenue corresponding to the total sales volume of the activity through multi-dimensional multiplication operation; and determining the final incremental revenue after deducting the basic revenue.

[0012] As a further improvement of the present invention, the adaptive dynamic deduction step further includes: calculating the estimated excess inventory based on the difference between the expected sales volume derived from the total revenue feature vector and the current inventory threshold; calculating the unit storage cost based on the warehousing cost parameters in the store operation characteristics; calculating the profit dilution amount by using the difference between the activity discount rate and the gross profit of a single product and the activity sales volume; calculating the expected user increase cost by combining the proportion of high-value users and the user retention coefficient in the user value stratification indicators; and calculating the risk cost factor based on the estimated excess inventory, unit storage cost, profit dilution amount, and expected user increase cost.

[0013] As a further improvement of the present invention, the calculation of the risk cost factor includes: constructing a dynamic cost correlation matrix by combining the inventory turnover capacity in the store's operational characteristics, the user loyalty coefficient in the user value stratification index, and the industry risk transmission model; the correlation matrix quantifies the mutual influence coefficients between the estimated excess inventory, unit storage cost, profit dilution amount, and user expected increase cost; setting a time decay factor based on the activity cycle, dynamically adjusting the inventory backlog cost according to the activity's progress stage, and adjusting the user expected increase cost in the opposite direction according to the time decay coefficient; introducing a risk threshold triggering mechanism, when any cost item reaches the store's preset risk threshold, automatically increasing the weight ratio of that cost item and the cost items with transmission correlation through the correlation matrix, and summing the product hierarchy of the original cost item value, mutual influence coefficient, time decay factor, and dynamic weight to obtain the risk cost factor, the dynamic weight being adaptively allocated based on the store's risk tolerance, the priority of activity objectives, and the real-time risk status.

[0014] As a further improvement of the present invention, the adaptation decision strategy includes a preset adaptation decision threshold system and multi-dimensional weight rules corresponding to the activity prediction results. The decision threshold system includes individual thresholds corresponding to the expected sales volume, profit, and number of new high-value users. The multi-dimensional weight rules adaptively allocate the weight ratio of each expected indicator based on the strategic goals in the merchant preference parameters. The actual predicted value of each expected indicator is quantitatively compared with the corresponding individual threshold to obtain the degree of compliance of each indicator. The degree of compliance of all indicators is comprehensively calculated by combining the weights allocated by the multi-dimensional weight rules to obtain the overall adaptation score. The overall adaptation score is compared with the preset adaptation threshold. When the overall adaptation score reaches or exceeds the preset adaptation threshold, the judgment result of activity adaptation and the corresponding optimal combination of activity parameters are output. When the overall adaptation score does not reach the preset adaptation threshold, the judgment result of activity incompatibility is output.

[0015] The beneficial effects of this invention are as follows: This invention solves the problems of existing e-commerce activities, such as reliance on experience, homogenized feature processing, one-sided benefit assessment, and lack of deviation correction, through multi-source data standardization governance, personalized dynamic prediction model construction, and multi-dimensional dynamic benefit extrapolation. It dynamically adjusts model parameters based on store operating characteristics and strategic goals to adapt to the personalized needs of stores of different industries and sizes. It decouples and integrates basic benefits, incremental benefits, and various risk costs, avoiding decision-making biases caused by single-dimensional evaluation. Through personalized correction of historical activity deviations and dynamic weight adjustment, it reduces the gap between predicted values ​​and actual results. The intelligent decision-making logic adaptively allocates indicator weights based on the merchant's strategic goals, outputting targeted adaptation conclusions and optimization suggestions, reducing merchant operational risks, and improving marketing activity benefits and resource utilization efficiency. Attached Figure Description

[0016] Figure 1This is a flowchart of an intelligent adaptation method for e-commerce activities based on multi-dimensional revenue dynamic extrapolation, according to the present invention. Detailed Implementation

[0017] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Identical components are denoted by the same reference numerals. It should be noted that the terms "front," "rear," "left," "right," "upper," and "lower" used in the following description refer to directions in the accompanying drawings, and the terms "bottom surface," "top surface," "inner," and "outer" refer to directions toward or away from the geometric center of a specific component, respectively.

[0018] This invention proposes an intelligent adaptation method for e-commerce activities based on multi-dimensional dynamic revenue extrapolation, such as... Figure 1 As shown, it includes: The data collection and processing steps involve collecting historical transaction data, user behavior event stream data, market trend data, and merchant preference parameters from the target store as raw data. Missing value processing and outlier processing are performed on the raw data, and key feature vectors are extracted, including historical price elasticity coefficients, user value stratification indicators, store operation characteristics, and market environment characteristics.

[0019] Data acquisition aims to integrate multi-source data, employing appropriate acquisition technologies and source channels tailored to the characteristics of different data types to ensure data integrity, timeliness, and effectiveness.

[0020] Historical transaction data is collected by connecting to the data interface opened by the e-commerce platform. The collection scope covers data directly related to the operation, such as product sales records, transaction amounts, payment methods, and order status within a preset period of the target store. During the collection process, a data verification mechanism is used to remove redundant data that is transmitted repeatedly, and only valid transaction records confirmed by the platform are retained.

