A suggestion generation method based on enterprise data differentiation analysis

Abstract

The application provides a suggestion generation method based on enterprise data differentiation analysis, comprising: obtaining product sales data of an enterprise from a same online transaction platform, wherein the product sales data comprises order data, product data and user data; preprocessing the order data, the product data and the user data; performing associated aggregation analysis on the preprocessed order data, product data and user data to construct a differentiation stage sample; and performing differentiation analysis on the differentiation stage sample to determine product promotion suggestions. Under the condition of limited data dimensions, a set of data processing schemes is constructed to deeply aggregate three types of data, i.e. orders, exposures and user basic information, and an analysis mode with user behavior as the core is constructed, so as to generate product promotion suggestions through intelligent analysis under limited data, and assist enterprises to complete the product promotion stage as much as possible.

Description

Technical Field

[0001] This application relates to the field of online product promotion technology, and more specifically, to a suggestion generation method based on enterprise data differentiation analysis. Background Technology

[0002] The e-commerce sector is developing rapidly, with many large e-commerce platforms serving as product promotion channels. Many companies are also choosing to promote their products on e-commerce platforms, gradually reducing the proportion of traditional distribution methods.

[0003] In the process of promoting products on e-commerce platforms, many factors influence product sales. Although partnerships between businesses and platforms can provide some key data (such as sales figures, page views, clicks, favorites, and orders), much data involving user privacy (such as detailed user profiles) is unavailable; only publicly available basic data is accessible. How to conduct intelligent analysis with limited data to provide more suitable product promotion suggestions and improve product sales and profits (depending on the company's goals at different stages of product promotion) is a technical problem that needs to be solved in this field. Summary of the Invention

[0004] The purpose of this application is to provide a suggestion generation method based on enterprise data differentiation analysis, so as to generate product promotion suggestions through intelligent analysis with limited data, and to assist enterprises in achieving their product promotion goals as much as possible.

[0005] To achieve the above objectives, the embodiments of this application are implemented in the following manner:

[0006] This application provides a suggestion generation method based on enterprise data differentiation analysis, including: acquiring product sales data of the enterprise from the same online trading platform, wherein the product sales data includes order data, product data, and user data; preprocessing the order data, product data, and user data; performing correlation and aggregation analysis on the preprocessed order data, product data, and user data to construct a differentiation stage sample; and performing differentiation analysis on the differentiation stage sample to determine product placement and promotion suggestions.

[0007] Furthermore, correlation and aggregation analysis is performed on the preprocessed order data, product data, and user data to construct differentiated stage samples. This includes: analyzing the preprocessed order data to determine the order information for each order, including order amount, discount level, product quantity, product pricing, and order user ID; analyzing the preprocessed product data to determine the product exposure information, including all user IDs who viewed the product, clicked on it, added it to their favorites, and ordered it; analyzing the preprocessed user data to determine the user information for each user, including user ID and basic information, covering all users who viewed the product; and performing correlation and aggregation analysis on all order information, exposure information, and user information to construct differentiated stage samples.

[0008] Furthermore, a correlation and aggregation analysis is performed on all order information, exposure information, and user information to construct differentiated stage samples. This includes: aggregating order information, exposure information, and user information with user ID as the core to generate user behavior data corresponding to each user ID; dividing users into several user categories based on basic user information and assigning user IDs to the corresponding user categories, where basic information includes age and gender; and extracting features from the user behavior data in each user category to construct differentiated stage samples.

[0009] Furthermore, feature extraction is performed on user behavior data in each user category to construct differentiated stage samples, including: for each user category: statistics on browsing users, clicking users, favorite users, ordering users, order amount, discount level, product quantity, and product pricing under the current user category; for each user behavior data under the current user category: based on user behavior data, differentiated stage samples are constructed, including browsing-click stage samples, click-order stage samples, click-favorite stage samples, and favorite-order stage samples.

[0010] Furthermore, based on user behavior data, differentiated stage samples are constructed, including: generating browse-click stage samples based on all user behavior data under the current user category, wherein the browse-click stage samples include valid click samples and invalid click samples, where valid click samples indicate that the user clicked after browsing, and invalid click samples indicate that the user did not click after browsing; generating click-order stage samples based on all user behavior data under the current user category, wherein the click-order stage samples include valid order samples and invalid order samples, where valid order samples indicate that the user placed an order after clicking, and invalid order samples indicate that the user placed an order after clicking. This indicates that the user did not place an order after clicking. Based on all user behavior data under the current user category, a click-to-favorite phase sample is generated. The click-to-favorite phase sample includes valid favorites and invalid favorites. Valid favorites indicate that the user added the item to their favorites after clicking, while invalid favorites indicate that the user did not add the item to their favorites after clicking. Based on all user behavior data under the current user category, a favorite-to-order phase sample is generated. The favorite-to-order phase sample includes valid direct purchases and invalid direct purchases. Valid direct purchases indicate that the user placed an order after adding the item to their favorites, while invalid direct purchases indicate that the user did not place an order after adding the item to their favorites.

