Intelligent promotion comprehensive processing system based on e-commerce information

Abstract

The invention discloses an intelligent promotion comprehensive processing system based on e-commerce information, and relates to the technical field of intelligent promotion, and the system screens user multi-source data, and constructs a personalized user portrait according to the screened user multi-source data; the personalized user portraits and the market multi-source data are analyzed, and the user recommendation condition and the market recommendation condition of each product are identified; for each product, obtaining a promotion score according to the user recommendation condition and the market recommendation condition, and further determining a promoted product for the user according to the promotion score; the promotion process of the promoted product is monitored in real time, and the promotion effect of the promoted product is predicted according to the real-time monitoring of the promotion process; screening the promoted products according to the promotion effect of the promoted products, updating the promotion score of each product, and adjusting and optimizing the promotion of the products. Personalized recommendation and dynamic optimization are realized, so that the popularization efficiency and the user satisfaction are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent promotion, and in particular to an intelligent promotion comprehensive processing system based on e-commerce information. Background Art

[0002] With the continuous advancement and popularization of Internet technology, the e-commerce market has shown explosive growth. More and more consumers choose to shop online. Enterprises are facing increasingly fierce market competition. In order to stand out from the competition, enterprises need more effective promotion methods to attract users and increase sales. At the same time, consumer needs are becoming more and more diversified and personalized. They hope to get product recommendations that meet their interests and needs. The traditional one-size-fits-all promotion method can no longer meet the needs of users. Enterprises need to adopt more precise promotion strategies to improve user satisfaction and loyalty.

[0003] Existing technologies often adopt extensive promotion methods and lack in-depth understanding of individual user characteristics and needs. For example, traditional email marketing may send the same promotional information to all users without considering their interests and purchase history, which results in a large amount of promotional information being ignored by users and poor promotion effects. Some existing promotion systems only focus on user data and ignore the dynamic changes in the market. Market trends, competitor strategies and other factors will have an impact on the promotion effect of the product. In addition, during the promotion process, existing technologies make it difficult to monitor the effects in real time and adjust strategies in a timely manner. Once the promotion activity begins, it is difficult to dynamically optimize according to actual conditions.

[0004] Therefore, in response to the above problems, there is an urgent need for an intelligent promotion comprehensive processing system based on e-commerce information. Summary of the invention

[0005] In view of the deficiencies in the prior art, the present invention provides an intelligent promotion comprehensive processing system based on e-commerce information, which solves the problems of inaccurate user portraits and inaccurate promotion strategies in e-commerce promotion.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: an intelligent promotion comprehensive processing system based on e-commerce information, including: a data acquisition module, used to acquire user multi-source data and market multi-source data; a user portrait construction module, used to screen user multi-source data, and then build a personalized user portrait based on the screened user multi-source data; a recommendation status identification module, used to analyze the personalized user portrait and market multi-source data respectively, and identify the user recommendation status and market recommendation status of each product; a promotion product screening module, used to obtain a promotion score for each product based on the user recommendation status and market recommendation status, and then determine the promotion product for the user based on the promotion score; an effect prediction module, used to monitor the promotion process of the promotion product in real time, and then predict the promotion effect of the promotion product based on the real-time monitoring of the promotion process; a promotion optimization and update module, used to screen the promotion products based on the promotion effect of the promotion products, and then update the promotion score of each product, and adjust and optimize the promotion of the product.

[0007] Furthermore, the user multi-source data is screened, and then the specific analysis of constructing a personalized user portrait based on the screened user multi-source data is as follows: based on user authorization, user multi-source data is collected, and the multiple sources include but are not limited to network platforms, social media, and e-commerce platforms. The user multi-source data is but not limited to user gender, user age, user frequently purchased product categories, user consumption amount, user purchase record, user collection record, and user browsing record; the user multi-source data is normalized, and missing values ​​and outliers are processed, and then the Pierre coefficient is used to calculate the correlation between the user multi-source data and the user shopping behavior; the correlation is compared with the correlation threshold, and the user multi-source data with a correlation lower than the correlation threshold is screened out; the user multi-source data with a correlation greater than or equal to the correlation threshold is integrated, and normalized and marked as a personalized user portrait feature set, and a personalized user portrait is constructed based on the personalized user portrait feature set.

[0008] Furthermore, the personalized user portrait and the market multi-source data are analyzed respectively to identify the user recommendation status and the market recommendation status of each product. The specific analysis is as follows: extracting the personalized user portrait feature set and the market multi-source data, the market multi-source data including but not limited to product category, product inventory, product sales, product praise rate, product positive demand trend, product market growth rate; taking the personalized user portrait and the market multi-source data as input respectively, using the Logistic regression analysis method to construct a user product interest evaluation model and a product market promotion potential evaluation model respectively, and outputting the user personalized recommendation coefficient and the market recommendation coefficient.