[0021] User behavior event stream data is achieved by deploying a behavior collection module on the front-end page of the e-commerce platform and user terminal applications. This module captures user behaviors related to the target store in real time, such as browsing, clicking, adding to cart, favoriting, and placing orders. It records the time, context, and associated product information of the behavior. After collection, the behavior data is stored in an orderly manner according to the time sequence to ensure the continuity of the behavior trajectory.

[0022] Market trend data is collected through a multi-channel integration approach. On the one hand, it obtains macro data such as the overall transaction volume, growth rate, and category distribution of the industry by subscribing to standardized data products from authoritative data service agencies in the industry. On the other hand, it obtains publicly available industry dynamic information through network data collection technology. After data deduplication and format standardization, a unified market trend dataset is formed.

[0023] Merchant preference parameters are collected through a visual interactive interface provided by the e-commerce platform. The interface settings include a series of options related to activity decisions, such as the core business objectives that merchants focus on, the acceptable range of activity costs, and the preferred activity formats. After the merchant completes the selection, the system automatically converts the interactive information into standardized parameters to ensure that the parameters can be directly used for subsequent model calculations.

[0024] Specifically, such as Figure 1 As shown, the data acquisition and processing steps include: identifying abnormal data through box plot analysis and selecting data to delete based on data importance; filling time series data using forward and backward filling methods; filling numerical data using mean processing; and filling categorical data using mode processing.

[0025] Anomaly identification employs box plot analysis, which quantifies data distribution characteristics to define reasonable data intervals, thereby accurately identifying outliers that deviate from the normal distribution.

[0026] During implementation, the original data of a single dimension is first sorted and organized, and the first quartile, the second quartile (i.e. the median), and the third quartile are divided according to the distribution ratio of the total data. The difference between the third quartile and the first quartile is used as the interquartile range, thereby determining the normal distribution range of the data. The upper limit of this range is the sum of the third quartile and the preset multiple of the interquartile range, and the lower limit is the difference between the first quartile and the preset multiple of the interquartile range.

[0027] Each data point to be analyzed is compared with the aforementioned normal distribution range. Data exceeding the upper or lower limit of the range is identified as anomalous data. The importance of data is determined based on its impact on subsequent feature extraction and model calculation. Data directly related to core features and affecting the model's prediction accuracy is considered important data, and such data is deleted directly after being confirmed as anomalous. Data related to minor features or with minimal impact on model calculation is considered unimportant data, and such data can be processed using subsequent filling methods to avoid insufficient data sample size due to data deletion.

[0028] Missing data imputation employs targeted imputation strategies based on data type differences to ensure that the imputed data maintains the distribution characteristics and inherent correlations of the original data without introducing additional bias.

[0029] Time series data imputation is based on the temporal correlation of the data. Time series data refers to continuous data recorded in chronological order, such as daily sales or weekly visitor counts. Forward imputation is suitable for scenarios where valid data exists immediately preceding a missing data point; it directly uses the valid data from that previous moment as the filler value. Backward imputation is suitable for scenarios where valid data exists immediately following a missing data point; it uses the valid data from that subsequent moment to fill the missing position, ensuring the continuity and trend consistency of the time series.

[0030] Numerical data is imputed using the mean. Numerical data refers to data that can be quantified, such as average order value, inventory quantity, and discount percentage. The process begins by filtering all valid data within the same dimension. The mean is then calculated based on the average level of all valid data and used as the imputed value for missing positions. The mean reflects the overall distribution center of the data, ensuring that the imputed value matches the overall characteristics of the original data.

[0031] Categorical data is imputed using the mode method. Categorical data refers to data used to distinguish different category attributes, such as product category, user location, and activity type. During implementation, the frequency of occurrence of each category under the same dimension is counted, and the category with the highest frequency is selected as the mode. This mode is then used to fill in the missing positions, ensuring that the category distribution after imputed matches the category distribution characteristics of the original data.

[0032] Specifically, such as Figure 1 As shown, the calculation of the key feature vector includes: obtaining the historical price elasticity coefficient by fitting the data of past discounts and sales changes of store products using the least squares method; grouping user behavior data by clustering algorithm and calculating user value stratification indicators by combining average purchase frequency and average order value; calculating average gross profit margin, inventory turnover days, and visitor conversion rate as store operation features based on historical operating data; and extracting industry popularity index, holiday weight coefficient, and competitor activity data as market environment features.

[0033] The historical price elasticity coefficient is extracted based on the store's past discount and sales data. First, records of discount changes and corresponding sales changes for the same product across different time periods are selected to ensure a match in the time dimensions of the two sets of data. The least squares method is used for fitting calculations, and the fitting parameters are adjusted to minimize the deviation between the fitting result and the actual data. This quantifies the correlation between discount changes and sales changes, ultimately yielding a historical price elasticity coefficient that reflects the sensitivity of product sales to discounts.