[0011] Furthermore, if the samples in the differentiation stage are from the browsing-click stage, a differentiation analysis is performed on these samples to determine product placement and promotion recommendations. This includes: for each user category: dividing the product into multiple pricing ranges based on product pricing; determining the number of valid and invalid click samples within each pricing range; generating observation frequency tables and expected frequency tables based on the number of valid and invalid click samples within each pricing range; using a chi-square test calculator to perform chi-square tests on the observation and expected frequency tables to obtain the chi-square test analysis results; determining the target user category based on the chi-square test analysis results for each user category; performing logistic regression analysis on the target user category to determine the pricing-click function; and generating product pricing recommendations for the target user category based on the pricing-click function.

[0012] Furthermore, if the samples in the differentiation stage are from the click-to-order stage, a differentiation analysis is performed on these samples to determine product placement and promotion recommendations. This includes: for each user category: dividing the product into multiple discount ranges based on the discount level; determining the number of valid and invalid orders within each discount range; generating observation frequency tables and expected frequency tables based on the number of valid and invalid orders within each discount range; using a chi-square test calculator to perform chi-square tests on the observation and expected frequency tables to obtain the chi-square test analysis results; determining the target user category based on the chi-square test analysis results for each user category; performing logistic regression analysis on the target user category to determine the discount-to-order function; and generating discount level recommendations for the target user category based on the discount-to-order function.

[0013] Furthermore, if the samples in the differentiation stage are from the click-to-favorite stage, a differentiation analysis is performed on these samples to determine product placement and promotion recommendations. This includes: for each user category: dividing the product into multiple pricing ranges based on product pricing; determining the number of valid and invalid favorites within each pricing range; generating observation frequency tables and expected frequency tables based on the number of valid and invalid favorites within each pricing range; using a chi-square test calculator to perform chi-square tests on the observation and expected frequency tables to obtain the chi-square test analysis results; determining the target user category based on the chi-square test analysis results for each user category; performing logistic regression analysis on the target user category to determine the pricing-favorite function; and generating product pricing recommendations for the target user category based on the pricing-favorite function.

[0014] Furthermore, if the samples in the differentiation stage are from the collection-order stage, a differentiation analysis is performed on the samples in the differentiation stage to determine product placement and promotion suggestions, including: for each user category: dividing the product into multiple discount ranges according to the discount level; determining the number of valid direct purchase samples and the number of invalid direct purchase samples within each discount range; generating an observation frequency table and an expected frequency table based on the number of valid and invalid direct purchase samples within each discount range; using a chi-square test calculator to perform chi-square test calculations on the observation frequency table and the expected frequency table to obtain the chi-square test analysis results; determining the target user category based on the chi-square test analysis results for each user category; performing logistic regression analysis on the target user category to determine the discount-direct purchase function; and generating discount level suggestions for the target user category based on the discount-direct purchase function.

[0015] Furthermore, after determining the pricing-click function, discount-order function, pricing-favorite function, and discount-direct purchase function for the same target user category, the method further includes: obtaining the product cost; inputting the product cost, pricing-click function, discount-order function, pricing-favorite function, and discount-direct purchase function into the expected profit model, wherein the expected profit model has an expected profit function: Expected Profit = a × Click-through Rate × Order Rate × (Product Price × (1 - Discount Amount) - Product Cost) + b × Click-through Rate × Favorite Rate × Direct Purchase Rate × (Product Price × (1 - Discount Amount) - Product Cost), where a and b are set weight values; solving the expected profit function using a nonlinear programming algorithm to obtain the optimal solution that maximizes the expected profit, and generating a promotion strategy plan that includes the product price value, the discount amount value, and the expected profit value.

[0016] Beneficial effects:

[0017] This solution acquires product sales data from the same online trading platform, including order data, product data, and user data. It preprocesses these data, then performs correlation and aggregation analysis to construct differentiated stage samples. Finally, it conducts differentiation analysis on these samples to determine product placement and promotion recommendations. Under the condition of limited data dimensions (inability to obtain detailed user profiles), this solution takes a different approach. By constructing a data processing scheme, it avoids the technical route of traditional e-commerce analysis that relies on the platform's user tagging system (such as purchasing power level and interest preferences). Under data constraints, it deeply aggregates three types of data: orders, exposure, and basic user information, constructing an analysis method centered on user behavior. It breaks down the complete user journey on the platform into four differentiated stages: browse-click, click-order, click-favorite, and favorite-order, and constructs a sample set containing positive and negative samples (valid and invalid) for each stage. By employing a refined sample construction approach, traditional macro-level conversion rate metrics are broken down into multiple micro-level conversion stages that can be independently analyzed and optimized, providing a data foundation for attribution analysis and strategy generation. In the specific differentiation stage analysis, a two-stage analysis approach is used: chi-square test screening and logistic regression. For each user category and behavioral stage, the solution uses a chi-square test to quickly scan discrete pricing or discount ranges. This rapidly identifies target user categories with significant differences from combinations of user category, behavioral stage, and influencing factors, completing the initial screening and avoiding the waste of resources from directly modeling all combinations. For the identified valuable targets, logistic regression is used to fit product pricing or discount intensity as a continuous variable, outputting continuous functions such as pricing-click function, discount-order function, pricing-favorite function, and discount-direct purchase function. These functions can serve as the basis for generating suggestions for differentiation stage analysis and also as the basis for subsequent comprehensive expected profit analysis.