[0009] Furthermore, in the above-mentioned construction of the user product interest evaluation model and the product marketing potential evaluation model respectively by using the Logistic regression analysis method, the specific expression of the user product interest evaluation model is: , where represents the user personalized recommendation coefficient, Indicates that the user is interested in a product and purchases it. Represents a personalized user portrait feature set, , Represents the total number of personalized user portrait features, represents the coefficient of the user product interest evaluation model; the specific expression of the product marketing potential evaluation model is: , where represents the market recommendation coefficient, It means that a product has promotion potential in the market. Represents a collection of multi-source market data. , Indicates the total number of multi-source market data, Represents the coefficient of the product marketing potential evaluation model.

[0010] Furthermore, a promotion score is obtained based on the user recommendation status and the market recommendation status, and then a specific analysis of the promotion products for the user is determined based on the promotion score: for each product, the user personalized recommendation coefficient and the market recommendation coefficient are extracted, and the user interest level and the market recommendation level of each product are divided according to the user personalized recommendation coefficient and the market recommendation coefficient, the user interest level includes a first interest level, a second interest level and a third interest level, and the market recommendation level includes a first recommendation level, a second recommendation level and a third recommendation level; the first interest level and the first recommendation level score are recorded as 2, the second interest level and the second recommendation level score are recorded as 1, and the third interest level and the third recommendation level score are recorded as 0; the sum of the user interest level and the market recommendation level score of each product is taken as the promotion score of each product; products with a promotion score of not less than 3 are marked as the user's promotion products, and then the user's promotion products are recommended and displayed using the e-commerce visualization platform interface.

[0011] Furthermore, a specific analysis of user interest level and market recommendation level of each product divided according to user personalized recommendation coefficient and market recommendation coefficient is as follows: for the division of user interest level, the 20th percentile of the sample size is used, and the user's interest in the product is divided into three levels according to the size of the user personalized recommendation coefficient: when the user personalized recommendation coefficient is below 0.5997, it is the third interest level, when the user personalized recommendation coefficient is between 0.5997 and 0.7996, it is the second interest level, and when the user personalized recommendation coefficient is higher than 0.7996, it is the first interest level; for the division of market recommendation level, the 20th percentile of the sample size is used, and the product's market promotion potential is divided into three levels according to the size of the market recommendation coefficient: when the market recommendation coefficient is below 0.5997, it is the third recommendation level, when the market recommendation coefficient is between 0.5997 and 0.7996, it is the second recommendation level, and when the market recommendation coefficient is higher than 0.7996, it is the first recommendation level.

[0012] Furthermore, the specific analysis of the promotion effect of the promotion product predicted based on the real-time monitoring of the promotion process is: using the promotion product return rate as the evaluation basis for the promotion effect;

[0013] Obtain monitoring parameters of the promoted product, including the number of interactions and exposures; construct a time series model of the return rate of the promoted product, taking the number of interactions and exposures as exogenous variables. The specific expression of the time series model of the return rate of the promoted product is:

[0014] ; In the formula Indicates the time point number. Indicates time point The predicted value of the return rate of the promoted product, represents a lag term of the predicted value of the promotion product return rate, It represents a difference term of the predicted value of the promotion product return rate. Indicates time point The number of interactions, represents a lagged term of the number of interactions, Indicates time point The exposure represents a lag term of exposure, , The coefficient of the autoregressive part indicates the autocorrelation of the predicted value of the promotion product return rate. The coefficient of the difference part indicates the stability of the predicted value of the promotion product return rate. , , , The coefficients of the exogenous variables represent the impact of the number of interactions and exposure on the predicted value of power generation, represents the error term, which represents unexplained random fluctuations.

[0015] Furthermore, the specific analysis for screening the promotional products based on the promotional effects of the promotional products is as follows: obtaining a return rate threshold, comparing the promotional product return rate with the return rate threshold, and when the promotional product return rate is lower than the return rate threshold, judging that the promotional effect of the promotional product is poor, and reducing the recommendation display of the promotional product on the e-commerce visualization platform interface; when the promotional product return rate is greater than the return rate threshold, judging that the promotional effect of the promotional product is good, and increasing the recommendation display of the promotional product on the e-commerce visualization platform interface.

[0016] The present invention has the following beneficial effects:

[0017] The intelligent promotion comprehensive processing system based on e-commerce information can accurately analyze multi-source data of users and build personalized user portraits, so that the promotion can be closer to the actual needs and preferences of users, and improve the pertinence and effectiveness of the promotion; the recommendation status identification module analyzes personalized user portraits and multi-source market data, and can accurately identify the user recommendation status and market recommendation status of each product, which helps the system to screen out the products that are most suitable for current users and market trends for promotion, and improve the accuracy of recommendation and user satisfaction; the promotion product screening module and the promotion optimization and update module work together to realize the dynamic screening and optimization of promotion products, and can monitor the promotion effect in real time, and adjust the promotion strategy according to the effect, so as to ensure the continuous optimization and improvement of promotion activities; the automated data processing and analysis process, as well as the intelligent promotion strategy formulation, greatly improve the efficiency of promotion work, can quickly respond to market changes, and timely adjust the promotion strategy to adapt to the changing user needs and market environment; through accurate recommendation and dynamic optimization, it can reduce unnecessary promotion waste, concentrate resources on the most potential products, help reduce promotion costs, and improve the overall return on investment, personalized promotion content and accurate product recommendations can enhance the user's shopping experience, thereby improving satisfaction and loyalty to the e-commerce platform.