[0034] The extraction of user value stratification indicators is based on user behavior data. First, a clustering algorithm is used to group all user behavior data. Clustering is based on multi-dimensional behavioral characteristics such as browsing time, click frequency, add-to-cart frequency, and purchase history, grouping users with similar behavioral characteristics into the same group. For each group, average purchase frequency and average order value are calculated. Average purchase frequency reflects the group's user activity level, while average order value reflects the group's user spending power. Combining the comprehensive evaluation results of these two indicators, different levels of user value stratification indicators are derived.

[0035] The extraction of store operation characteristics is based on the store's historical operating data. The average gross profit margin is calculated by dividing the difference between the store's total revenue and total cost within a preset period by the total revenue, reflecting the store's profitability. The inventory turnover days are calculated by dividing the store's average inventory quantity by the average daily sales quantity within a preset period, reflecting the store's inventory turnover efficiency. The visitor conversion rate is calculated by dividing the number of customers who place orders by the total number of visitors within a preset period, reflecting the store's ability to convert visitors into consumers. These three indicators together constitute the store's operation characteristics.

[0036] The extraction of market environment characteristics combines macro-level industry data with competitive environment data. The industry popularity index is obtained by integrating multi-dimensional data such as overall transaction volume, search volume, and attention within the industry, and after standardization processing, it reflects the overall prosperity of the industry. The holiday weight coefficient is set according to the degree of impact of holidays on industry consumer demand in different time periods; the higher the degree of impact, the larger the weight coefficient. Competitor activity data is obtained by monitoring the public marketing activities of competitors in the same industry, and sorting out key information such as activity format, discount level, and activity cycle to form competitor activity data that can reflect the competitive environment. The three together constitute the market environment characteristics.

[0037] The prediction model construction steps include building an activity fit prediction model. The activity fit prediction model architecture includes a feature dynamic encoding layer, a revenue factor decoupling layer, a personalized bias correction layer, and a multi-dimensional output layer. The key feature vectors are integrated to generate store-specific feature vectors, and the platform's default activity parameters and merchant preference parameters are integrated to generate activity parameter vectors. These are input into the activity fit prediction model. Dynamic weights are calculated based on the semantic similarity between the activity objective and the features, and the degree of influence of the features on the revenue output. The initial input matrix is ​​weighted and encoded using dynamic weights to obtain a personalized encoded feature vector.

[0038] Specifically, such as Figure 1As shown, the calculation of the dynamic weights includes: clarifying the correlation dimensions corresponding to each feature by combining the store's personalized operation characteristics and activity objectives; quantifying the semantic similarity between each feature and the activity objectives by using cosine similarity; determining the degree of influence of each feature on revenue output based on the industry's general data pre-trained influence factor library and integrating the actual effect of each feature on revenue output in the store's historical activities; multiplying the semantic similarity of a single feature with its corresponding degree of influence to obtain the comprehensive contribution value of that feature; and performing linear normalization on the comprehensive contribution values ​​of all features to complete the allocation of dynamic weights for each feature.

[0039] The adaptation dynamic extrapolation step decouples the personalized coding feature vector to obtain the basic revenue, incremental revenue, and risk cost factor. The weights are adjusted according to the store's risk tolerance and the priority of the activity objectives, and the total revenue feature vector is regenerated. After correction based on historical activity deviation data, the activity prediction result is output. The activity prediction result includes expected sales, profit, and the number of new high-value users.

[0040] The return factor decoupling takes personalized encoded feature vectors as input and obtains two positive return factors, basic return and incremental return, as well as a negative cost factor, risk cost factor, through targeted quantitative calculations, ensuring the accurate separation of return and cost.

[0041] Specifically, such as Figure 1 As shown, the basic revenue calculation of the adaptation dynamic deduction step includes: using the historical price elasticity coefficient correlation data in the key feature vector, statistically analyzing the store's historical average daily sales in the same period, combining the average price of individual products and the fixed cost rate to calculate the basic revenue, and calculating the fixed cost rate based on the operating costs and warehousing costs in the store's historical transaction data.

[0042] The basic revenue reflects the benchmark revenue level that the target store could have obtained based on historical operating patterns without this promotion. Its calculation logic revolves around the correlation of historical data and the calculation of core parameters.

[0043] By correlating historical price elasticity coefficients in key feature vectors, a historical timeframe is defined. This timeframe must align with the current campaign's duration, seasonal characteristics, and market environment, eliminating confounding factors such as special marketing activities and unforeseen events from previous periods to ensure statistical comparability. Based on this defined timeframe, all valid sales records for the corresponding time period are extracted, and the total sales volume for that period is calculated. This total sales volume is then divided by the number of days in the period to obtain the historical average daily sales volume for the same period.

[0044] The average price of a single item is obtained by dividing the total effective transaction amount of the corresponding product in the target store during the same period in history by the sales volume during the same period, ensuring that the average price can reflect the normal pricing level of the product.

[0045] The calculation of the fixed cost ratio is based on the operating costs and warehousing costs in the store's historical transaction data. Operating costs include fixed expenses such as personnel salaries, platform service fees, and equipment depreciation incurred in the daily operation of the store. Warehousing costs include fixed expenses such as warehouse rental fees, warehouse management labor costs, and goods storage fees. The total fixed cost is obtained by summing the two types of costs. Then, the total fixed cost is divided by the total revenue in the same period of the previous year to obtain the fixed cost ratio. This ratio reflects the proportion of the store's fixed costs in its revenue.