[0018] This solution, building upon the analysis of differentiated stages, introduces an expected profit model based on maximizing expected profit. This shifts the focus of analysis from maximizing conversion rate to maximizing profit. After obtaining the pricing-click function, discount-order function, pricing-favorite function, and discount-direct purchase function, these are integrated with product costs to construct a unified expected profit function. Based on actual conditions, two conversion paths are cleverly designed: immediate conversion through direct click and order (weighted by parameter a), and delayed conversion through ordering after adding to favorites (weighted by parameter b). A nonlinear programming algorithm is used to solve the expected profit function, automatically calculating the specific product pricing and discount values ​​that maximize expected profit. This generates a directly executable, quantifiable, and optimally profitable promotion strategy, providing businesses with a more advantageous strategy for product promotion.

[0019] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 A flowchart illustrating a suggestion generation method based on enterprise data differentiation analysis provided in this application embodiment. Detailed Implementation

[0022] The technical solutions in the embodiments of this application will now be described with reference to the accompanying drawings.

[0023] Please see Figure 1 , Figure 1 A flowchart illustrating a suggestion generation method based on enterprise data differentiation analysis provided in this application embodiment.

[0024] First, proceed with step S10.

[0025] Step S10: Obtain the enterprise's product sales data from the same online trading platform, including order data, product data, and user data.

[0026] In this embodiment, when enterprises promote their products online, they can cooperate with e-commerce platforms to legally obtain product sales data, including order data (including order amount, discount level, product quantity, product pricing, order user ID, etc., as well as other data such as order number, transaction number, etc., which are not the data used in this embodiment), product data (including product pricing, discount level, etc., and users who interact with the product, such as browsing user ID, clicking user ID, favorite user ID, ordering user ID), and user data (user ID and basic information, such as age and gender, and deeper data such as user profiles and user tags that are difficult to obtain).

[0027] It should be noted that the method in this embodiment analyzes product sales data from the same online trading platform; therefore, the data source must be the same online trading platform.

[0028] After obtaining product sales data, you can proceed with step S20.

[0029] Step S20: Preprocess the order data, product data, and user data.

[0030] In this embodiment, order data, product data, and user data can be preprocessed. Preprocessing includes deduplication and outlier handling, which are conventional preprocessing methods and are not the focus of this embodiment.

[0031] After preprocessing is complete, step S30 can be run.

[0032] Step S30: Perform correlation and aggregation analysis on the preprocessed order data, product data, and user data to construct differentiated stage samples.

[0033] In this embodiment, the preprocessed order data can be analyzed to determine the order information for each order, including order amount, discount level, product quantity, product price, and order user ID. The preprocessed product data can be analyzed to determine the product's exposure information, including all user IDs who viewed the product, user IDs who clicked on it, user IDs who added it to their favorites, and user IDs who placed orders. Furthermore, the preprocessed user data can be analyzed to determine the user information for each user, including user ID and basic information. The user data covers all users who viewed the product.

[0034] The user ID serves as the link between these three types of information, thereby establishing a connection between order information, exposure information, and user information, which forms the basis for aggregated analysis.

[0035] Once all order information, exposure information, and user information are identified, correlation and aggregation analysis can be performed on all order information, exposure information, and user information to construct differentiated stage samples.

[0036] For example, using the user ID as the core, order information, exposure information, and user information are aggregated to generate user behavior data corresponding to each user ID. That is, using the user ID as the core, a connection is established between order information, exposure information, and user information. Of course, for users who have not placed an order, there is no order information.

[0037] After aggregating all order information, exposure information, and user information, users can be divided into several user categories based on their basic information, and user IDs can be assigned to the corresponding user categories. Basic information includes age and gender.

[0038] For example, users can be categorized by age as follows: under 18, 18 (inclusive) - 25 (inclusive), 25 (exclusive) - 30 (inclusive), 30 (exclusive) - 40 (inclusive), and over 40. They can also be categorized by gender as male, female, and unknown, resulting in 15 user categories. Among these, the under 18 category is more complex (because the user information obtained is limited, and a large portion of users in this age group on the platform are not their actual age).

[0039] Once the user categories are determined, feature extraction can be performed on the user behavior data in each user category to construct differentiated stage samples.

