Tobacco marketing method and device based on tobacco sales prediction and electronic equipment

Abstract

The invention discloses a tobacco marketing method and device based on tobacco sales prediction and electronic equipment. The method comprises the steps of determining a region comprehensive score of each region; determining predicted tobacco sales time sequence data of each region based on a Prophet model, and determining a tobacco sales predicted value of each region based on a LightGBM model; dividing each region into an experimental sub-region and a contrast sub-region, and determining a marketing strategy of each experimental sub-region; determining a marketing effect evaluation index of each experimental sub-region; performing high-frequency lightweight fine tuning on the model; marketing rules are optimized according to marketing effect evaluation indexes, and intermediate-frequency increment updating is carried out on the multi-factor feature weight matrix of each region; performing low-frequency global model retraining on the model, and performing weight calibration on the multi-factor feature weight matrix; the accuracy of tobacco sales prediction and the precision of the tobacco marketing strategy are improved, and the method can be applied to the technical field of artificial intelligence.

Description

Tobacco marketing methods, devices, and electronic equipment based on tobacco sales forecasting Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a tobacco marketing method, apparatus, and electronic device based on tobacco sales forecasting. Background Technology

[0002] Traditional tobacco sales forecasting and marketing management often rely on a singular, experience-based approach: at the forecasting level, they mainly depend on historical sales data for simple statistical analysis or calculations using a single model (such as a trend forecasting model based solely on time series). Data sources are limited to historical sales records and do not fully integrate multi-dimensional dynamic data such as geographic information, consumer profiles, and macroeconomic data. Furthermore, data processing often employs a full-scale calculation model, lacking efficient incremental updates and dimensionality reduction optimization mechanisms. At the marketing decision-making level, they rely primarily on manual experience and judgment, and marketing strategy formulation lacks systematic linkage with forecast results.

[0003] Traditional technologies suffer from three key drawbacks: First, it is difficult to balance prediction accuracy and efficiency. Traditional single-data-dimensional or single-model approaches cannot fully cover the complex influencing factors of tobacco sales, making it difficult for prediction results to reflect market trends and regional differences. Furthermore, the full-data-processing model consumes a large amount of computing resources, resulting in low data processing and prediction response efficiency. Second, marketing decisions lack precision and interpretability. Traditional human experience-driven or universally applicable marketing plans do not link multi-dimensional factor scores with sales prediction results, making it impossible to develop tailored strategies for the specific characteristics of different regions, leading to wasted marketing resources and unsatisfactory results. Third, the system lacks dynamic adaptation and iteration capabilities. Traditional technologies lack mechanisms such as data drift monitoring, layered optimization, and closed-loop effect feedback. They cannot quickly respond to short-term market fluctuations (such as changes in consumer preferences or policy adjustments) and cannot achieve long-term iterative upgrades of model parameters, weighting systems, and marketing strategies, resulting in poor system adaptability and limited long-term application value.

[0004] The above problems urgently need to be addressed.

[0005] Terminology Explanation: Prophet Model: A time series forecasting model that excels at capturing underlying trends, seasonal fluctuations, and holiday effects in data, and is highly robust to missing and outlier values.

[0006] LightGBM model: An ensemble learning model based on gradient boosting decision trees, which efficiently handles high-dimensional features and large-scale data, and is suitable for prediction tasks with feature table-like data.

[0007] Principal Component Analysis (PCA): A data dimensionality reduction technique that optimizes data storage and computation efficiency by extracting core features and removing redundant information.

[0008] Incremental update mechanism: This is an efficient data update method that processes only newly added or changed data, avoiding full data computation and reducing computing resource consumption.

[0009] Rule Engine: A decision-making system built on "IF-THEN" logical rules, enabling transparent and interpretable automated policy generation and execution.

[0010] Data drift index: an indicator that quantifies the difference between the current data distribution and the historical baseline distribution, used to determine whether the model needs to be adjusted to adapt to data changes.

[0011] Entropy weighting: A method that determines weights by calculating the entropy of feature information. The smaller the entropy, the higher the importance of the feature. It is suitable for dynamic weight adjustment. Summary of the Invention

[0012] The purpose of this invention is to at least partially solve one of the technical problems existing in the prior art.

[0013] Therefore, one objective of this invention is to provide a tobacco marketing method based on tobacco sales forecasting, which improves the accuracy of tobacco sales forecasting and the precision of tobacco marketing strategies.

[0014] Another objective of this invention is to provide a tobacco marketing device based on tobacco sales forecasting.

[0015] To achieve the above-mentioned technical objectives, the technical solution adopted in this embodiment of the invention includes: On one hand, this embodiment of the invention provides a tobacco marketing method based on tobacco sales forecasting, comprising the following steps: determining multi-dimensional feature values ​​and corresponding multi-factor feature weight matrices for each region based on geographic information data of multiple regions, historical tobacco sales time-series data, and consumer profile data; determining a comprehensive regional score for each region based on the multi-dimensional feature values ​​and the multi-factor feature weight matrix; determining predicted tobacco sales time-series data for each region based on the historical tobacco sales time-series data using the Prophet model; determining predicted tobacco sales values ​​for each region based on the predicted tobacco sales time-series data and the comprehensive regional score using the LightGBM model; dividing each region into experimental sub-regions and control sub-regions; and determining marketing strategies for each experimental sub-region based on the comprehensive regional score, the predicted tobacco sales values, and preset marketing rules; obtaining various... The marketing effectiveness evaluation index for each experimental sub-region is determined based on the first actual tobacco sales value of the control sub-region and the second actual tobacco sales value of each experimental sub-region after implementing the marketing strategy. The sales data drift index for each control sub-region is determined based on the first actual tobacco sales value and the predicted tobacco sales value, and the Prophet model and the LightGBM model are fine-tuned using a high-frequency, lightweight approach based on the sales data drift index. The marketing rules are optimized based on the marketing effectiveness evaluation index, and the multi-factor feature weight matrix for each region is updated incrementally using a medium-frequency approach. The Prophet model and the LightGBM model are retrained using low-frequency global model retraining based on real tobacco sales time-series data and actual tobacco sales values ​​from multiple marketing cycles, and the multi-factor feature weight matrix is ​​calibrated based on the subjective regional weights given by business experts.

[0016] Furthermore, in one embodiment of the present invention, the step of determining multi-dimensional feature values ​​and corresponding multi-factor feature weight matrices for each region based on geographic information data, historical tobacco sales time-series data, and consumer profile data, and determining the comprehensive regional score for each region based on the multi-dimensional feature values ​​and the multi-factor feature weight matrix, specifically includes: performing feature selection and principal component analysis on the geographic information data, the historical tobacco sales time-series data, and the consumer profile data to obtain multi-dimensional regional evaluation factors; standardizing the regional evaluation factors of each dimension to obtain the multi-dimensional feature values; determining the corresponding factor feature weights based on the historical data regression analysis coefficients and expert experience of each regional evaluation factor to obtain the multi-factor feature weight matrix; and performing weighted summation on the multi-dimensional feature values ​​based on the multi-factor feature weight matrix to obtain the comprehensive regional score.