[0018] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is a structural diagram of an intelligent promotion comprehensive processing system based on e-commerce information of the present invention.

[0020] Figure 2 The present invention is a method flow chart of an intelligent promotion comprehensive processing system based on e-commerce information. DETAILED DESCRIPTION

[0021] The embodiment of the present application realizes personalized recommendation and dynamic optimization through an intelligent promotion comprehensive processing system based on e-commerce information, thereby improving promotion efficiency and user satisfaction.

[0022] The overall idea of ​​the embodiments of the present application is: to collect multi-source user data and multi-source market data to provide a basis for subsequent analysis and decision-making; to screen and analyze multi-source user data, build personalized user portraits, and deeply understand the characteristics and needs of each user; to analyze personalized user portraits and multi-source market data, and then comprehensively calculate promotion scores, and select the most suitable promotion products for specific users from a large number of products; to monitor the promotion process of promotion products in real time, predict the promotion effect based on the monitoring data, and provide a reference for the adjustment of promotion strategies; to screen products according to the actual promotion effects of promotion products, update promotion scores, adjust promotion strategies, and achieve continuous optimization of promotion effects.

[0023] See also Figure 1 , Figure 2 The embodiment of the present invention provides a technical solution: an intelligent promotion comprehensive processing system based on e-commerce information, including: a data acquisition module, used to acquire user multi-source data and market multi-source data; a user portrait construction module, used to screen the user multi-source data, and then build a personalized user portrait based on the screened user multi-source data; a recommendation status identification module, used to analyze the personalized user portrait and the market multi-source data respectively, and identify the user recommendation status and market recommendation status of each product; a promotion product screening module, used to obtain a promotion score for each product based on the user recommendation status and the market recommendation status, and then determine the promotion product for the user based on the promotion score; an effect prediction module, used to monitor the promotion process of the promotion product in real time, and then predict the promotion effect of the promotion product based on the real-time monitoring of the promotion process; a promotion optimization and update module, used to screen the promotion products based on the promotion effect of the promotion products, and then update the promotion score of each product, and adjust and optimize the promotion of the product.

[0024] Specifically, the user multi-source data is screened, and then the specific analysis of building a personalized user portrait based on the screened user multi-source data is as follows: based on user authorization, user multi-source data is collected, and the multi-sources include but are not limited to network platforms, social media, and e-commerce platforms. The user multi-source data includes but is not limited to user gender, user age, user frequently purchased product categories, user consumption amount, user purchase record, user collection record, user browsing record; the user multi-source data is normalized, and missing values ​​and outliers are processed, and then the Pierre coefficient is used to calculate the correlation between the user multi-source data and the user shopping behavior; the correlation is compared with the correlation threshold, and the user multi-source data with a correlation lower than the correlation threshold is screened out; the user multi-source data with a correlation greater than or equal to the correlation threshold is integrated, and normalized and marked as a personalized user portrait feature set, and a personalized user portrait is built based on the personalized user portrait feature set.

[0025] In this implementation scheme, the specific steps of calculating the correlation using the Pearson coefficient are as follows: extract the feature data S (such as user age, user spending amount, etc.) and user shopping behavior data G (such as purchase frequency, purchase amount, etc.) to be analyzed from the multi-source user data after normalization and processing of missing values ​​and outliers; assume , , where b is the number of samples; calculate the mean and standard deviation of S and G respectively, and then obtain the covariance of S and G based on the mean of S and G; compare the covariance of S and G with the product of the standard deviation of S and G to obtain the correlation between user multi-source data and user shopping behavior.

[0026] The association threshold is obtained in the following way: according to the previous business experience and data analysis results, a suitable association threshold is set. For example, in similar user portrait construction projects, it is found that the features with a Pearson coefficient greater than 0.5 are strongly correlated with the user shopping behavior, so the association threshold can be set to 0.5. A large amount of historical data can also be analyzed to observe the distribution of the Pearson coefficient and select a suitable quantile as the association threshold, such as the 75th quantile, that is, retaining the features with the top 25% of the Pearson coefficient. Alternatively, user portrait construction experiments under different association thresholds can be conducted to evaluate the impact of different thresholds on the accuracy of user portraits and marketing effects, and the association threshold that can achieve the best marketing effect (such as the highest conversion rate, the highest customer satisfaction, etc.) can be selected as the final association threshold. Machine learning models (such as decision trees, random forests, etc.) can also be used to rank the features by importance, and the Pearson coefficients corresponding to the features with higher importance can be used as a reference to determine the association threshold.