[0046] The basic revenue is calculated by correlating historical average daily sales, average price per item, and fixed cost rate. The core logic is to use the average daily sales as a basis, combine it with the average price per item to obtain the average daily revenue, and then deduct the corresponding fixed cost portion from the average daily revenue to obtain the average daily basic revenue. If the activity period is multiple calendar days, the total basic revenue is obtained by accumulating the number of days in the activity period.

[0047] Specifically, such as Figure 1 As shown, the incremental revenue calculation in the adaptive dynamic deduction step includes: using the historical average daily sales volume as a benchmark, combining the historical price elasticity coefficient and the activity discount rate to calculate the sales growth to obtain incremental revenue; integrating the activity traffic growth and activity conversion improvement rate obtained through nonlinear mapping of personalized coding feature vectors; obtaining the revenue corresponding to the total sales volume of the activity through multi-dimensional multiplication operation; and determining the final incremental revenue after deducting the basic revenue.

[0048] Using historical average daily sales as the benchmark, the historical price elasticity coefficient reflects the sensitivity of product sales to discount changes. The promotional discount rate is the discount percentage set for this promotion. Through correlation analysis between the two, the effect of the promotional discount on sales is determined, resulting in a sales increase percentage based on price factors. For example, the larger the absolute value of the price elasticity coefficient, the more sensitive the product is to discounts, and the higher the sales increase percentage under the same discount rate.

[0049] The activity traffic increase and conversion rate improvement are derived from the personalized coding feature vector using a non-linear transformation algorithm. This personalized coding feature vector contains multi-dimensional feature information such as store operations, market environment, and user behavior. The non-linear transformation algorithm can uncover the complex relationships between features, transforming the abstract feature vector into concrete increases in traffic and conversion metrics. Specifically, the activity traffic increase reflects the percentage increase in the number of store visitors during the activity period compared to the same period in previous years, while the activity conversion rate improvement reflects the percentage increase in the proportion of visitors who placed orders during the activity period compared to the same period in previous years.

[0050] The calculation of incremental revenue follows a progressive logic: First, based on the historical average daily sales volume for the same period, combined with the sales growth rate based on price factors, a preliminary sales volume considering the impact of discounts is obtained; then, the preliminary sales volume is multiplied by the increase in activity traffic to obtain the sales volume considering traffic growth; subsequently, this sales volume is multiplied by the activity conversion rate improvement to obtain the average daily total sales volume during the activity period; based on the average daily total sales volume and the average price of individual products, the average daily total revenue of the activity is calculated; finally, the average daily basic revenue is subtracted from the average daily total revenue of the activity to obtain the average daily incremental revenue, which is accumulated over the number of days in the activity period to obtain the total incremental revenue.

[0051] Specifically, such as Figure 1 As shown, the adaptive dynamic deduction step further includes: calculating the estimated excess inventory based on the difference between the expected sales volume derived from the total revenue feature vector and the current inventory threshold; calculating the unit storage cost based on the warehousing cost parameters in the store operation characteristics; calculating the profit dilution amount by using the difference between the activity discount rate and the single product gross profit and the activity sales volume; calculating the expected user increase cost by combining the proportion of high-value users and the user retention coefficient in the user value stratification indicators; and calculating the risk cost factor based on the estimated excess inventory, unit storage cost, profit dilution amount, and expected user increase cost.

[0052] The calculation of estimated excess inventory is based on the expected sales volume derived from the total revenue feature vector. The current inventory threshold is a reasonable upper limit for inventory determined by factors such as the store's historical inventory turnover capacity, safety stock level, and supply chain replenishment cycle, used to avoid the risk of inventory backlog or stockouts. When the expected sales volume is lower than the current inventory threshold, the excess portion is the estimated excess inventory, and this portion of inventory will incur additional warehousing and management costs.

[0053] Unit storage cost is extracted from the warehousing cost parameters of store operation characteristics. This parameter is calculated based on data such as the store's historical total warehousing cost, warehouse storage capacity, and average inventory quantity, and can accurately reflect the warehousing cost burden of a single product.

[0054] The core of calculating profit dilution lies in quantifying the impact of discounts on the gross profit of a single item. The gross profit of a single item is the amount after deducting the variable cost of the single item from the normal selling price of the product. The discount rate of the promotion leads to a decrease in the actual selling price of the product, which in turn reduces the actual gross profit of the single item. The product of the reduced amount and the sales volume of the promotion is the profit dilution amount, which reflects the degree to which the promotion discount erodes the overall profit.

[0055] The calculation of user expectation-driven cost is closely related to user value segmentation indicators. First, the scope and proportion of high-value users are determined based on these indicators. Then, the user retention coefficient is calculated by combining historical retention data of these users. This coefficient reflects the probability of continued retention of high-value users after the promotion. User expectation-driven cost refers to the potential cost arising from the promotion's discounts leading to higher user expectations for subsequent product pricing, which may affect users' willingness to purchase at normal pricing. The specific amount of this potential cost is quantified through correlation analysis between the proportion of high-value users and the user retention coefficient.