[0017] Furthermore, in one embodiment of the present invention, the step of determining the predicted tobacco sales time-series data for each region based on the Prophet model according to the historical tobacco sales time-series data, and determining the predicted tobacco sales volume for each region based on the LightGBM model according to the predicted tobacco sales time-series data and the region's comprehensive score, specifically includes: inputting the historical tobacco sales time-series data into the pre-trained Prophet model to obtain the predicted tobacco sales time-series data for each region in a preset future time period; and inputting the predicted tobacco sales time-series data and the region's comprehensive score into the pre-trained LightGBM model to obtain the predicted tobacco sales volume for each region in the preset future time period.

[0018] Furthermore, in one embodiment of the present invention, the step of dividing each region into experimental sub-regions and control sub-regions, and determining the marketing strategy for each experimental sub-region based on the region's comprehensive score, the tobacco sales forecast, and preset marketing rules, specifically includes: dividing each region into experimental sub-regions and control sub-regions of comparable size based on the store distribution and pedestrian flow distribution of each region; determining a first tobacco sales forecast for the control sub-region and a second tobacco sales forecast for the experimental sub-region based on the tobacco sales forecast; and querying the marketing rules based on the region's comprehensive score and the second tobacco sales forecast to obtain the corresponding marketing strategy.

[0019] Furthermore, in one embodiment of the present invention, the step of determining the marketing effect evaluation index for each of the experimental sub-regions based on the first true value of tobacco sales and the second true value of tobacco sales specifically includes: determining the tobacco sales growth value for each of the experimental sub-regions based on the first true value of tobacco sales and the second true value of tobacco sales; calculating the strategy response rate, sales growth rate, and input-output ratio for each of the experimental sub-regions based on the tobacco sales growth value to obtain the marketing effect evaluation index.

[0020] Further, in one embodiment of the present invention, the step of determining the sales data drift index of each of the control sub-regions based on the first actual tobacco sales value and the predicted tobacco sales value, and performing high-frequency lightweight fine-tuning of the Prophet model and the LightGBM model based on the sales data drift index, specifically includes: determining the first predicted tobacco sales value of each of the control sub-regions based on the predicted tobacco sales value; determining the sales data drift index of each of the control sub-regions in each time period based on the first actual tobacco sales value and the first predicted tobacco sales value; when the sales data drift index is greater than a preset first threshold, determining the corresponding time period as a data drift period; when the number of data drift periods reaches a preset second threshold, using streaming gradient descent to fine-tune and update the weights of some leaf nodes of the LightGBM model, or fine-tuning the residual scaling factor of the Prophet model.

[0021] Furthermore, in one embodiment of the present invention, the step of optimizing the marketing rules according to the marketing effectiveness evaluation index and performing mid-frequency incremental updates on the multi-factor feature weight matrix of each region specifically includes: determining the average effect confidence interval of the corresponding marketing strategy according to the marketing effectiveness evaluation index, and optimizing the marketing rules according to the average effect confidence interval; determining the information entropy of multiple dimensions of regional evaluation factors in each region during the execution of the marketing strategy, and calculating the normalized weights of multiple dimensions of regional evaluation factors in each region according to the information entropy; and incrementally adjusting the factor feature weights of the regional evaluation factors according to the normalized weights to obtain the updated multi-factor feature weight matrix.

[0022] Further, in one embodiment of the present invention, the step of performing low-frequency global model retraining on the Prophet model and the LightGBM model based on real tobacco sales time-series data and actual tobacco sales values ​​over multiple marketing cycles, and calibrating the multi-factor feature weight matrix based on subjective regional weights given by business experts, specifically includes: constructing a first training sample based on the real tobacco sales time-series data and historical tobacco sales time-series data of each region over multiple marketing cycles; retraining the Prophet model based on the first training sample; constructing a second training sample based on the actual tobacco sales values ​​of each region over multiple marketing cycles, the predicted tobacco sales time-series data, and the regional comprehensive score; retraining the LightGBM model based on the second training sample; obtaining the subjective regional weights of multiple dimensions of regional evaluation factors for each region given by business experts in the next marketing cycle; and weighting and summing the subjective regional weights and the incrementally updated factor feature weights based on the objective data confidence coefficient to obtain the weight-calibrated multi-factor feature weight matrix.

[0023] On the other hand, embodiments of the present invention provide a tobacco marketing device based on tobacco sales forecasting, comprising: a multi-dimensional factor determination module, used to determine multi-dimensional feature values ​​and corresponding multi-factor feature weight matrices for each region based on geographic information data of multiple regions, historical tobacco sales time-series data, and consumer profile data, and to determine a regional comprehensive score for each region based on the multi-dimensional feature values ​​and the multi-factor feature weight matrix; a tobacco sales forecasting module, used to determine predicted tobacco sales time-series data for each region based on the historical tobacco sales time-series data using a Prophet model, and to determine predicted tobacco sales values ​​for each region based on the predicted tobacco sales time-series data and the regional comprehensive score using a LightGBM model; a marketing strategy determination module, used to divide each region into experimental sub-regions and control sub-regions, and to determine a marketing strategy for each experimental sub-region based on the regional comprehensive score, the predicted tobacco sales value, and preset marketing rules; and a marketing effect evaluation module, used to obtain a first [data point] for each control sub-region. The system comprises: a first real tobacco sales value and a second real tobacco sales value for each experimental sub-region after implementing the marketing strategy; a marketing effectiveness evaluation index for each experimental sub-region determined based on the first real tobacco sales value and the second real tobacco sales value; a high-frequency lightweight fine-tuning module for determining the sales data drift index for each control sub-region based on the first real tobacco sales value and the predicted tobacco sales value, and performing high-frequency lightweight fine-tuning on the Prophet model and the LightGBM model based on the sales data drift index; a mid-frequency incremental learning module for optimizing the marketing rules based on the marketing effectiveness evaluation index, and performing mid-frequency incremental updates on the multi-factor feature weight matrix for each region; and a low-frequency global calibration module for retraining the Prophet model and the LightGBM model using real tobacco sales time-series data and real tobacco sales values ​​from multiple marketing cycles, and calibrating the multi-factor feature weight matrix based on the subjective regional weights given by business experts.

[0024] On the other hand, embodiments of the present invention provide an electronic device, including: at least one processor; at least one memory for storing at least one program; when the at least one program is executed by the at least one processor, the at least one processor enables the above-described tobacco marketing method based on tobacco sales forecasting.

[0025] On the other hand, embodiments of the present invention also provide a computer-readable storage medium storing a processor-executable computer program that, when executed by a processor, implements the above-described tobacco marketing method based on tobacco sales forecasting.

[0026] On the other hand, embodiments of the present invention also provide a computer program product, including a computer program that, when executed by a processor, implements the above-described tobacco marketing method based on tobacco sales forecasting.