[0027] Normalizing user multi-source data can eliminate the impact of dimensions and scales between different features, making the data comparable in subsequent analysis. Processing missing values ​​and outliers can avoid these data from interfering with the analysis results, ensuring the accuracy and reliability of the data, thus laying the foundation for building accurate user portraits. The Pearson coefficient is used to calculate the correlation between user multi-source data and user shopping behavior, and the data with low correlation is screened out by comparing with the correlation threshold. In this way, we can focus on data closely related to user shopping behavior, reduce the interference of irrelevant data, and improve the accuracy and pertinence of user portraits. Accurate user portraits can more accurately reflect users' shopping preferences and behavioral habits, and help companies formulate more effective marketing strategies. Integrating user multi-source data with correlation greater than or equal to the correlation threshold into a personalized user portrait feature set facilitates the subsequent construction and management of user portraits, making the data more structured, facilitating the system to analyze and apply user portraits, and improving the availability and operability of the data. Based on accurate personalized user portraits, companies can provide users with product recommendations and marketing activities that better meet their needs, thereby increasing users' willingness to buy and loyalty, and improving marketing effects and the economic benefits of the company.

[0028] Specifically, personalized user portraits and market multi-source data are analyzed respectively to identify the user recommendation status and market recommendation status of each product. The specific analysis is as follows: extract personalized user portrait feature sets and market multi-source data, and the market multi-source data includes but is not limited to product categories, product inventory, product sales, product praise rate, product positive demand trends, and product market growth rate; use personalized user portraits and market multi-source data as input, and use the Logistic regression analysis method to construct a user product interest evaluation model and a product market promotion potential evaluation model, and output the user personalized recommendation coefficient and the market recommendation coefficient.

[0029] In this implementation scheme, the user product interest evaluation model and the product marketing potential evaluation model are respectively constructed using the Logistic regression analysis method. The specific expression of the user product interest evaluation model is: , where represents the user personalized recommendation coefficient, Indicates that the user is interested in a product and purchases it. Represents a personalized user portrait feature set, , Represents the total number of personalized user portrait features, Represents the coefficient of the user product interest evaluation model.

[0030] The specific expression of the product marketing potential evaluation model is:

[0031] , where represents the market recommendation coefficient, It means that a product has promotion potential in the market. Represents a collection of multi-source market data, , Indicates the total number of multi-source market data, Represents the coefficient of the product marketing potential evaluation model.

[0032] Multi-source market data is obtained from e-commerce platform databases, market research institutions, industry reports and other channels; users’ interest in a product and their purchase is determined by their actual purchasing behavior. If a user purchases a product, the value is 1, otherwise it is 0; whether a product has promotion potential in the market is comprehensively judged based on factors such as product sales growth and market share changes. For example, if a product’s sales volume grows rapidly over a period of time and its market share gradually expands, it is considered that the product has promotion potential and the value is 1, otherwise it is 0.

[0033] The specific steps of building a model using the Logistic regression analysis method are as follows: collect and organize personalized user portrait feature sets and market multi-source data sets as input data for the user product interest evaluation model and the product market promotion potential evaluation model respectively; for the user product interest evaluation model, the target variable is whether the user is interested in and purchases a certain product, and for the product market promotion potential evaluation model, the target variable is whether a certain product has promotion potential in the market; divide the data into training sets and test sets, generally using a ratio of 70%-30% or 80%-20% for division; initialize the Logistic regression model and set the model parameters, such as the regularization parameter etc.; use the training set data to train the model, and estimate the model coefficients through methods such as maximum likelihood estimation. During the training process, the model will continuously adjust the coefficients to minimize the error between the model's predicted results and the actual results; use the test set data to evaluate the trained model. Common evaluation indicators include accuracy, recall rate, F1 value, etc. The performance of the model is judged by the evaluation indicators. If the model performance is not ideal, the model parameters can be adjusted or the features can be reselected; when the model performance meets the requirements, it is applied to actual data, and the personalized user portrait feature set and the market multi-source data set are input to calculate the user personalized recommendation coefficient and the market recommendation coefficient respectively.

[0034] The coefficients of the user product interest evaluation model and the product marketing potential evaluation model are obtained by using the maximum likelihood estimation method to find a set of coefficients so that given these coefficients, the probability of the observed sample data occurring is maximized. The coefficients of the model can be obtained by solving the maximum value of the log-likelihood function. In practical applications, optimization algorithms (such as gradient descent, Newton's method, etc.) can be used to solve the maximum likelihood estimation problem.

[0035] By constructing a user product interest evaluation model, the user personalized recommendation coefficient can be accurately calculated based on personalized user portraits, reflecting the user's interest in different products, which helps e-commerce platforms accurately recommend products that meet their interests to users, improve users' purchase conversion rate and satisfaction, and improve the platform's sales performance; the product market promotion potential evaluation model uses multi-source market data to calculate the market recommendation coefficient to help companies understand the product's promotion potential in the market. Companies can reasonably allocate promotion resources based on the market recommendation coefficient, focus on products with high promotion potential, and improve market promotion efficiency and effectiveness; the model is constructed based on the Logistic regression analysis method, so that the entire recommendation and promotion process is based on data-based decision-making. The data-driven approach reduces the influence of subjective factors, improves the scientificity and accuracy of decision-making, and can better adapt to market changes and user needs; the Logistic regression model has good interpretability, and the model coefficient can reflect the influence of each feature on user interest and product market promotion potential. Companies can adjust marketing strategies based on the model coefficient, such as optimizing features that have a greater impact on user interest, or focusing on factors that have a greater impact on market promotion potential.