[0056] Specifically, such as Figure 1 As shown, the calculation of the risk cost factor includes: constructing a dynamic cost correlation matrix by combining the inventory turnover capacity in store operation characteristics, the user loyalty coefficient in user value stratification indicators, and the industry risk transmission model. The correlation matrix quantifies the mutual influence coefficients between the estimated excess inventory, unit storage cost, profit dilution amount, and user expected increase cost; setting a time decay factor based on the activity cycle, dynamically adjusting the inventory backlog cost according to the activity progress stage, and adjusting the user expected increase cost in the opposite direction according to the time decay coefficient; introducing a risk threshold trigger mechanism, when any cost item reaches the store's preset risk threshold, automatically increasing the weight ratio of that cost item and the cost items with transmission correlation through the correlation matrix, and summing the product hierarchy of the original cost item value, mutual influence coefficient, time decay factor, and dynamic weight to obtain the risk cost factor. The dynamic weight is adaptively allocated based on the store's risk tolerance, the priority of activity objectives, and the real-time risk status.

[0057] The construction of the dynamic cost correlation matrix is ​​based on inventory turnover capacity (a key characteristic of store operations), user loyalty coefficient (a key indicator of user value segmentation), and an industry risk transmission model. Stronger inventory turnover capacity results in a lower correlation coefficient between estimated excess inventory and unit storage cost; higher user loyalty coefficients mean a smaller impact of expected user-increased costs on other cost items. The industry risk transmission model references cost transmission patterns among similar stores in the industry, providing initial correlation coefficients for the matrix. The final matrix quantifies the mutual influence among estimated excess inventory, unit storage cost, profit dilution amount, and expected user-increased costs. Changes in one cost item are transmitted to other related cost items through correlation coefficients.

[0058] The time decay factor is set based on the characteristics of different stages of the activity cycle. In the early stage of the activity, the marketing effect is gradually released, user participation steadily increases, and the risk of inventory backlog is low, so the time decay factor for inventory backlog costs is set at a low value. As the activity progresses to the middle and later stages, the inventory consumption rate slows down, the risk of backlog increases, and the time decay factor gradually increases, thereby dynamically adjusting the accounting weight of inventory backlog costs. The adjustment logic of the time decay factor for user expectations increasing costs is the opposite of that for inventory backlog costs. In the early stage of the activity, user expectations have not yet formed, the cost impact is small, and the decay factor is set at a high value. In the later stage, user expectations are solidified, the cost impact increases, and the decay factor gradually decreases.

[0059] The risk threshold triggering mechanism first presets a risk threshold for each cost item. This threshold is determined based on the store's historical risk loss data, operational capacity, and industry risk benchmarks. When the calculated value of any cost item reaches the corresponding risk threshold, the system identifies other cost items with a transmission relationship to that cost item through a dynamic cost association matrix. It then automatically increases the weight of that cost item and related cost items, ensuring that the calculated proportion of high-risk cost items in the total risk cost matches the actual risk level.

[0060] The final risk cost factor is obtained through hierarchical calculation. The calculation logic is as follows: First, the original accounting value of each cost item is multiplied by the corresponding mutual influence coefficient in the dynamic cost association matrix to obtain the adjusted value after considering the transmission effect between costs; then, the adjusted value is multiplied by the corresponding time decay factor to obtain the cost value combined with the time dynamic characteristics; subsequently, the cost value is multiplied by the dynamic weight adaptively allocated based on the store's risk tolerance, the priority of the activity target, and the real-time risk status to obtain the final accounting weight value of each cost item; finally, the final accounting weight values ​​of all cost items are summed to obtain the risk cost factor.

[0061] The generation of the total revenue feature vector is based on basic revenue, incremental revenue, and risk cost factors, with the weights of each factor adjusted according to the store's risk tolerance and the priority of activity objectives. The store's risk tolerance is quantified using indicators such as historical risk response data and financial status. The priority of activity objectives is determined based on the strategic goals in the merchant's preference parameters. If the goal is to attract new customers, the weight of incremental revenue is increased; if the goal is to maintain profits, the weight of the risk cost factor is increased. By adjusting the weights, the influence of each factor on the total revenue is matched with the merchant's needs. Finally, each factor is multiplied by its corresponding weight and summed to obtain the total revenue feature vector.

[0062] Personalized deviation correction is based on historical store activity deviation data. First, a historical deviation record database is established, storing the deviation information between predicted and actual values ​​for each past activity, including the deviation magnitude and reasons for each revenue factor and total revenue. An exponential moving average method is used to fit the trend of the deviation data, assigning higher weight to recent deviation data to obtain an initial correction coefficient that reflects the current stage's prediction deviation pattern. The confidence level of the deviation data is then combined with this data; the confidence level is calculated based on the sample size and dispersion of the deviation data—a larger sample size and smaller dispersion result in a higher confidence level. The total revenue feature vector is then used to correct the deviation using the correlation logic between the initial correction coefficient and the confidence level, resulting in a corrected total revenue feature vector. Based on the corrected total revenue feature vector, and combined with the correlation logic between each revenue factor and prediction indicators, the activity prediction result is output. The expected sales volume is derived from the total revenue feature vector, corresponding to the calculation result of the total sales volume of the activity; the profit is obtained by subtracting the risk cost factor and other related costs from the total revenue in the total revenue feature vector; the number of new high-value users is derived from the user conversion data, user value stratification indicators and historical high-value user conversion ratio corresponding to the incremental revenue. The three together constitute the complete activity prediction result, providing data support for subsequent activity decisions.