[0027] The advantages and beneficial effects of this invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention: Embodiments of this invention determine multi-dimensional feature values ​​and corresponding multi-factor feature weight matrices for each region based on geographic information data, historical tobacco sales time-series data, and consumer profile data. A comprehensive regional score is determined for each region based on the multi-dimensional feature values ​​and multi-factor feature weight matrices. Predicted tobacco sales time-series data for each region is determined based on the Prophet model using historical tobacco sales time-series data. Predicted tobacco sales volume for each region is determined based on the LightGBM model using predicted tobacco sales time-series data and the comprehensive regional score. Each region is divided into experimental sub-regions and control sub-regions. Marketing strategies for each experimental sub-region are determined based on the comprehensive regional score, predicted tobacco sales volume, and preset marketing rules. The actual first tobacco sales value of each control sub-region and the actual second tobacco sales value of each experimental sub-region after implementing the marketing strategy are obtained. The marketing effect evaluation index of each experimental sub-region is determined based on the actual first and second tobacco sales values. The sales data drift index of each control sub-region is determined based on the actual first and predicted tobacco sales values. The Prophet and LightGBM models are fine-tuned at high frequency based on the sales data drift index. The marketing rules are optimized based on the marketing effect evaluation index. The multi-factor feature weight matrix of each region is updated at medium frequency. The Prophet and LightGBM models are retrained at low frequency based on the real tobacco sales time series data and actual tobacco sales values ​​of multiple marketing cycles. The multi-factor feature weight matrix is ​​calibrated based on the subjective regional weights given by business experts. This invention, based on the Prophet and LightGBM models, utilizes geographic information data, historical tobacco sales time-series data, and consumer profile data to predict tobacco sales in various regions, thereby improving the accuracy of tobacco sales prediction. Based on preset marketing rules, it uses regional comprehensive scores and tobacco sales forecasts to determine marketing strategies for each region, improving the precision of tobacco marketing strategies. Furthermore, through high-frequency lightweight fine-tuning of the Prophet and LightGBM models, optimization of marketing rules, mid-frequency incremental updates to the multi-factor feature weight matrix, low-frequency global model retraining of the Prophet and LightGBM models, and weight calibration of the multi-factor feature weight matrix, the accuracy of tobacco sales prediction and the precision of tobacco marketing strategies are further improved. Attached Figure Description

[0028] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the embodiments of the present invention are described below. It should be understood that the drawings described below are only for the convenience of clearly describing some embodiments of the technical solutions of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1 is a flowchart of a tobacco marketing method based on tobacco sales forecasting provided in an embodiment of the present invention; Figure 2 is a structural block diagram of a tobacco marketing device based on tobacco sales forecasting provided in an embodiment of the present invention; Figure 3 is a structural block diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the embodiments of this invention; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this invention as detailed in the appended claims.

[0031] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein is for the purpose of describing embodiments of the invention only and is not intended to limit the invention.

[0032] The tobacco marketing method based on tobacco sales forecasting provided in this invention can be applied to a terminal, a server, or software running on either a terminal or a server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, or in-vehicle terminal, but is not limited to these. The server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The server can also be a node server in a blockchain network. The software can be an application that implements the tobacco marketing method based on tobacco sales forecasting, but is not limited to the above forms.

[0033] This invention can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This invention can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This invention can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0034] It should be noted that in various specific embodiments of the present invention, when processing data related to user identity or characteristics, such as user information, user behavior data, user historical data, and user location information, user permission or consent is obtained first. Furthermore, the collection, use, and processing of this data comply with relevant laws, regulations, and standards. In addition, when embodiments of the present invention require access to sensitive personal information of users, separate permission or consent from the user is obtained through pop-ups or redirection to a confirmation page. Only after obtaining the user's separate permission or consent is the necessary user-related data for the normal operation of the embodiments of the present invention acquired.

[0035] Referring to Figure 1, this embodiment of the invention provides a tobacco marketing method based on tobacco sales forecasting, specifically including the following steps: S101, determining the multi-dimensional feature values ​​and corresponding multi-factor feature weight matrices for each region based on geographic information data of multiple regions, historical tobacco sales time-series data, and consumer profile data, and determining the regional comprehensive score for each region based on the multi-dimensional feature values ​​and multi-factor feature weight matrices; S102, determining the predicted tobacco sales time-series data for each region based on the Prophet model and historical tobacco sales time-series data, and determining the predicted tobacco sales value for each region based on the LightGBM model and the predicted tobacco sales time-series data and regional comprehensive score; S103, dividing each region into experimental sub-regions and control sub-regions, and determining the marketing strategy for each experimental sub-region based on the regional comprehensive score, the predicted tobacco sales value, and preset marketing rules; S104, obtaining the control sub-regions... The first real tobacco sales value and the second real tobacco sales value of each experimental sub-region after implementing the marketing strategy are used to determine the marketing effect evaluation index for each experimental sub-region. S105. The sales data drift index of each control sub-region is determined based on the first real tobacco sales value and the predicted tobacco sales value, and the Prophet model and LightGBM model are fine-tuned at high frequency based on the sales data drift index. S106. The marketing rules are optimized based on the marketing effect evaluation index, and the multi-factor feature weight matrix of each region is updated at medium frequency. S107. The Prophet model and LightGBM model are retrained at low frequency globally based on the real tobacco sales time series data and the real tobacco sales value of multiple marketing cycles, and the multi-factor feature weight matrix is ​​calibrated based on the subjective regional weights given by business experts.

[0036] This invention, based on the Prophet and LightGBM models, utilizes geographic information data, historical tobacco sales time-series data, and consumer profile data to predict tobacco sales in various regions, thereby improving the accuracy of tobacco sales prediction. Based on preset marketing rules, it uses regional comprehensive scores and tobacco sales forecasts to determine marketing strategies for each region, improving the precision of tobacco marketing strategies. Furthermore, through high-frequency lightweight fine-tuning of the Prophet and LightGBM models, optimization of marketing rules, mid-frequency incremental updates to the multi-factor feature weight matrix, low-frequency global model retraining of the Prophet and LightGBM models, and weight calibration of the multi-factor feature weight matrix, the accuracy of tobacco sales prediction and the precision of tobacco marketing strategies are further improved.

[0037] As an optional implementation, multi-dimensional feature values ​​and corresponding multi-factor feature weight matrices for each region are determined based on geographic information data, historical tobacco sales time-series data, and consumer profile data from multiple regions. A comprehensive regional score for each region is then determined based on these multi-dimensional feature values ​​and the multi-factor feature weight matrix. Specifically, this includes: S201, performing feature selection and principal component analysis on geographic information data, historical tobacco sales time-series data, and consumer profile data to obtain multi-dimensional regional evaluation factors; S202, standardizing the regional evaluation factors for each dimension to obtain multi-dimensional feature values; S203, determining the corresponding factor feature weights based on the historical data regression analysis coefficients and expert experience for each regional evaluation factor to obtain a multi-factor feature weight matrix; and S204, performing a weighted summation of the multi-dimensional feature values ​​based on the multi-factor feature weight matrix to obtain the comprehensive regional score.

[0038] Specifically, by collecting and integrating multi-dimensional dynamic data, a data foundation is provided for subsequent forecasting and marketing decisions. Dynamic data includes geographic information, historical sales, consumer profiles, macroeconomic data, policies and regulations, etc. In order to optimize system performance and reduce computing resource consumption, feature selection and compression are performed at the data level to ensure the storage and query efficiency of the database.