[0040] For each user category, the system can collect statistics on browsing users, click-through users, favorite users, order-taking users, order amounts, discount levels, product quantities, and product pricing (forming structured tabular data). For example, the statistical data for a specific user category is shown in Table 1 below:

[0041] Table 1. Examples of statistics in User Category 4

[0042]

[0043] Therefore, for each user behavior data under the current user category, differentiated stage samples can be constructed based on the user behavior data. The differentiated stage samples include browsing-click stage samples, click-order stage samples, click-favorite stage samples, and favorite-order stage samples.

[0044] For example, browsing-click phase samples can be generated based on all user behavior data under the current user category. The browsing-click phase samples include valid click samples and invalid click samples. Valid click samples indicate that the user clicked after browsing, while invalid click samples indicate that the user did not click after browsing.

[0045] That is, assuming that in user category 5, there are 10,000 visitors, of whom 1,500 click, 100 favorite, 450 place an order after clicking, and 50 place an order after favorite. Then, in the visit-click phase, there are a total of 10,000 samples, of which 1,500 are valid clicks and 8,500 are invalid clicks.

[0046] For example, click-to-order phase samples can be generated based on all user behavior data under the current user category. The click-to-order phase samples include valid order samples and invalid order samples. Valid order samples indicate that the user made an order after clicking, while invalid order samples indicate that the user did not make an order after clicking.

[0047] Taking user category 5 above as an example, in the click-to-order stage, there are a total of 1,500 samples, of which 450 are valid orders and 1,000 are invalid orders (50 samples that were placed after being added to favorites need to be excluded).

[0048] For example, based on all user behavior data under the current user category, a click-to-favorite phase sample can be generated. The click-to-favorite phase sample includes valid favorite samples and invalid favorite samples. Valid favorite samples indicate that the user has a favorite behavior after clicking, while invalid favorite samples indicate that the user has no favorite behavior after clicking.

[0049] Taking user category 5 as an example again, in the click-to-favorite stage, there are a total of 1500 samples, of which 100 are valid favorites and 1400 are invalid favorites.

[0050] In addition, based on all user behavior data under the current user category, a collection-order stage sample is generated. The collection-order stage sample includes valid direct purchase samples and invalid direct purchase samples. Valid direct purchase samples indicate that the user has made an order after collecting the data, while invalid direct purchase samples indicate that the user has not made an order after collecting the data.

[0051] Taking user category 5 as an example again, in the collection-order stage, there are a total of 100 samples in the collection-order stage, of which 50 are valid direct purchase samples and 50 are invalid direct purchase samples.

[0052] Based on this, differentiated stage samples can be constructed for each user category, resulting in browsing-click stage samples, click-order stage samples, click-favorite stage samples, and favorite-order stage samples for the corresponding user category.

[0053] After completing the sample construction for the differentiation phase, step S40 can be run.

[0054] Step S40: Conduct a differentiation analysis on the samples in the differentiation stage to determine product launch and promotion recommendations.

[0055] In this embodiment, for the differential analysis of the browsing-click stage samples under each user category, multiple product pricing ranges can be divided according to product pricing, such as 280 (inclusive) - 290 (exclusive), 290 (inclusive) - 300 (exclusive), and 300 (inclusive) - 310 (exclusive). This embodiment uses three product pricing ranges as an example for ease of explanation; the actual pricing may be more complex (product ranges are divided according to the highest and lowest prices; the interval between each pricing range is calculated based on the difference between the highest and lowest prices. For differences not exceeding 200, a 10 yuan interval is used; for differences higher than 200 but not higher than 1000, a 50 yuan interval is used; for differences exceeding 1000, a 100 yuan interval is used; and in other cases, the difference divided by 10 is used as the interval). Of course, there are also cases where platforms require uniform pricing. This situation is not within the scope of protection of this embodiment for the time being. In the differentiation analysis, this embodiment is aimed at the situation where multiple different product prices can be implemented during the product promotion stage, and will also involve suggestions on product pricing. Therefore, this situation of uniform pricing will not be elaborated.

[0056] After dividing the product pricing ranges, the number of valid click samples and invalid click samples within each product pricing range are counted. Then, based on the number of valid click samples and invalid click samples within each product pricing range, an observation frequency table and an expected frequency table are generated. Taking a small portion of the data given in user category 44 as an example, examples of generating the observation frequency table and expected frequency table are shown in Tables 2 and 3 below:

[0057] Table 2. Observation Frequency Table

[0058]

[0059] Table 3. Expected Frequency Table

[0060]

[0061] Statistics and calculations for observation frequency tables and expected frequency tables are common techniques, and there are readily available tools that can be used by directly calling interfaces. For example, in Python, the chi2_contingency function in scipy.stats can be used directly to generate observation frequency tables and expected frequency tables by using the statistical data as a basis and calling this interface function.