[0063] The activity decision generation step involves analyzing the activity prediction results using a preset adaptation decision strategy to obtain the final activity adaptation decision and outputting the activity adaptation judgment result.

[0064] The adaptive decision threshold system is a benchmark standard for judging whether each predicted indicator has met the merchant's expectations. Its preset process needs to combine multi-dimensional references to ensure the rationality and relevance of the threshold, rather than simply setting a fixed value.

[0065] The individual thresholds are set for three core predictive metrics: expected sales volume, profit, and the number of new high-value users. The threshold for each metric is determined through multi-level data integration and analysis. For the expected sales volume threshold, the system first statistically analyzes the target store's historical sales data during the same period without promotions and the sales performance of similar past promotions. It also refers to the sales levels of stores of similar size and product category in the industry under similar promotional scenarios. Combined with the sales growth expectations set by the merchant in the preference parameters, the system determines the individual threshold for expected sales volume that meets the merchant's basic requirements through data trend analysis and target decomposition.

[0066] The profit threshold is set based on the store's cost structure and profit target. It integrates the fixed cost and variable cost accounting results from historical operating data, refers to the industry average profit margin level, combines the minimum acceptable profit margin specified in the merchant preference parameters, and also considers the additional costs that may be generated in this event, such as promotion expenses and event subsidies. It is derived through profit balance analysis to ensure that the threshold can cover costs and achieve the preset profit target.

[0067] The threshold for the number of new high-value users is set based on the historical acquisition cost and lifetime value data of high-value users in the store, combined with the average conversion efficiency of high-value users in the industry, and also takes into account the requirements for the quality of new user acquisition in the merchant preference parameters. Through user value assessment and cost-benefit analysis, the benchmark for the number of new high-value users that can bring positive value to the long-term operation of the store is determined.

[0068] The core of the multi-dimensional weighting rule is to ensure that the weight ratio of each predictive indicator is highly matched with the merchant's strategic goals, so as to achieve personalized adaptation of decision-making. Its adaptive allocation logic revolves around the merchant's preference parameters and historical data feedback.

[0069] The system first performs an in-depth analysis of the strategic objectives in the merchant preference parameters to identify the core demand direction. If the merchant preference parameters clearly indicate a preference for expanding market share and accumulating user resources, then the strategic objective is determined to be customer acquisition, and the weight of the indicator for the number of new high-value users will be increased. If the merchant preference parameters emphasize ensuring operating revenue and controlling cost risks, then the strategic objective is determined to be profit preservation, and the weight of the profit indicator will be significantly increased. If the merchant preference parameters emphasize increasing market activity and clearing inventory, then the weight of the expected sales volume indicator will increase accordingly.

[0070] The weighting allocation is not solely based on qualitative judgments of strategic goals, but also dynamically adjusted in conjunction with historical store activity data. The system extracts activity data from similar strategic goals in the past, analyzes the actual contribution of each indicator in successful activity cases, and if historical data shows that a certain indicator has a more significant impact on the overall effect of the activity, then the weight of that indicator is appropriately increased in this weighting allocation. At the same time, the initial weights are calibrated by referring to the weighting allocation experience values ​​of similar strategic goals in the industry, ultimately forming a multi-dimensional weighting rule that both fits the needs of merchants and conforms to actual operational rules.

[0071] The quantitative comparison of the degree of compliance is the process of accurately matching and analyzing the predicted indicators with the thresholds, while the comprehensive score calculation is the core link that integrates the compliance status and weight of each indicator. Together, they constitute the quantitative basis for decision-making.

[0072] During the quantitative comparison process, for each predicted indicator, the system performs correlation analysis between the actual predicted value and the corresponding individual threshold to determine the achievement level of that indicator. For the expected sales volume indicator, the degree of sales achievement is determined by analyzing the relationship and the magnitude of the difference between the predicted value and the individual threshold. If the predicted value is significantly higher than the threshold, the degree of achievement is high; if the predicted value is close to the threshold, the degree of achievement is medium; if the predicted value is lower than the threshold, the degree of achievement is reduced accordingly based on the magnitude of the difference.

[0073] The quantitative logic for achieving the target levels of profit and the number of new high-value users is consistent with that of expected sales. Both are based on the relative levels of predicted values ​​and individual thresholds to ensure that the achievement of each indicator can be accurately quantified and to avoid ambiguous judgments.

[0074] The calculation of the overall fit score follows a weighted integration logic. The system first converts the compliance level of each indicator into a standardized quantitative value. Then, it correlates this value with the corresponding weights assigned in the multi-dimensional weighting rules, ensuring that the higher the weight of an indicator, the more significant its impact on the overall score. Subsequently, the system integrates the results of all weighted calculations to form an overall fit score that comprehensively reflects the overall fit level of the activity. The score directly corresponds to the degree of fit between the activity and the store.