[0039] By utilizing an incremental update mechanism to avoid the computational overhead of full data processing, a structured database containing multi-dimensional data such as geographic information and consumer behavior is constructed. Geographic information includes transportation hubs, commercial outlets, and population flow; consumer behavior includes consumption frequency, brand loyalty, and price sensitivity. These factors are used as candidate regional evaluation factors.

[0040] This system integrates geographic information, historical sales data, low-frequency macroeconomic indicators, and compliantly acquired terminal sales data. To balance data comprehensiveness and system performance, an incremental update mechanism is adopted, and feature selection and principal component analysis (PCA) are applied for dimensionality reduction to construct a structured data warehouse suitable for modeling. At this stage, the initial weights of each dimension are not dynamically optimized but are determined through quarterly review meetings, combining regression analysis coefficients from historical data with expert experience, forming a static quarterly weight vector. ,in Indicates quarterly period. Region. Static comprehensive score From the formula Calculations show that For the region Regional evaluation factors The standardized feature values. This score serves as a stable input for subsequent prediction models.

[0041] These data not only serve as input for sales forecasting but also form the foundation for generating precise marketing strategies. The constructed database allows for real-time updates and optimization of factor scores for each region.

[0042] As a further optional implementation, the predicted tobacco sales time series data for each region is determined based on the Prophet model using historical tobacco sales time series data, and the predicted tobacco sales volume for each region is determined based on the LightGBM model using the predicted tobacco sales time series data and the regional comprehensive score. Specifically, this includes: S301, inputting historical tobacco sales time series data into the pre-trained Prophet model to obtain the predicted tobacco sales time series data for each region in a preset future period; S302, inputting the predicted tobacco sales time series data and the regional comprehensive score into the pre-trained LightGBM model to obtain the predicted tobacco sales volume for each region in a preset future period.

[0043] Specifically, the system employs a lightweight ensemble model, with its core consisting of the Prophet model for processing time series data and the LightGBM model for processing feature tables. Prophet is responsible for capturing underlying trends, seasonality, and holiday effects, and its prediction results... As an important feature, it is consistent with the factor scores generated in the aforementioned steps. Other features All data is input into the LightGBM model for final prediction:

[0044] in, This is the final sales forecast. This is the prediction function for the LightGBM model. The prediction results are from the Prophet model; For regional static comprehensive scoring; To supplement special unforeseen factors; Focusing on patterns in the time dimension, Focusing on the fundamental characteristics of the region, By supplementing special unforeseen factors, the three factors comprehensively cover the factors affecting sales from different dimensions, thus solving the problem that a single model can only capture local patterns.

[0045] For example: a certain area High rating (densely populated, commercially developed), but If the period after the Spring Festival is traditionally a slow season, LightGBM will use feature interaction to predict that the sales volume in that region after the Spring Festival will be "higher than other regions, but lower than its own pre-Spring Festival level," which is consistent with the actual market logic.

[0046] The model is trained using a sliding window of data and has a defined update frequency: the prediction task is executed daily on a rolling basis, and the entire model is retrained monthly. The system outputs point predictions and confidence intervals for the next 7 and 30 days.

[0047] To achieve a dynamic early warning mechanism that adapts to market fluctuations, an adaptive threshold determination logic is introduced: by analyzing the distribution characteristics of historical sales data for the same period and combining it with the regional sales fluctuation coefficient, the early warning threshold is automatically calculated. When the lower limit of the confidence interval of the predicted value on the t-th day is lower than this dynamic threshold, it is determined that "sales significantly deviate from the normal range," and the system automatically triggers a business early warning signal, simultaneously pushing auxiliary information such as the basis for prediction and fluctuation characteristic analysis, providing timely and accurate guidance for manual intervention decisions. For example, when the lower limit of the confidence interval of the predicted value on the t-th day is lower than this dynamic threshold, it is determined that "sales significantly deviate from the normal range," and the system automatically triggers a business early warning signal, simultaneously pushing auxiliary information such as the basis for prediction and fluctuation characteristic analysis, providing timely and accurate guidance for manual intervention decisions. Lower limit of the predicted value for the day satisfy hour, This is the historical average for the same period, triggering a business early warning signal.

[0048] As an optional implementation method, each region is further divided into experimental sub-regions and control sub-regions. The marketing strategy for each experimental sub-region is determined based on the regional comprehensive score, tobacco sales forecast, and preset marketing rules. Specifically, this includes: S401, dividing each region into experimental sub-regions and control sub-regions of similar size based on the distribution of stores and pedestrian traffic in each region; S402, determining the first tobacco sales forecast for the control sub-region and the second tobacco sales forecast for the experimental sub-region based on the tobacco sales forecast; S403, querying the marketing rules based on the regional comprehensive score and the second tobacco sales forecast to obtain the corresponding marketing strategy.

[0049] Specifically, this embodiment of the invention employs rule-driven precision marketing strategy recommendations. Based on the prediction results and early warning signals from the preceding steps, this step transforms prediction into action suggestions. At this stage, instead of introducing complex reinforcement learning, a transparent and interpretable marketing strategy rule engine is constructed. This engine contains a series of "IF-THEN" rules, the condition portion of which is related to the aforementioned factor scoring. and prediction results Direct correlation. For example, a rule can be defined as: if the "price sensitivity" factor of a certain region... Above the threshold And the predicted sales volume is lower than the target value. If the conditions are met, the "targeted discount coupon" strategy is recommended. The system will automatically generate a draft marketing plan for eligible regions, including specific strategies, expected goals, and a list of target stores, along with the basis for the triggering rules. All strategy drafts are submitted to marketing management personnel for final approval and execution, forming a human-machine collaborative decision-making process.

[0050] As an optional implementation method, marketing effectiveness evaluation indicators for each experimental sub-region are determined based on the first and second true tobacco sales values. Specifically, this includes: S501, determining the tobacco sales growth value for each experimental sub-region based on the first and second true tobacco sales values; S502, calculating the strategy response rate, sales growth rate, and input-output ratio for each experimental sub-region based on the tobacco sales growth value to obtain the marketing effectiveness evaluation indicators.

[0051] Specifically, marketing effectiveness monitoring and data feedback: After the marketing campaign is launched, this step is responsible for quantitatively evaluating the effectiveness and collecting feedback data. The system monitors key indicators through the sales terminal network and constructs a marketing effectiveness evaluation matrix. The matrix contains information such as policy response rate. Sales growth rate Key metrics include Return on Investment (ROI). For activities that have undergone A / B testing, the system calculates the incremental improvement of the experimental group relative to the control group, using this as an estimate of the strategy's net effect. All assessment data and corresponding strategies - background information The data will be stored in a structured format to form a "strategy-effect" case library. This case library is the key fuel for subsequent system optimization, and its data quality directly determines the success or failure of model iteration.

[0052] The embodiments of the present invention adopt a hierarchical asynchronous optimization and real-time learning framework to balance response speed and computing resources, and divide the optimization task into three levels: high-frequency lightweight fine-tuning, mid-frequency incremental learning and low-frequency global recalibration.