[0036] Specifically, a promotion score is obtained based on the user recommendation status and the market recommendation status, and then the specific analysis of the promotion products for the user is determined based on the promotion score: for each product, the user personalized recommendation coefficient and the market recommendation coefficient are extracted, and the user interest level and the market recommendation level of each product are divided according to the user personalized recommendation coefficient and the market recommendation coefficient. The user interest level includes the first interest level, the second interest level and the third interest level, and the market recommendation level includes the first recommendation level, the second recommendation level and the third recommendation level; the first interest level and the first recommendation level score are recorded as 2, the second interest level and the second recommendation level score are recorded as 1, and the third interest level and the third recommendation level score are recorded as 0; the sum of the user interest level and the market recommendation level score of each product is taken as the promotion score of each product; products with a promotion score of not less than 3 are marked as the user's promotion products, and then the user's promotion products are recommended and displayed using the e-commerce visualization platform interface.

[0037] In this implementation scheme, the specific analysis of user interest level and market recommendation level of each product according to the user personalized recommendation coefficient and the market recommendation coefficient is as follows: for the division of user interest level, the 20th percentile point of the sample size is used, and the user's interest in the product is divided into three levels according to the size of the user personalized recommendation coefficient: when the user personalized recommendation coefficient is below 0.5997, it is the third interest level, when the user personalized recommendation coefficient is between 0.5997 and 0.7996, it is the second interest level, and when the user personalized recommendation coefficient is higher than 0.7996, it is the first interest level; for the division of market recommendation level, the 20th percentile point of the sample size is used, and the market promotion potential of the product is divided into three levels according to the size of the market recommendation coefficient: when the market recommendation coefficient is below 0.5997, it is the third recommendation level, when the market recommendation coefficient is between 0.5997 and 0.7996, it is the second recommendation level, and when the market recommendation coefficient is higher than 0.7996, it is the first recommendation level.

[0038] The impact of different percentile points on promotion effect is as follows:

[0039] Regarding the classification based on low percentiles (such as the 10% percentile): more products will be classified into higher levels. For example, when classifying user interest levels or market recommendation levels, a large number of products will be identified as the first or second level based on the lower percentiles. This may lead to a large selection range of promotional products. Enterprises need to invest more resources in promotion, increasing promotion costs. Since the screening criteria are relatively loose, some products with low actual promotion value may be included, which will reduce the accuracy of the promotion and disperse the promotion information. Users may receive a large amount of promotional content that they are not interested in, thereby reducing the user's response rate and conversion rate to the promotion.

[0040] Regarding the classification by high percentiles (such as the 90% percentile): it will greatly narrow the selection range of promoted products. Only a very small number of products with outstanding performance will be classified into a higher level and recommended to users. This may cause the company to miss some potential products and fail to fully tap market opportunities. Although the screened products may have high quality and promotion value, due to their small number, they may not be able to meet the diverse needs of users and reduce their willingness to buy. Too narrow a product range may not be able to cover a wider user group, affecting the company's market share and brand influence.

[0041] The advantages of using the 20th percentile of the sample size are as follows: the 20th percentile division can screen out relatively high-quality products while ensuring that a certain number of products are recommended. It will neither recommend too many products like the low percentiles, resulting in a waste of resources, nor recommend too few products like the high percentiles, missing market opportunities. For example, when dividing user interest levels, about 20% of the products will be classified as the first level. These products are the ones that users are more interested in. Enterprises can concentrate resources on promoting these products and improve promotion efficiency; dividing user interests or market recommendation potential into three levels can classify products more finely. Through the 20th percentile division, the products that users are most interested in or have the greatest market promotion potential can be screened out and accurately recommended to users, which helps to increase users' attention and response rate to promotion content, thereby improving promotion effects; the products and user needs on the market are diverse. The 20th percentile division can take into account different types of products and users to a certain extent. For some niche but potential products, there is also a chance to be classified into a higher level for promotion, which meets the diverse needs of the market and expands the market coverage of the enterprise, thereby ensuring the stability of the promotion effect.

[0042] Combining the user personalized recommendation coefficient and the market recommendation coefficient to determine the promotion score can comprehensively consider two important aspects: user interest and market potential. It avoids only focusing on individual user needs and ignoring market trends, and also prevents only focusing on market prospects and ignoring user actual interests, making the selection of promotion products more scientific and reasonable. By dividing the user interest level and the market recommendation level, the products are classified, which can more accurately screen out promotion products suitable for users. Products with a promotion score of not less than 3 are marked as promotion products for recommendation and display, which improves the pertinence of the recommendation and helps to improve user acceptance and conversion rate of promotion products. User interest and market promotion potential are divided into three levels and assigned corresponding scores, which simplifies the evaluation process. This quantitative method makes the calculation of promotion scores more intuitive and convenient, and facilitates the system to automatically process and manage. Using the e-commerce visualization platform interface to recommend and display promotion products can present product information to users in an intuitive way to attract users' attention. Accurate recommendations can reduce the time users spend looking for products of interest in a large number of products, improve marketing efficiency, and also improve user experience.