[0062] Next, the chi-square test calculator can be used to perform chi-square tests on the observed frequency table and the expected frequency table to obtain the chi-square test analysis results. Based on the chi-square test analysis results for each user category, the target user categories are then determined. This process primarily filters out user categories with p-values ​​less than the significance level (e.g., p < 0.05). The user behavior (e.g., click-through rate) in these categories is significantly affected by changes in the influencing factor (product pricing), and therefore they are identified as target user categories worthy of in-depth analysis. Conversely, categories with p-values ​​greater than 0.05 are considered insensitive to the influencing factor (product pricing) and are not considered for further optimization.

[0063] After completing the chi-square test analysis, several target user categories are identified. Logistic regression analysis can then be performed on these target user categories to determine the corresponding pricing-click function. Based on this pricing-click function, product pricing recommendations can be generated for each target user category.

[0064] Specifically, for each target user category: raw granular data is extracted from the corresponding browse-click phase samples. Each data record contains two core fields: independent variable (X): product pricing (e.g., 299.00 yuan); dependent variable (Y): click behavior (a binary variable, 1 represents clicking after browsing, 0 represents not clicking after browsing). A logistic regression model is used to fit the above data. The functional form of the model is:

[0065] , (1)

[0066] in, This represents the probability of a click event occurring at a given price. The logistic regression model automatically calculates the optimal regression coefficients using the maximum likelihood estimation method. (Intercept) and (Slope). After the model is trained, a pricing-click function specific to this target user category is obtained:

[0067] , (2)

[0068] in, Indicates the target user category The pricing-click function is used to quantify the predicted click probability corresponding to any pricing point (the more data, the better the prediction effect). The intercept (represents the baseline click probability when the price is zero; in reality, the price will not be zero, therefore...) It is a mathematical benchmark). The slope The direction and size reveal the relationship between pricing and click-through rate: if A negative value indicates that the click-through rate decreases as the price increases. The larger the value, the greater the impact. If the slope... A positive value indicates that the click-through rate increases as the price increases. The larger the value, the greater the impact.

[0069] Based on this function, corresponding suggestions can be generated according to the goals that the company needs to achieve in the actual promotion stage (such as profit or sales volume). For example, it can be suggested that the product price be reduced to XX, which can increase the click-through rate by X%.

[0070] The above is an introduction to the differential analysis and product placement and promotion suggestion generation plan for the browsing-click stage.

[0071] Similarly, for the click-to-order stage: for each user category, multiple discount ranges can be defined based on the discount level (divided according to set values: 5% and below, 5%-10%, 10%-15%, 15%-20%, 20%-30%, and above 30%, with the starting value of each range excluding the value itself and the ending value including the value). Then, the number of valid and invalid orders within each discount range is determined. Based on these numbers, an observed frequency table and an expected frequency table are generated. Next, a chi-square test calculator is used to perform chi-square tests on the observed and expected frequency tables, yielding the chi-square test analysis results. Based on the chi-square test analysis results for each user category, several target user categories are determined. For a detailed explanation of the process, please refer to the differential analysis process for the browse-click stage; it will not be elaborated here.

[0072] For each target user category, logistic regression analysis can determine the corresponding discount-order function. Based on this function, a discount recommendation for the target user category can be generated. For details, please refer to the logistic regression modeling process for the browse-click phase. The generated discount-order function is as follows:

[0073] , (3)

[0074] in, Indicates the target user category The discount-to-order function is used to quantify the probability that a user will convert a click into an order given a certain discount level. The intercept (representing the baseline order probability when the discount is zero); The slope The direction and magnitude reveal the relationship between discount strength and order rate: if A positive value indicates that the order rate increases as the discount increases. The larger the value, the greater the impact. If the slope... A negative value indicates that the order rate decreases as the discount increases. The larger the value, the greater the impact.

[0075] Based on this function, corresponding suggestions can be generated according to the goals that the company needs to achieve in the actual promotion stage (such as profit or sales volume). For example, it can be suggested to increase the discount to XX, which can increase the order rate by X%.

[0076] The above is an introduction to the differentiated analysis and product placement and promotion suggestion generation plan for the click-to-order stage.

[0077] Similarly, for the click-to-favorite stage: for each user category, multiple product pricing ranges are divided according to product pricing (e.g., 280 (inclusive) - 290 (exclusive), 290 (inclusive) - 300 (exclusive), 300 (inclusive) - 310 (exclusive). These ranges can be divided based on the highest and lowest prices. The interval between each pricing range is calculated based on the difference between the highest and lowest prices: for a difference not exceeding 200, a 10 yuan interval is used; for a difference between 200 and 1000, a 50 yuan interval is used; for a difference exceeding 1000, a 100 yuan interval is used; and in other cases, the difference divided by 10 is used as the interval). Then, the number of valid and invalid favorites within each product pricing range is determined. Based on these numbers, observation frequency tables and expected frequency tables are generated. Next, the chi-square test calculator is used to perform chi-square tests on the observed frequency table and the expected frequency table to obtain the chi-square test analysis results. Based on the chi-square test analysis results for each user category, the target user category is determined. For details, please refer to the differential analysis process for the browsing-click stage, which will not be elaborated here.