[0053] As a further optional implementation, the sales data drift index of each control sub-region is determined based on the first actual tobacco sales value and the predicted tobacco sales value. The Prophet model and the LightGBM model are then fine-tuned using a high-frequency, lightweight approach based on the sales data drift index. Specifically, this includes: S601, determining the first predicted tobacco sales value for each control sub-region based on the predicted tobacco sales value; S602, determining the sales data drift index of each control sub-region at each time period based on the first actual tobacco sales value and the first predicted tobacco sales value; S603, when the sales data drift index is greater than a preset first threshold, determining the corresponding time period as a data drift period; S604, when the number of data drift periods reaches a preset second threshold, using streaming gradient descent to fine-tune and update the weights of some leaf nodes in the LightGBM model, or fine-tuning the residual scaling factor of the Prophet model.

[0054] Specifically, this step involves high-frequency, lightweight fine-tuning, focusing on rapidly adapting the prediction model without changing its structure. Only some parameters are adjusted to maintain a data drift index for tobacco sales data in each region or store cluster. The calculation formula is as follows:

[0055] in, It is the regional / store cluster data drift index at time t. Index for data sampling time, It is the squared Euclidean distance between the actual feature vector and the expected feature vector. It's a summation operation observed in the window. It is the actual feature vector. It is a predicted vector based on the distribution of historical data. This is the viewing window.

[0056] Data drift index The core objective is to quantify whether the current feature data distribution deviates from the historical baseline distribution. The calculation process follows the logic of "local aggregation - average quantization," and the specific steps are as follows: Determine the calculation range: Taking the current time t as the endpoint, truncate forward by a length of... The observation window is configured to ensure that drift is calculated based solely on recent data, avoiding interference from older data; the deviation at a single moment is calculated: for each moment within the window... Calculate the actual feature vectors respectively With the expected feature vector square Euclidean distance This process obtains the instantaneous drift value at each moment; sums the total deviation within the window: summing the instantaneous drift values ​​at all moments yields the total deviation within the entire observation window, reflecting the cumulative effect of data drift within the window period; and calculates the average drift index: dividing the total deviation by the observation window size. The average drift exponent at time t is obtained. This value eliminates the influence of window length on the results, making the drift index comparable under different window sizes.

[0057] when exceeding the threshold continuously This triggers online learning and fine-tuning of the prediction model for that region. For example, streaming gradient descent can be used to make small updates to the weights of some leaf nodes in the LightGBM model, or the residual scaling factor of the Prophet model can be adjusted. This process is computationally inefficient and enables on-demand, near real-time model adjustments, ensuring the prediction's responsiveness to rapid local changes.

[0058] As an optional implementation method, marketing rules are optimized based on marketing effectiveness evaluation indicators, and the multi-factor feature weight matrix of each region is updated incrementally at medium frequency. Specifically, this includes: S701, determining the average effect confidence interval of the corresponding marketing strategy based on the marketing effectiveness evaluation indicators, and optimizing the marketing rules based on the average effect confidence interval; S702, determining the information entropy of the regional evaluation factors of multiple dimensions in each region during the execution of the marketing strategy, and calculating the normalized weights of the regional evaluation factors of multiple dimensions in each region based on the information entropy; S703, incrementally adjusting the factor feature weights of the regional evaluation factors based on the normalized weights to obtain the updated multi-factor feature weight matrix.

[0059] Specifically, this step involves mid-frequency incremental learning and strategy optimization, responsible for incrementally optimizing marketing strategy rules and regional feature weights. The system continuously monitors the "strategy-effect" case library, and when the lower limit of the confidence interval for the average effect of a certain strategy category falls below a preset threshold... This layer's optimization is triggered when the global importance ranking of a certain feature dimension changes significantly (detected using the rank correlation coefficient within a sliding window). The optimization primarily includes: 1) updating the marketing strategy rule base through association rule mining based on new success stories; 2) weighting regional features using entropy weighting or a simplified online AHP method based on recent data. Incremental adjustments are made using the following formula:

[0060] in, As a smoothing factor, It is the updated factor characteristic weight of the i-th regional evaluation factor. It is the weight of the i-th region feature before the update. Let be the recent information entropy of the i-th region feature. Let be the reciprocal of the feature information entropy of the i-th region. It is the sum of the reciprocals of the feature entropy of all regions. The recent normalized weight of the i-th regional feature; the core purpose of this formula is to perform "incremental optimization" of the regional feature weights based on recent market data, so as to retain the stability of historical weights and adapt to recent market changes. The calculation process is divided into the following 4 steps: (1) Calculate the recent feature information entropy. For each regional characteristic (such as price sensitivity, population mobility, brand loyalty, etc.), recent data (such as dynamic data from the past month or the past three months) is extracted and processed using the information entropy formula. calculate, This represents the probability distribution of recent data for this feature; the more dispersed the recent data for a feature (e.g., the proportion of consumers with high and low price sensitivity is similar), the better. The larger the value, the weaker the feature's ability to distinguish sales in the recent period; conversely, the smaller the value, the stronger the ability to distinguish. (2) Transform and normalize the importance of recent features. Take the reciprocal Transforming uncertainty into importance scores, i.e. The smaller the feature size, the stronger the discriminative power. The larger the value, the higher the importance score.

[0061] Using a single feature Divide by all features The sum ,get (2) Ensure that the sum of the recent importance percentages of all features is 1, forming a recent weight that can be directly compared; (3) Integrate historical weights and recent weights, and use a weighted summation method to adjust the historical weights before the update. Integrating with recent normalized weights: Part 1 Preserve the core influence of historical weights and avoid sudden weight changes due to short-term fluctuations; Part Two Introduce the importance of features from recent data to adapt to the latest market changes. For example, if a certain feature has significantly increased its impact on sales recently, then the proportion of that feature will increase; (4) Output the updated weights. The final updated weights are obtained by weighting and summing the two parts. These weights not only continue the accumulation of historical models, but also achieve "incremental optimization" through recent data, ensuring that the feature weights evolve in sync with market driving factors.

[0062] As an optional implementation, the Prophet and LightGBM models are retrained using low-frequency global model retraining based on real tobacco sales time-series data and actual tobacco sales values ​​across multiple marketing cycles. The multi-factor feature weight matrix is ​​then calibrated based on subjective regional weights provided by business experts. Specifically, this includes: S801, constructing a first training sample based on real and historical tobacco sales time-series data for each region across multiple marketing cycles, and retraining the Prophet model using the first training sample; S802, constructing a second training sample based on actual tobacco sales values, predicted tobacco sales time-series data, and regional comprehensive scores for each region across multiple marketing cycles, and retraining the LightGBM model using the second training sample; S803, obtaining the subjective regional weights of multiple dimensions of regional evaluation factors for each region in the next marketing cycle, as provided by business experts; and S804, weighting and summing the subjective regional weights and the incrementally updated factor feature weights based on the objective data confidence coefficients to obtain the weight-calibrated multi-factor feature weight matrix.