[0043] Specifically, the specific analysis of the promotion effect of the promotion product based on the real-time monitoring and prediction of the promotion process is as follows: the promotion product return rate is used as the evaluation basis for the promotion effect; the monitoring parameters of the promotion product are obtained, and the monitoring parameters include the number of interactions and the exposure; a time series model of the promotion product return rate is constructed, and the number of interactions and the exposure are used as exogenous variables. The specific expression of the time series model of the promotion product return rate is: ; In the formula Indicates the time point number. Indicates time point The predicted value of the return rate of the promoted product, represents a lag term of the predicted value of the promotion product return rate, It represents a difference term of the predicted value of the promotion product return rate. Indicates time point The number of interactions, represents a lagged term of the number of interactions, Indicates time point The exposure represents a lag term of exposure, , The coefficient of the autoregressive part indicates the autocorrelation of the predicted value of the promotion product return rate. The coefficient of the difference part indicates the stability of the predicted value of the promotion product return rate. , , , The coefficients of the exogenous variables represent the impact of the number of interactions and exposure on the predicted value of power generation, represents the error term, which represents unexplained random fluctuations.

[0044] The specific analysis for screening promotional products based on their promotional effects is as follows: obtaining a return rate threshold, comparing the promotional product return rate with the return rate threshold; when the promotional product return rate is lower than the return rate threshold, it is judged that the promotion effect of the promotional product is poor, and the recommendation display of the promotional product on the e-commerce visualization platform interface is reduced; when the promotional product return rate is greater than the return rate threshold, it is judged that the promotion effect of the promotional product is good, and the recommendation display of the promotional product on the e-commerce visualization platform interface is increased.

[0045] In this implementation scheme, the difference term is used to convert a non-stationary time series into a stationary series. Taking the predicted value of the return rate of the promotion product as an example, the difference term represents the difference between the predicted value of the return rate at the current time point and the predicted value of the return rate at the previous time point. Through the difference operation, non-stationary factors such as trends and seasonality in the time series can be eliminated, so that the model can better capture the changing patterns of the data.

[0046] A lag term refers to the value of a time series at a certain moment in the past. For example, a lag term of the predicted value of the return rate of a promoted product represents the predicted value of the return rate at the previous point in time, and a lag term of the number of interactions represents the number of interactions at the previous point in time. The lag term is used to reflect the autocorrelation of the time series, that is, the relationship between the current value and the past value.

[0047] The specific steps to build a time series model of the return rate of promoted products are: collect historical data such as the return rate, number of interactions and exposure of promoted products to ensure the accuracy and completeness of the data; pre-process the data, including processing missing values, outliers, etc. Interpolation and smoothing methods can be used to handle missing values, and statistical methods can be used to identify and correct outliers; unit root tests (such as ADF tests) and other methods can be used to test the stationarity of the promotion product return rate time series. If the time series is not stationary, differential processing is required until the series becomes stationary; the autocorrelation function (ACF) and partial autocorrelation function (PACF) of the stationary time series are observed to determine the order of the autoregressive part and the order of the differential part; based on business experience and data analysis, the lag order of exogenous variables (interaction times and exposure) is determined; the least squares method, maximum likelihood estimation method and other methods are used to estimate the parameters in the model; the estimated model is tested, including the autocorrelation test and heteroscedasticity test of the residuals. If the model residuals have autocorrelation or heteroscedasticity, the model needs to be adjusted and improved; the estimated model is used to predict the promotion product return rate, and the predicted results are compared with the actual return rate to evaluate the prediction performance of the model.

[0048] , , , , , , It is obtained through maximum likelihood estimation fitting, and the specific acquisition analysis is as follows: Error term The mean is 0 and the variance is The specific expression of the likelihood function is: ; In the formula Represents the total number of samples, that is, the total number of time points within the promotion product return rate prediction period. represents the model parameters ( , , , , , , ), Indicates time point The predicted value of the return rate of the promoted product, Indicates time point The exogenous variables include the number of interactions and exposure. By taking the logarithm of the likelihood function, we get the log-likelihood function. The specific expression of the log-likelihood function is: ; Maximizing the log-likelihood function is equivalent to minimizing the sum of squares of errors, so by maximizing the log-likelihood function, we get , , , , , , .

[0049] The rate of return measures the proportional relationship between the income and investment brought by the promotion activities, and reflects the economic benefits of the promotion activities. It is obtained by deducting the promotion costs from the sales revenue of the promoted products and comparing them with the promotion costs. The sales revenue can be obtained from the sales data of the e-commerce platform, and the promotion costs include advertising costs, marketing activity costs, etc., which can be obtained from the company's financial records.

[0050] The number of interactions refers to the number of interactive behaviors between users and promoted products, such as the sum of clicks, comments, shares and other operations. It reflects the user's attention and participation in the promoted products. It is obtained from the log records and data analysis tools of the e-commerce platform. The e-commerce platform will record every interactive behavior of the user. The number of interactions can be obtained by counting and analyzing these records.