[0078] For each target user category, logistic regression analysis can determine the corresponding pricing-favorite function. Based on this function, product pricing suggestions for the target user category can be generated. For details, please refer to the logistic regression modeling process for the browse-click phase. The generated pricing-favorite function is as follows:

[0079] , (4)

[0080] in, Indicates the target user category The pricing-favorites function is used to quantify the probability that a user will convert a click into a favorite given a product price. The intercept represents the baseline probability of collecting when the price is zero; in reality, the price will not be zero, therefore... It is a mathematical benchmark). The slope The direction and size reveal the relationship between pricing and collectability: if A negative value indicates that the collection rate decreases as the price increases. The larger the value, the greater the impact. If the slope... A positive value indicates that the collection rate increases as the price rises. The larger the value, the greater the impact.

[0081] Based on this function, corresponding suggestions can be generated according to the goals that the company needs to achieve in the actual promotion stage (such as profit or sales volume). For example, it can be suggested that the product price be reduced to XX, which can increase the collection rate by X%.

[0082] The above is an introduction to the differentiated analysis of the click-to-favorite stage and the generation of product placement and promotion suggestions.

[0083] Similarly, for the collection-order stage: for each user category, multiple discount ranges can be defined based on the discount level (divided according to set values: 5% and below, 5%-10%, 10%-15%, 15%-20%, 20%-30%, and above 30%, with the starting value of each range excluding the value itself and the ending value including the value). Then, the number of valid and invalid direct purchase samples within each discount range is determined. Based on these numbers, an observed frequency table and an expected frequency table are generated. Next, a chi-square test calculator is used to perform chi-square tests on the observed and expected frequency tables, yielding the chi-square test analysis results. Based on the chi-square test analysis results for each user category, several target user categories are determined. For a detailed explanation of the process, please refer to the differential analysis process for the collection-order stage; it will not be elaborated here.

[0084] For each target user category, logistic regression analysis can determine the corresponding discount-to-purchase function. Based on this function, a discount recommendation for the target user category can be generated. For details, please refer to the logistic regression modeling process for the browsing-click phase. The generated discount-to-purchase function is as follows:

[0085] , (5)

[0086] in, Indicates the target user category The discount-to-purchase function is used to quantify the probability that a user will convert their collection into a purchase under a given discount level. The intercept (represents the baseline probability of direct purchase when the discount is zero); The slope The direction and magnitude reveal the relationship between the discount strength and the direct purchase rate (i.e., the conversion rate from adding to favorites to ordering): if A positive value indicates that the direct purchase rate increases with the discount level. The larger the value, the greater the impact. If the slope... A negative value indicates that the direct purchase rate decreases as the discount increases. The larger the value, the greater the impact.

[0087] Based on this function, corresponding suggestions can be generated according to the goals that the company needs to achieve in the actual promotion stage (such as profit or sales volume). For example, it can be suggested to increase the discount to XX, which can increase the direct purchase rate by X%.

[0088] The above is an introduction to the differentiated analysis and product placement and promotion suggestions generated during the collection-order stage.

[0089] After completing the differentiated analysis of each target user category at each stage (browse-click stage, click-order stage, click-favorite stage, favorite-order stage), a comprehensive analysis and recommendations can be made based on the situation of each target user category.

[0090] First, the product cost can be obtained; this cost is the comprehensive cost per unit of product set by the company. Then, the product cost, pricing-click function, discount-order function, pricing-favorite function, and discount-direct purchase function can be input into the expected profit model. The expected profit model includes an expected profit function:

[0091] Expected profit = a × click-through rate × order rate × (product price × (1 - discount) - product cost) + b × click-through rate × collection rate × direct purchase rate × (product price × (1 - discount) - product cost).

[0092] Where a and b are the set weight values ​​(the set values ​​can be fixed values ​​or based on the ratio of the number of users from click to order to the number of users from favorite to order).

[0093] Based on this, the expected profit function can be solved using a nonlinear programming algorithm to obtain the optimal solution that maximizes the expected profit, thereby generating a promotion strategy plan that includes product pricing, discount level, and expected profit value.

[0094] The expected profit function is defined as:

[0095] , (6)

[0096] in, It is the product pricing (decision variable). It is the discount level (decision variable, ranging from 0 to 1). It is the product cost (a known constant). and It is a weight value. For pricing - click function, For discount-order function, For pricing-favorites function, This is a discount-direct purchase function.

[0097] Constraints:

[0098] (1) (The lower limit of product pricing can be cost C or higher);

[0099] (2) (The upper limit for product pricing is set based on market affordability, with the highest product price in the company's data as the upper limit.)

[0100] (3) (The maximum discount level can be set to 1, but it is usually set to 0.5).

[0101] The goal is to find under constraints and The value of makes maximize.