[0063] Specifically, this step involves low-frequency global recalibration and architecture iteration, performing computationally intensive tasks to provide a solid foundation for the system. The third layer is triggered at fixed intervals (a preset number of marketing cycles) and mainly includes: global retraining and selection of the prediction model: retraining all candidate models using all historical data and selecting the optimal model or ensemble combination for the next cycle through time-series cross-validation; expert calibration of the weight system: automatically adjusting the weights of the mid-frequency layer. Compared to subjective weights given by business experts based on strategic direction The data is then merged to generate the authoritative weight benchmark for the next cycle. The fusion formula is:

[0064] in, The confidence coefficient for objective data. As the authoritative weighting benchmark for the next cycle, The weights are automatically adjusted for the intermediate frequency layer. This is the softmax normalized result of the weights.

[0065] Automatically adjusted weights for the intermediate frequency layer Normalization is performed using the softmax function: The original automatically adjusted weights are converted into standardized weights with values ​​between (0,1) and the sum of all feature weights is 1, which solves the problem of inconsistent weight levels of different features and provides a basis for integration with expert subjective weights V. It assigns a confidence level γ to the weights of normalized objective data, reflecting the importance of features reflected in recent market data (e.g., if a feature has a significant impact on sales recently, its normalized weight is higher, and the contribution of this part also increases accordingly). This refers to assigning confidence levels to the expert's subjective weight V. This reflects the expert's strategic-level judgment. For example, based on policy requirements, experts may increase the subjective weight of compliance characteristics, which will directly increase its proportion in the final weight. The weighted results of the above two parts are added together to obtain the authoritative weight benchmark for the next cycle. This result preserves the data's adaptability to market dynamics while incorporating experts' considerations for long-term strategies, avoiding the short-term utilitarianism that may result from a single data-driven approach or the "subjective bias" that may exist in a single expert's judgment.

[0066] In some optional embodiments, the present invention also employs offline training based on a reinforcement learning policy simulator: after accumulating sufficient data, a reinforcement learning model is trained offline using historical and simulation data to prepare for future replacement or enhancement of the rule engine.

[0067] Offline training of the reinforcement learning policy simulator, as a key component of the system's "low-frequency global recalibration and architecture iteration," serves both as a forward-looking upgrade reserve for the existing rule engine and a core support for the system's long-term intelligent evolution. This training is based on massive historical data accumulated from a case library of policies and effects, combined with market scenario simulation data to construct rich training samples. The reinforcement learning model is trained and optimized offline, effectively avoiding interference from online training on real-time business operations while ensuring the depth and accuracy of model training. Once the model training matures, it will not only serve as a powerful supplement to the existing rule engine, further enhancing the flexibility and accuracy of marketing strategy recommendations, but also has the potential to replace the traditional IF-THEN rule engine, achieving a leap from rule-driven to intelligent autonomous decision-making, and providing more adaptive and forward-looking technical support for tobacco sales forecasting and marketing decisions.

[0068] It should be noted that the Prophet model and LightGBM model in the embodiments of the present invention can be used to accurately predict tobacco sales in various regions after multiple optimizations and updates, and the optimized marketing rules can make marketing decisions based on the tobacco sales prediction results and the regional comprehensive score.

[0069] The method steps of the embodiments of the present invention have been described above. It can be understood that the embodiments of the present invention, based on the Prophet model and the LightGBM model, utilize geographic information data, historical tobacco sales time-series data, and consumer profile data to predict tobacco sales in various regions, thereby improving the accuracy of tobacco sales prediction. Based on preset marketing rules, the embodiments utilize regional comprehensive scores and tobacco sales prediction values ​​to determine marketing strategies for each region, thereby improving the accuracy of tobacco marketing strategies. Furthermore, through high-frequency lightweight fine-tuning of the Prophet and LightGBM models, optimization and adjustment of marketing rules, mid-frequency incremental updates of the multi-factor feature weight matrix, low-frequency global model retraining of the Prophet and LightGBM models, and weight calibration of the multi-factor feature weight matrix, the accuracy of tobacco sales prediction and the precision of tobacco marketing strategies are further improved.

[0070] Compared with existing technologies, the embodiments of this invention also have the following advantages: 1) A lightweight integrated model is adopted, combining the advantages of time series processing models and feature table processing models. On the one hand, it accurately captures the basic trend, seasonal fluctuations, and holiday effects of sales. On the other hand, it deeply integrates regional comprehensive scores and other auxiliary features. Through model collaborative computation, it outputs the sales point prediction results and corresponding confidence intervals for the next 7 days and 30 days, effectively taking into account both the trend regularity of the time dimension and the market stability of the feature dimension. The prediction accuracy is significantly better than the prediction effect of a single model. Relying on the dynamic update mechanism of daily rolling prediction and monthly full model retraining, coupled with real-time monitoring of data drift status, it ensures that the model can quickly adapt to short-term market fluctuations and potential changes. When the lower limit of the confidence interval of predicted sales is lower than the normal level threshold, the system automatically triggers a business early warning signal, providing timely and clear guidance for manual decision-making, effectively avoiding operational risks such as inventory backlog and market share loss caused by a sharp decline in sales. This collaborative mode of "dynamic prediction + intelligent early warning" not only realizes the accuracy and quantitative presentation of tobacco sales prediction, but also greatly improves the response speed to market changes, providing a solid quantitative basis and effective risk warning for marketing decisions.

[0071] 2) Through a transparent and interpretable rule engine, combined with regional comprehensive scores and sales forecast results, precise marketing strategy drafts are generated based on the market characteristics and sales expectations of different regions. For example, for regions with high price sensitivity and predicted sales below the target value, targeted discount strategies are automatically matched, realizing the rapid implementation from forecast data to practical solutions. Step four monitors key indicators of marketing activities in real time through the sales terminal network, and builds an effect evaluation system that includes core indicators such as strategy response rate, sales improvement rate, and return on investment. At the same time, it integrates information such as strategy details and market background to form a "strategy-effect" case library, providing core data support for subsequent system optimization. Step five's layered asynchronous optimization framework quickly adapts to local market changes through high-frequency lightweight fine-tuning, optimizes strategy rules and feature weights through medium-frequency incremental learning, and improves the model architecture and weight system through low-frequency global recalibration. This ensures the system's rapid response to short-term market fluctuations and achieves long-term steady iteration of strategy rules, model parameters, and system architecture. The entire closed loop is driven by data, taking into account both the immediacy of decision-making and the evolution of system development, ultimately achieving a simultaneous improvement in the accuracy of tobacco sales forecasting, the effectiveness of marketing strategy implementation, and the system's ability to adapt to market changes.

[0072] 3) By comprehensively integrating multi-dimensional dynamic data such as geographic information, historical sales, consumer profiles, macroeconomic data, and policies and regulations, and employing an incremental update mechanism and principal component analysis for dimensionality reduction, a structured data warehouse is constructed to generate regional static comprehensive scores. This ensures data comprehensiveness to cover core market influencing factors while effectively optimizing system performance and reducing computational resource consumption, providing stable and efficient data support for subsequent predictive analysis and marketing decisions. Static quarterly weights are determined jointly by combining historical data regression analysis results and expert experience, enabling the regional comprehensive scores to objectively reflect the inherent market potential and basic conditions of different regions. Standardized feature data further ensures the scientific nature and horizontal comparability of the scores. This step successfully addresses the pain points of traditional forecasting, such as single data dimensions and low processing efficiency. Through structured storage, real-time updates of factor scores for each region are achieved, building a crucial bridge between data collection and model prediction, and providing a high-quality, highly reliable input foundation for subsequent accurate forecasting.