[0051] Exposure refers to the number of times a promoted product is displayed to users, reflecting the potential audience range of the promoted product. It is obtained from the advertising delivery system and data analysis tools of the e-commerce platform. The advertising delivery system will record the number of times an advertisement is displayed, and the exposure is obtained by aggregating and counting these data.

[0052] The return rate threshold can be obtained by analyzing the return rate data of the company's past promotional activities, calculating the average return rate or median return rate, and using it as the return rate threshold. You can also set different return rate thresholds for each product or product category based on the historical return rates of different products. You can also refer to the promotion return rate data of other companies in the same industry to understand the industry average level or the return rate standards of excellent companies, and set the company's return rate threshold based on this. Or, based on the company's promotion costs, expected profits and market conditions, calculate a minimum return rate that can guarantee profitability, and use it as the return rate threshold. For example, if a company hopes to achieve a certain profit margin in its promotional activities, it can determine a suitable return rate threshold by calculating costs and expected benefits.

[0053] Using the return rate of promoted products as the basis for evaluation can directly reflect the effectiveness of promotion activities at the economic level, avoid the one-sidedness of evaluation based solely on traffic indicators (such as the number of interactions and exposures), and enable companies to more accurately understand the output of promotion inputs; by real-time monitoring of parameters such as the number of interactions and exposures, and building a time series model to predict the return rate of promoted products, companies can adjust their promotion strategies in a timely manner based on the predicted results, reduce recommended displays for products with low returns, and increase recommended displays for products with high returns, thereby optimizing resource allocation and improving overall promotion efficiency; incorporating the number of interactions and exposures as exogenous variables into the time series model takes into account the impact of these factors on the return rate of promoted products, making the prediction results more accurate and reliable and better reflecting the actual promotion situation; screening and adjusting recommended displays of promoted products based on the return rate threshold can help companies concentrate resources on promoting products with high returns and avoid wasting resources on products with low returns, thereby improving promotion efficiency and return on investment.

[0054] In summary, this application has at least the following effects:

[0055] Through the user portrait construction module, the multi-source data of users is screened and analyzed to build personalized user portraits, so that the system can deeply understand the interests, preferences and purchasing habits of each user, so as to provide users with highly personalized product recommendations; when recommending products, not only the individual needs of users are considered, but also the market trends and competitive situations can be grasped; the promotion products that are most suitable for specific users are screened out based on the user recommendation status and the market recommendation status, which reduces unnecessary waste of promotion resources, improves promotion efficiency, and reduces marketing costs; the promotion process is monitored in real time and the promotion effect is predicted, so that enterprises can understand the direction of promotion activities in advance. If the prediction results are not ideal, enterprises can adjust the promotion strategy in time to avoid further waste of resources. If the prediction effect is good, the promotion efforts can be increased to further increase the revenue. Products are screened and the promotion scores are updated according to the actual promotion effect of the promotion products. This enables the system to continuously learn and adapt to market changes and dynamic changes in user needs, and continuously optimize the promotion strategy.

[0056] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as systems. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0057] The present invention is described with reference to the block diagram of the system according to the embodiment of the present invention. It should be understood that the combination of each structure in the block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate a device for implementing the functions specified in each structure in the block diagram.

[0058] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a product including an instruction device that implements the functions specified in each structure of the structural diagram.

[0059] These computer program instructions may also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in each structure in the structural diagram.

[0060] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0061] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. An intelligent promotion integrated processing system based on e-commerce information, characterized in that: include: Data acquisition module, used to acquire multi-source user data and multi-source market data; The user portrait construction module is used to filter the multi-source user data and then construct a personalized user portrait based on the filtered multi-source user data; The recommendation status identification module is used to analyze personalized user portraits and multi-source market data to identify the user recommendation status and market recommendation status of each product; The promotion product screening module is used to obtain promotion scores for each product based on user recommendation status and market recommendation status, and then determine the promotion products for the user based on the promotion scores; The effect prediction module is used to monitor the promotion process of the promotion product in real time, and then predict the promotion effect of the promotion product based on the real-time monitoring of the promotion process; The promotion optimization and update module is used to screen the promotion products according to their promotion effects, and then update the promotion score of each product to adjust and optimize the promotion of the products.

2. According to claim 1, the intelligent promotion integrated processing system based on e-commerce information is characterized in that: The specific analysis of filtering multi-source user data and then building a personalized user portrait based on the filtered multi-source user data is as follows: Collect multi-source data of users based on user authorization, including but not limited to online platforms, social media, and e-commerce platforms. The multi-source data of users includes but is not limited to user gender, user age, user's frequently purchased product categories, user spending amount, user purchase history, user collection history, and user browsing history; Normalize the user multi-source data, handle missing values ​​and outliers, and then use the Pierre coefficient to calculate the correlation between the user multi-source data and the user's shopping behavior; Compare the correlation with the correlation threshold, and filter out the multi-source data of users whose correlation is lower than the correlation threshold; Integrate multi-source user data with correlation greater than or equal to the correlation threshold, normalize and label them as a personalized user portrait feature set, and build a personalized user portrait based on the personalized user portrait feature set.