[0102] Due to the expected profit function Since the objective function is non-linear and involves exponential functions, this embodiment employs the gradient ascent method to quickly converge to a local maximum using the gradient information of the objective function. Based on this, the expected profit function can be solved, yielding the optimal solution for maximizing expected profit (including the optimal solution at time...). , , This generates a promotional strategy that includes product pricing, discount levels, and expected profit.

[0103] In summary, this application provides a suggestion generation method based on enterprise data differentiation analysis. It obtains product sales data from the same online trading platform, including order data, product data, and user data. The method preprocesses the order data, product data, and user data; performs correlation and aggregation analysis on the preprocessed order data, product data, and user data to construct differentiation stage samples; and conducts differentiation analysis on the differentiation stage samples to determine product placement and promotion suggestions. Under the condition of limited data dimensions (inability to obtain detailed user profiles), this solution takes a different approach. By constructing a data processing scheme, it avoids the technical route of traditional e-commerce analysis that relies on the platform's user tag system (such as purchasing power level and interest preferences). Under data constraints, it deeply aggregates three types of data—orders, exposures, and basic user information—to construct an analysis method centered on user behavior. It breaks down the complete user journey on the platform into four differentiation stages: browsing-click, click-order, click-favorite, and favorite-order, and constructs a sample set containing positive and negative samples (valid and invalid) for each stage. By employing a refined sample construction approach, traditional macro-level conversion rate metrics are broken down into multiple micro-level conversion stages that can be independently analyzed and optimized, providing a data foundation for attribution analysis and strategy generation. In the specific differentiation stage analysis, a two-stage analysis approach is used: chi-square test screening and logistic regression. For each user category and behavioral stage, the solution uses a chi-square test to quickly scan discrete pricing or discount ranges. This rapidly identifies target user categories with significant differences from combinations of user category, behavioral stage, and influencing factors, completing the initial screening and avoiding the waste of resources from directly modeling all combinations. For the identified valuable targets, logistic regression is used to fit product pricing or discount intensity as a continuous variable, outputting continuous functions such as pricing-click function, discount-order function, pricing-favorite function, and discount-direct purchase function. These functions can serve as the basis for generating suggestions for differentiation stage analysis and also as the basis for subsequent comprehensive expected profit analysis.

[0104] This solution, building upon the analysis of differentiated stages, introduces an expected profit model based on maximizing expected profit. This shifts the focus of analysis from maximizing conversion rate to maximizing profit. After obtaining the pricing-click function, discount-order function, pricing-favorite function, and discount-direct purchase function, these are integrated with product costs to construct a unified expected profit function. Based on actual conditions, two conversion paths are cleverly designed: immediate conversion through direct click and order (weighted by parameter a), and delayed conversion through ordering after adding to favorites (weighted by parameter b). A nonlinear programming algorithm is used to solve the expected profit function, automatically calculating the specific product pricing and discount values ​​that maximize expected profit. This generates a directly executable, quantifiable, and optimally profitable promotion strategy, providing businesses with a more advantageous strategy for product promotion.

[0105] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A suggestion generation method based on enterprise data differentiation analysis, characterized in that, include: Obtain product sales data from the same online trading platform for the enterprise. The product sales data includes order data, product data, and user data. Preprocess order data, product data, and user data; Perform correlation and aggregation analysis on preprocessed order data, product data, and user data to construct differentiated stage samples; Conduct differentiation analysis on samples from the differentiation phase to determine product launch and promotion recommendations; Perform correlation and aggregation analysis on preprocessed order data, product data, and user data to construct differentiated stage samples, including: The preprocessed order data is analyzed to determine the order information for each order, including order amount, discount level, product quantity, product price, and order user ID. The preprocessed product data is analyzed to determine the product's exposure information, which includes all user IDs of users who viewed the product, clicked on it, added it to their favorites, and ordered it. The preprocessed user data is analyzed to determine the user information for each user. The user information includes user ID and basic information. The user data covers all users who browse the product. Perform correlation and aggregation analysis on all order information, exposure information, and user information to construct differentiated stage samples; A correlation and aggregation analysis is performed on all order information, exposure information, and user information to construct differentiated stage samples, including: Using user ID as the core, order information, exposure information and user information are aggregated to generate user behavior data corresponding to each user ID; Based on users' basic information, users are divided into several user categories, and user IDs are assigned to the corresponding user categories. The basic information includes age and gender. Feature extraction is performed on user behavior data in each user category to construct differentiated stage samples; Feature extraction is performed on user behavior data for each user category to construct differentiated stage samples, including: For each user category: Statistics on browsing users, click users, favorite users, order users, order amount, discount level, product quantity, and product price under the current user category; For each user behavior data point within the current user category: Based on user behavior data, differentiated stage samples are constructed, including browsing-click stage samples, click-order stage samples, click-favorite stage samples, and favorite-order stage samples. If the differentiation stage sample is a browse-click stage sample, perform differentiation analysis on the differentiation stage sample to determine product placement and promotion recommendations, including: For each user category: Divide products into multiple pricing ranges based on their pricing structure; Determine the number of valid click samples and invalid click samples within each product's pricing range; Based on the number of valid click samples and invalid click samples within each product's pricing range, generate an observation frequency table and an expected frequency table; Use the chi-square test calculator to perform chi-square test calculations on the observed frequency table and the expected frequency table, and obtain the chi-square test analysis results; Based on the chi-square test results of each user category, the target user category is determined; Logistic regression analysis was performed on the target user categories to determine the pricing-click function; Based on the pricing-click function, generate product pricing suggestions for the target user category.