[0073] Referring to Figure 2, this embodiment of the invention provides a tobacco marketing device based on tobacco sales forecasting, comprising: a multi-dimensional factor determination module, used to determine the multi-dimensional feature values ​​and corresponding multi-factor feature weight matrices of each region based on geographic information data of multiple regions, historical tobacco sales time-series data, and consumer profile data, and to determine the regional comprehensive score of each region based on the multi-dimensional feature values ​​and the multi-factor feature weight matrix; a tobacco sales forecasting module, used to determine the predicted tobacco sales time-series data of each region based on the Prophet model and historical tobacco sales time-series data, and to determine the predicted tobacco sales value of each region based on the predicted tobacco sales time-series data and the regional comprehensive score based on the LightGBM model; a marketing strategy determination module, used to divide each region into experimental sub-regions and control sub-regions, and to determine the marketing strategy of each experimental sub-region based on the regional comprehensive score, the predicted tobacco sales value, and preset marketing rules; and a marketing effect evaluation module, used to obtain the comprehensive score of each control sub-region. The system comprises four modules: a first real tobacco sales value and a second real tobacco sales value for each experimental sub-region after implementing the marketing strategy; a high-frequency lightweight fine-tuning module to determine the sales data drift index for each control sub-region based on the first real tobacco sales value and the predicted tobacco sales value, and to perform high-frequency lightweight fine-tuning on the Prophet and LightGBM models based on the sales data drift index; a mid-frequency incremental learning module to optimize marketing rules based on the marketing effectiveness evaluation indicators and to perform mid-frequency incremental updates on the multi-factor feature weight matrix for each region; and a low-frequency global calibration module to retrain the Prophet and LightGBM models using real tobacco sales time-series data and real tobacco sales values ​​from multiple marketing cycles, and to calibrate the multi-factor feature weight matrix based on the subjective regional weights given by business experts.

[0074] It is understood that the content of the above method embodiments is applicable to the present device embodiments. The specific functions implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0075] Referring to FIG3, an embodiment of the present invention provides an electronic device, including: at least one processor; at least one memory for storing at least one program; when the at least one program is executed by the at least one processor, the at least one processor implements the above-described tobacco marketing method based on tobacco sales forecasting.

[0076] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0077] This invention also provides a computer-readable storage medium storing a processor-executable computer program that, when executed by a processor, implements the above-described tobacco marketing method based on tobacco sales forecasting.

[0078] This invention provides a computer-readable storage medium that can execute a tobacco marketing method based on tobacco sales forecasting provided in the method embodiments of this invention. It can execute any combination of the implementation steps of the method embodiments and has the corresponding functions and beneficial effects of the method.

[0079] This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described tobacco marketing method based on tobacco sales forecasting.

[0080] It is understood that the content of the above method embodiments is applicable to the embodiments of this program product. The specific functions implemented by the embodiments of this program product are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0081] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0082] The embodiments described in this invention are for the purpose of more clearly illustrating the technical solutions of the embodiments of this invention, and do not constitute a limitation on the technical solutions provided by the embodiments of this invention. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this invention are also applicable to similar technical problems.

[0083] The terms "first," "second," "third," "fourth," etc. (if present) in the specification and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0084] In some alternative embodiments, the functions / operations mentioned in the block diagrams may not occur in the order shown in the operation diagrams. For example, depending on the functions / operations involved, two consecutively shown blocks may actually be executed substantially simultaneously, or the aforementioned blocks may sometimes be executed in reverse order. Furthermore, the embodiments presented and described in the flowcharts of this invention are provided by way of example to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is changed and sub-operations described as part of a larger operation are executed independently.

[0085] Furthermore, although the invention has been described in the context of functional modules, it should be understood that, unless otherwise stated, one or more of the aforementioned functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the invention. Rather, given the properties, functions, and internal relationships of the various functional modules in the apparatus disclosed herein, the actual implementation of the module will be understood within the scope of conventional skill of an engineer. Therefore, those skilled in the art can implement the invention as set forth in the claims using ordinary techniques without excessive experimentation. It is also understood that the specific concepts disclosed are merely illustrative and not intended to limit the scope of the invention, which is determined by the full scope of the appended claims and their equivalents.

[0086] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0087] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0088] More specific examples (a non-exhaustive list) of computer-readable media include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the aforementioned program can be printed, because the aforementioned program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0089] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0090] In the foregoing description of this specification, references to terms such as "one embodiment," "another embodiment," or "some embodiments" indicate that a specific feature, structure, material, or characteristic described in connection with an embodiment or example is included in at least one embodiment or example of the present invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0091] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

[0092] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the above embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of the present invention.

Claims

1. A tobacco marketing method based on tobacco sales forecasting, characterized in that, Includes the following steps: Based on geographic information data, historical tobacco sales time-series data, and consumer profile data from multiple regions, multi-dimensional feature values ​​and corresponding multi-factor feature weight matrices are determined for each region. A comprehensive regional score is then determined based on these multi-dimensional feature values ​​and the multi-factor feature weight matrices. Predicted tobacco sales time-series data for each region is determined using the Prophet model based on the historical tobacco sales time-series data. A predicted tobacco sales volume for each region is determined using the LightGBM model based on the predicted tobacco sales time-series data and the comprehensive regional score. Each region is divided into experimental sub-regions and control sub-regions. A marketing strategy for each experimental sub-region is determined based on the comprehensive regional score, the predicted tobacco sales volume, and preset marketing rules. The first actual tobacco sales volume for each control sub-region and the actual sales volume after implementing the marketing strategy for each experimental sub-region are obtained. The marketing effectiveness evaluation index for each experimental sub-region is determined based on the actual tobacco sales value of the second sub-region and the actual tobacco sales value of the first and second sub-regions. The sales data drift index for each control sub-region is determined based on the actual tobacco sales value of the first and the predicted tobacco sales value, and the Prophet and LightGBM models are fine-tuned using high-frequency lightweight adjustments based on the sales data drift index. The marketing rules are optimized based on the marketing effectiveness evaluation index, and the multi-factor feature weight matrix for each region is updated incrementally using medium frequency. The Prophet and LightGBM models are retrained using low-frequency global model retraining based on real tobacco sales time-series data and actual tobacco sales values ​​from multiple marketing cycles, and the multi-factor feature weight matrix is ​​calibrated based on the subjective regional weights given by business experts.