[0075] Specifically, such as Figure 1 As shown, the adaptation decision strategy includes a preset adaptation decision threshold system and multi-dimensional weight rules corresponding to the activity prediction results. The decision threshold system includes individual thresholds corresponding to the expected sales volume, profit, and number of new high-value users. The multi-dimensional weight rules adaptively allocate the weight ratio of each expected indicator based on the strategic goals in the merchant preference parameters. The actual predicted value of each expected indicator is quantitatively compared with the corresponding individual threshold to obtain the degree of compliance of each indicator. The degree of compliance of all indicators is comprehensively calculated by combining the weights allocated by the multi-dimensional weight rules to obtain the comprehensive adaptation score. The comprehensive adaptation score is compared with the preset adaptation threshold. When the comprehensive adaptation score reaches or exceeds the preset adaptation threshold, the judgment result of activity adaptation and the corresponding optimal combination of activity parameters are output. When the comprehensive adaptation score does not reach the preset adaptation threshold, the judgment result of activity incompatibility is output.

[0076] The fit assessment is based on the comparison between the overall fit score and the preset fit threshold. Combined with the quantitative analysis results, it outputs targeted decision conclusions and supporting solutions to ensure the practicality and guidance of the decision.

[0077] The preset adaptation threshold is the core benchmark for distinguishing whether an activity is adapted or not. It is set with reference to the general qualified standards for activity adaptation in the industry, combined with the lowest comprehensive score of successful adaptation cases in the store's historical activities, and also takes into account the merchant's tolerance for the activity effect. It is determined after multiple rounds of data calibration, so that neither setting the threshold too high will cause valid activities to be missed, nor lowering the standard will cause unsuitable activities to be misjudged.

[0078] When the overall adaptation score reaches or exceeds the preset adaptation threshold, the system determines that the activity is compatible with the store, and outputs the activity adaptation result. Simultaneously, the system extracts the best-performing set of activity parameters during the prediction and simulation process, including the activity discount rate, activity period, and promotional resource allocation. This set of parameters is determined based on the simulation result where each prediction indicator reaches or exceeds its corresponding threshold and has the highest overall score. This optimal combination of activity parameters is then output together, providing direct reference for merchants to implement their activities.

[0079] When the overall adaptation score fails to reach the preset adaptation threshold, the system determines that the activity and the store are not compatible, and outputs the result of the mismatch. Simultaneously, the system performs backtesting analysis on the achievement level and weighting of each predicted indicator, identifying core indicators with low achievement levels and high weighting that are not met. Based on the reasons for these non-compliance, such as insufficient profit due to excessively high discount rates leading to profit dilution, or insufficient new high-value users due to insufficient precision in promotional channels, the system generates targeted improvement suggestions, providing guidance for merchants to adjust their activity plans or optimize their business strategies.

[0080] The foregoing has illustrated and described the basic features, principles, and advantages of the present invention. It should be noted that the present invention is not limited to the above embodiments, but only to some embodiments. Any improvements and additions made without departing from the spirit and scope of the present invention are considered to be within the scope of protection of the present invention.

Claims

1. A method for intelligent adaptation of e-commerce activities based on multi-dimensional revenue dynamic extrapolation, characterized in that, include: The data collection and processing steps involve collecting historical transaction data, user behavior event stream data, market trend data, and merchant preference parameters from the target store as raw data. Missing values ​​and outliers are processed on the raw data, and key feature vectors are extracted, including historical price elasticity coefficients, user value stratification indicators, store operation characteristics, and market environment characteristics. The prediction model construction steps include building an activity fit prediction model. The activity fit prediction model architecture includes a feature dynamic encoding layer, a revenue factor decoupling layer, a personalized deviation correction layer, and a multi-dimensional output layer. The key feature vectors are integrated to generate store-specific feature vectors, and the platform activity default parameters and merchant preference parameters are integrated to generate activity parameter vectors. These are input into the activity fit prediction model. Dynamic weights are calculated based on the semantic similarity between the activity target and the features, and the degree of influence of the features on the revenue output. The initial input matrix is ​​weighted and encoded using dynamic weights to obtain a personalized encoded feature vector. The adaptive dynamic deduction step decouples the personalized coding feature vector to obtain the basic revenue, incremental revenue and risk cost factor. The weights are adjusted according to the store's risk tolerance and the priority of the activity objectives, and the total revenue feature vector is regenerated. After correction by combining historical activity deviation data, the activity prediction result is output. The activity prediction result includes expected sales, profit and the number of new high-value users. The activity decision generation step involves analyzing the activity prediction results using a preset adaptation decision strategy to obtain the final activity adaptation decision and outputting the activity adaptation judgment result.

2. The intelligent adaptation method for e-commerce activities based on multi-dimensional revenue dynamic extrapolation according to claim 1, characterized in that, The data acquisition and processing steps include identifying abnormal data through box plot analysis, selecting data to delete based on data importance, filling time series data using forward and backward filling methods, filling numerical data using mean processing, and filling categorical data using mode processing.