2. The suggestion generation method based on enterprise data differentiation analysis according to claim 1, characterized in that, Based on user behavior data, differentiated stage samples are constructed, including: Based on all user behavior data under the current user category, generate browsing-click stage samples. The browsing-click stage samples include valid click samples and invalid click samples. Valid click samples indicate that the user clicked after browsing, while invalid click samples indicate that the user did not click after browsing. Based on all user behavior data under the current user category, generate click-to-order phase samples. The click-to-order phase samples include valid order samples and invalid order samples. Valid order samples indicate that the user made an order after clicking, while invalid order samples indicate that the user did not make an order after clicking. Based on all user behavior data under the current user category, generate click-to-favorite stage samples. The click-to-favorite stage samples include valid favorite samples and invalid favorite samples. Valid favorite samples indicate that the user has favorited the item after clicking, while invalid favorite samples indicate that the user has not favorited the item after clicking. Based on all user behavior data under the current user category, a collection-order stage sample is generated. The collection-order stage sample includes valid direct purchase samples and invalid direct purchase samples. Valid direct purchase samples indicate that the user made an order after collecting the data, while invalid direct purchase samples indicate that the user did not make an order after collecting the data.

3. The suggestion generation method based on enterprise data differentiation analysis according to claim 1, characterized in that, If the differentiation stage sample is a click-to-order stage sample, perform differentiation analysis on the differentiation stage sample to determine product placement and promotion recommendations, including: For each user category: Divide the discount levels into multiple ranges; Determine the number of valid and invalid order samples within each discount range; Based on the number of valid and invalid order samples within each discount range, generate an observation frequency table and an expected frequency table. Use the chi-square test calculator to perform chi-square test calculations on the observed frequency table and the expected frequency table, and obtain the chi-square test analysis results; Based on the chi-square test results of each user category, the target user category is determined; Logistic regression analysis was performed on the target user categories to determine the discount-order function; Based on the discount-order function, generate discount suggestions for the target user category.

4. The suggestion generation method based on enterprise data differentiation analysis according to claim 3, characterized in that, If the sample in the differentiation stage is the click-to-favorite stage, perform differentiation analysis on the sample in the differentiation stage to determine product placement and promotion suggestions, including: For each user category: Divide products into multiple pricing ranges based on their pricing structure; Determine the number of valid and invalid collection samples within each product's pricing range; Based on the number of valid and invalid collection samples within each product's pricing range, generate an observation frequency table and an expected frequency table. Use the chi-square test calculator to perform chi-square test calculations on the observed frequency table and the expected frequency table, and obtain the chi-square test analysis results; Based on the chi-square test results of each user category, the target user category is determined; Logistic regression analysis was performed on the target user categories to determine the pricing-collection function; Based on the pricing-favorites function, product pricing suggestions are generated for the target user category.

5. The suggestion generation method based on enterprise data differentiation analysis according to claim 4, characterized in that, If the sample in the differentiation stage is the collection-order stage, perform differentiation analysis on the sample in the differentiation stage to determine product placement and promotion suggestions, including: For each user category: Divide the discount levels into multiple ranges; Determine the number of valid direct purchase samples and the number of invalid direct purchase samples within each discount range; Based on the number of valid direct purchase samples and the number of invalid direct purchase samples within each discount range, generate an observation frequency table and an expected frequency table; Use the chi-square test calculator to perform chi-square test calculations on the observed frequency table and the expected frequency table, and obtain the chi-square test analysis results; Based on the chi-square test results of each user category, the target user category is determined; Logistic regression analysis was performed on the target user categories to determine the discount-direct purchase function; Based on the discount-direct purchase function, generate discount suggestions for the target user category.

6. The suggestion generation method based on enterprise data differentiation analysis according to claim 5, characterized in that, After determining the pricing-click function, discount-order function, pricing-favorites function, and discount-direct purchase function for the same target user category, the method further includes: Obtain product cost; Input the product cost, pricing-click function, discount-order function, pricing-favorite function, and discount-direct purchase function into the expected profit model. The expected profit model includes the expected profit function: Expected profit = a × click-through rate × order rate × (product price × (1 - discount) - product cost) + b × click-through rate × collection rate × direct purchase rate × (product price × (1 - discount) - product cost). Where a and b are the set weight values; The expected profit function is solved using a nonlinear programming algorithm to obtain the optimal solution that maximizes the expected profit, and a promotion strategy plan is generated that includes product pricing, discount level, and expected profit value.

Images