2. The tobacco marketing method based on tobacco sales forecasting according to claim 1, characterized in that, The process of determining multi-dimensional feature values ​​and corresponding multi-factor feature weight matrices for each region based on geographic information data, historical tobacco sales time-series data, and consumer profile data, and then determining a comprehensive regional score for each region based on the multi-dimensional feature values ​​and the multi-factor feature weight matrix, specifically includes: performing feature selection and principal component analysis on the geographic information data, the historical tobacco sales time-series data, and the consumer profile data to obtain multi-dimensional regional evaluation factors; standardizing the regional evaluation factors in each dimension to obtain the multi-dimensional feature values; determining the corresponding factor feature weights based on the historical data regression analysis coefficients and expert experience of each regional evaluation factor to obtain the multi-factor feature weight matrix; and performing a weighted summation of the multi-dimensional feature values ​​based on the multi-factor feature weight matrix to obtain the comprehensive regional score.

3. The tobacco marketing method based on tobacco sales forecasting according to claim 1, characterized in that, The process of determining predicted tobacco sales time-series data for each region based on the Prophet model using historical tobacco sales time-series data, and determining predicted tobacco sales volume for each region based on the LightGBM model using the predicted tobacco sales time-series data and the region's comprehensive score, specifically includes: inputting the historical tobacco sales time-series data into the pre-trained Prophet model to obtain the predicted tobacco sales time-series data for each region in a preset future period; and inputting the predicted tobacco sales time-series data and the region's comprehensive score into the pre-trained LightGBM model to obtain the predicted tobacco sales volume for each region in the preset future period.

4. The tobacco marketing method based on tobacco sales forecasting according to claim 1, characterized in that, The process of dividing each region into experimental sub-regions and control sub-regions, and determining the marketing strategy for each experimental sub-region based on the region's comprehensive score, the tobacco sales forecast, and preset marketing rules, specifically includes: dividing each region into experimental sub-regions and control sub-regions of similar size based on the store distribution and pedestrian flow distribution of each region; determining a first tobacco sales forecast for the control sub-region and a second tobacco sales forecast for the experimental sub-region based on the tobacco sales forecast; and querying the marketing rules based on the region's comprehensive score and the second tobacco sales forecast to obtain the corresponding marketing strategy.

5. A tobacco marketing method based on tobacco sales forecasting according to claim 1, characterized in that, The step of determining the marketing effectiveness evaluation index for each experimental sub-region based on the first and second actual tobacco sales values ​​specifically includes: determining the tobacco sales growth value for each experimental sub-region based on the first and second actual tobacco sales values; and calculating the strategy response rate, sales increase rate, and input-output ratio for each experimental sub-region based on the tobacco sales growth value to obtain the marketing effectiveness evaluation index.

6. The tobacco marketing method based on tobacco sales forecasting according to claim 1, characterized in that, The step of determining the sales data drift index of each of the control sub-regions based on the first actual tobacco sales value and the predicted tobacco sales value, and performing high-frequency lightweight fine-tuning of the Prophet model and the LightGBM model based on the sales data drift index, specifically includes: determining the first predicted tobacco sales value of each of the control sub-regions based on the predicted tobacco sales value; determining the sales data drift index of each of the control sub-regions in each time period based on the first actual tobacco sales value and the first predicted tobacco sales value; when the sales data drift index is greater than a preset first threshold, determining the corresponding time period as a data drift period; when the number of data drift periods reaches a preset second threshold, using streaming gradient descent to fine-tune and update the weights of some leaf nodes of the LightGBM model, or fine-tuning the residual scaling factor of the Prophet model.

7. A tobacco marketing method based on tobacco sales forecasting according to claim 1, characterized in that, The step of optimizing the marketing rules based on the marketing effectiveness evaluation indicators and performing mid-frequency incremental updates on the multi-factor feature weight matrix for each region specifically includes: determining the average effect confidence interval of the corresponding marketing strategy based on the marketing effectiveness evaluation indicators, and optimizing the marketing rules based on the average effect confidence interval; determining the information entropy of multiple dimensions of regional evaluation factors in each region during the execution of the marketing strategy, and calculating the normalized weights of multiple dimensions of regional evaluation factors in each region based on the information entropy; and incrementally adjusting the factor feature weights of the regional evaluation factors based on the normalized weights to obtain the updated multi-factor feature weight matrix.

8. A tobacco marketing method based on tobacco sales forecasting according to claim 1, characterized in that, The process of retraining the Prophet and LightGBM models using low-frequency global model retraining based on real tobacco sales time-series data and actual tobacco sales values ​​across multiple marketing cycles, and calibrating the multi-factor feature weight matrix based on subjective regional weights provided by business experts, specifically includes: constructing a first training sample based on real tobacco sales time-series data and historical tobacco sales time-series data for each region across multiple marketing cycles; retraining the Prophet model based on the first training sample; constructing a second training sample based on actual tobacco sales values, predicted tobacco sales time-series data, and regional comprehensive scores for each region across multiple marketing cycles; retraining the LightGBM model based on the second training sample; obtaining the subjective regional weights of multiple dimensions of regional evaluation factors for each region in the next marketing cycle, provided by business experts; and weighting and summing the subjective regional weights and the incrementally updated factor feature weights based on the objective data confidence coefficient to obtain the weight-calibrated multi-factor feature weight matrix.

9. A tobacco marketing device based on tobacco sales forecasting, characterized in that, include: The multi-dimensional factor determination module is used to determine the multi-dimensional feature values ​​and corresponding multi-factor feature weight matrices for each region based on geographical information data, historical tobacco sales time-series data, and consumer profile data, and to determine the comprehensive regional score for each region based on the multi-dimensional feature values ​​and the multi-factor feature weight matrices. The tobacco sales forecasting module is used to determine the predicted tobacco sales time-series data for each region based on the historical tobacco sales time-series data using the Prophet model, and to determine the predicted tobacco sales value for each region based on the predicted tobacco sales time-series data and the comprehensive regional score using the LightGBM model. The marketing strategy determination module is used to divide each region into experimental sub-regions and control sub-regions, and to determine the marketing strategy based on the predicted tobacco sales time-series data and the comprehensive regional score. The marketing strategy for each experimental sub-region is determined based on the comprehensive regional score, the predicted tobacco sales value, and the preset marketing rules. A marketing effectiveness evaluation module is used to obtain the first actual tobacco sales value for each control sub-region and the second actual tobacco sales value for each experimental sub-region after implementing the marketing strategy. Based on the first and second actual tobacco sales values, a marketing effectiveness evaluation index for each experimental sub-region is determined. A high-frequency lightweight fine-tuning module is used to determine the sales data drift index for each control sub-region based on the first actual tobacco sales value and the predicted tobacco sales value, and to perform high-frequency lightweight fine-tuning on the Prophet model and the LightGBM model based on the sales data drift index. The mid-frequency incremental learning module is used to optimize the marketing rules based on the marketing effectiveness evaluation indicators and to perform mid-frequency incremental updates on the multi-factor feature weight matrix of each region. The low-frequency global calibration module is used to retrain the Prophet model and the LightGBM model using real tobacco sales time-series data and actual tobacco sales values ​​from multiple marketing cycles, and to calibrate the weights of the multi-factor feature weight matrix based on the subjective region weights given by business experts.

10. An electronic device, characterized in that, include: At least one processor; At least one memory for storing at least one program; when the at least one program is executed by the at least one processor, the at least one processor implements a tobacco marketing method based on tobacco sales forecasting as described in any one of claims 1 to 8.

Images