3. According to claim 1, the intelligent promotion integrated processing system based on e-commerce information is characterized in that: The personalized user portraits and multi-source market data are analyzed separately to identify the user recommendation status and market recommendation status of each product. The specific analysis is as follows: Extracting a personalized user portrait feature set and market multi-source data, wherein the market multi-source data includes but is not limited to product categories, product inventory, product sales, product praise rate, product positive demand trend, and product market growth rate; Taking personalized user portraits and multi-source market data as input, the Logistic regression analysis method is used to build a user product interest evaluation model and a product marketing potential evaluation model, and the output is the user personalized recommendation coefficient and the market recommendation coefficient.

4. The intelligent promotion integrated processing system based on e-commerce information according to claim 3 is characterized in that: In the above-mentioned construction of the user product interest evaluation model and the product marketing potential evaluation model respectively by using the Logistic regression analysis method, the specific expression of the user product interest evaluation model is: , where represents the user personalized recommendation coefficient, Indicates that the user is interested in a product and purchases it. Represents a personalized user portrait feature set, , Represents the total number of personalized user portrait features, Represents the coefficient of the user product interest evaluation model; The specific expression of the product marketing potential evaluation model is: , where represents the market recommendation coefficient, It means that a product has promotion potential in the market. Represents a collection of multi-source market data. , Indicates the total number of multi-source market data, Represents the coefficient of the product marketing potential evaluation model.

5. The intelligent promotion integrated processing system based on e-commerce information according to claim 3 is characterized in that: The promotion score is obtained based on the user recommendation status and the market recommendation status, and then the specific analysis of the promotion product for the user is determined based on the promotion score: For each product, extract the user personalized recommendation coefficient and the market recommendation coefficient, and divide the user interest level and the market recommendation level of each product according to the user personalized recommendation coefficient and the market recommendation coefficient, wherein the user interest level includes a first interest level, a second interest level and a third interest level, and the market recommendation level includes a first recommendation level, a second recommendation level and a third recommendation level; The first interest level and the first recommendation level score are recorded as 2, the second interest level and the second recommendation level score are recorded as 1, and the third interest level and the third recommendation level score are recorded as 0; The sum of the user interest level and market recommendation level score of each product is recorded as the promotion score of each product; Products with a promotion score of 3 or more are marked as the user's promotional products, and then the user's promotional products are recommended and displayed using the e-commerce visualization platform interface.

6. The intelligent promotion integrated processing system based on e-commerce information according to claim 5 is characterized in that: The specific analysis of user interest level and market recommendation level of each product according to the user personalized recommendation coefficient and market recommendation coefficient is as follows: For the division of user interest levels, the 20th percentile of the sample size is used to divide the user's interest in the product into three levels according to the size of the user's personalized recommendation coefficient: when the user's personalized recommendation coefficient is below 0.5997, it is the third interest level; when the user's personalized recommendation coefficient is between 0.5997 and 0.7996, it is the second interest level; when the user's personalized recommendation coefficient is higher than 0.7996, it is the first interest level; For the division of market recommendation levels, the 20th percentile of the sample size is used, and the product's marketing potential is divided into three levels according to the size of the market recommendation coefficient: when the market recommendation coefficient is below 0.5997, it is the third recommendation level; when the market recommendation coefficient is between 0.5997 and 0.7996, it is the second recommendation level; when the market recommendation coefficient is higher than 0.7996, it is the first recommendation level.

7. The intelligent promotion integrated processing system based on e-commerce information according to claim 6 is characterized in that: The specific analysis of the promotion effect of the promotion product based on real-time monitoring and prediction of the promotion process is as follows: The promotion effect is evaluated based on the return rate of the promoted product; Obtain monitoring parameters of the promoted product, including the number of interactions and exposure; Construct a time series model of the return rate of promoted products, taking the number of interactions and exposure as exogenous variables. The specific expression of the time series model of the return rate of promoted products is: ; In the formula Indicates the time point number. Indicates time point The predicted value of the return rate of the promoted product, represents a lag term of the predicted value of the promotion product return rate, It represents a difference term of the predicted value of the promotion product return rate. Indicates time point The number of interactions, represents a lagged term of the number of interactions, Indicates time point The exposure represents a lag term of exposure, , The coefficient of the autoregressive part indicates the autocorrelation of the predicted value of the promotion product return rate. The coefficient of the difference part indicates the stability of the predicted value of the promotion product return rate. , , , The coefficients of the exogenous variables represent the impact of the number of interactions and exposure on the predicted value of power generation, represents the error term, which represents unexplained random fluctuations.

8. The intelligent promotion integrated processing system based on e-commerce information according to claim 7 is characterized in that: The specific analysis of screening the promotion products according to the promotion effect of the promotion products is as follows: Obtain a return rate threshold, compare the return rate of the promoted product with the return rate threshold, and when the return rate of the promoted product is lower than the return rate threshold, determine that the promotion effect of the promoted product is poor, and reduce the recommended display of the promoted product on the e-commerce visualization platform interface; When the return rate of the promoted product is greater than the return rate threshold, it is determined that the promotion effect of the promoted product is good, and the promoted product is added to the recommended display on the e-commerce visualization platform interface.

Images