3. The intelligent adaptation method for e-commerce activities based on multi-dimensional revenue dynamic extrapolation according to claim 1, characterized in that, The calculation of the key feature vector includes: obtaining the historical price elasticity coefficient by fitting the data of past discounts and sales changes of store products using the least squares method; grouping user behavior data by clustering algorithm and calculating user value stratification indicators by combining average purchase frequency and average order value; calculating average gross profit margin, inventory turnover days, and visitor conversion rate as store operation features based on historical operating data; and extracting industry popularity index, holiday weight coefficient, and competitor activity data as market environment features.

4. The intelligent adaptation method for e-commerce activities based on multi-dimensional revenue dynamic extrapolation according to claim 1, characterized in that, The calculation of the dynamic weights includes: clarifying the correlation dimensions of each feature by combining the store's personalized operation characteristics and activity objectives; quantifying the semantic similarity between each feature and the activity objectives by using cosine similarity; determining the degree of influence of each feature on revenue output based on the industry's general data pre-trained influence factor library and integrating the actual effect of each feature on revenue output in the store's historical activities; multiplying the semantic similarity of a single feature with its corresponding degree of influence to obtain the comprehensive contribution value of that feature; and performing linear normalization on the comprehensive contribution values ​​of all features to complete the allocation of dynamic weights for each feature.

5. The intelligent adaptation method for e-commerce activities based on multi-dimensional revenue dynamic extrapolation according to claim 1, characterized in that, The basic revenue calculation in the adaptation dynamic deduction step includes: using the historical price elasticity coefficient correlation data in the key feature vector, statistically analyzing the store's historical average daily sales during the same period, combining the average price of individual products and the fixed cost rate to calculate the basic revenue, and calculating the fixed cost rate based on the operating costs and warehousing costs in the store's historical transaction data.

6. The intelligent adaptation method for e-commerce activities based on multi-dimensional revenue dynamic extrapolation according to claim 1, characterized in that, The incremental revenue calculation in the adaptation dynamic extrapolation step includes: using the average daily sales volume of the same period in history as a benchmark, combining the historical price elasticity coefficient and the activity discount rate to calculate the sales growth to obtain incremental revenue; integrating the activity traffic growth and activity conversion improvement rate obtained through nonlinear mapping of personalized coding feature vectors; obtaining the revenue corresponding to the total sales volume of the activity through multi-dimensional multiplication operation; and determining the final incremental revenue after deducting the basic revenue.

7. The intelligent adaptation method for e-commerce activities based on multi-dimensional revenue dynamic extrapolation according to claim 1, characterized in that, The adaptive dynamic deduction step also includes: calculating the estimated excess inventory based on the difference between the expected sales volume derived from the total revenue feature vector and the current inventory threshold; calculating the unit storage cost based on the warehousing cost parameters in the store operation characteristics; calculating the profit dilution amount by using the difference between the activity discount rate and the gross profit of a single product and the activity sales volume; calculating the expected user increase cost by combining the proportion of high-value users and the user retention coefficient in the user value stratification indicators; and calculating the risk cost factor based on the estimated excess inventory, unit storage cost, profit dilution amount, and expected user increase cost.

8. The intelligent adaptation method for e-commerce activities based on multi-dimensional revenue dynamic extrapolation according to claim 7, characterized in that, The risk cost factor calculation includes constructing a dynamic cost correlation matrix by combining inventory turnover capacity from store operational characteristics, user loyalty coefficient from user value stratification indicators, and an industry risk transmission model. This correlation matrix quantifies the mutual influence coefficients between estimated excess inventory, unit storage cost, profit dilution amount, and expected user-increased costs. A time decay factor is set based on the activity cycle, with inventory backlog costs dynamically adjusted according to the activity's progress stage, and expected user-increased costs adjusted inversely according to the time decay factor. The risk threshold triggering mechanism automatically increases the weight of the cost item and related cost items through the correlation matrix when any cost item reaches the store's preset risk threshold. The risk cost factor is obtained by summing the product of the original cost item value, mutual influence coefficient, time decay factor, and dynamic weight. The dynamic weight is adaptively allocated based on the store's risk tolerance, the priority of the activity target, and the real-time risk status.

9. The intelligent adaptation method for e-commerce activities based on multi-dimensional revenue dynamic extrapolation according to claim 1, characterized in that, The adaptation decision strategy includes a preset adaptation decision threshold system and multi-dimensional weight rules corresponding to the activity prediction results. The decision threshold system includes individual thresholds corresponding to the expected sales volume, profit, and number of new high-value users. The multi-dimensional weight rules adaptively allocate the weight ratio of each expected indicator based on the strategic goals in the merchant preference parameters. The actual predicted value of each expected indicator is quantitatively compared with the corresponding individual threshold to obtain the degree of compliance of each indicator. The degree of compliance of all indicators is comprehensively calculated by combining the weights allocated by the multi-dimensional weight rules to obtain the overall adaptation score. The overall adaptation score is compared with the preset adaptation threshold. When the overall adaptation score reaches or exceeds the preset adaptation threshold, the judgment result of activity adaptation and the corresponding optimal combination of activity parameters are output. When the overall adaptation score does not reach the preset adaptation threshold, the judgment result of activity incompatibility is output.

Images