Tobacco enterprise marketing scheme auxiliary optimization system based on big data

Abstract

The invention relates to the technical field of enterprise marketing, and discloses a tobacco enterprise marketing scheme auxiliary optimization system based on big data, and compared with a marketing management means implemented by a traditional tobacco enterprise depending on artificial experience and static data analysis, the marketing scheme auxiliary optimization system based on the big data has the advantages that the marketing management means is more accurate, and the marketing efficiency is improved. A whole-process closed-loop mechanism of data acquisition, model construction, strategy evaluation, continuous iteration and result feedback is broken through, the system not only realizes automatic processing and modeling of marketing data, but also has strategy optimization and dynamic learning capabilities, the response speed of an enterprise to market change and the accuracy of strategy execution are remarkably improved, and the marketing efficiency is improved. The system can effectively identify potential risks, provides an accurate intervention scheme, and enhances the robustness and stability of a sales system especially in changeable situations of frequent policy adjustment, consumption preference transfer, complex channel structure and the like.

Description

Technical Field

[0001] The present invention relates to the field of enterprise marketing technology, and in particular to a tobacco enterprise marketing plan auxiliary optimization system based on big data. Background Art

[0002] Data marketing is a marketing approach applied to the internet advertising industry based on massive amounts of data from multiple platforms and leveraging big data technology. The core of big data marketing is to deliver online ads to the right people at the right time, through the right media, and in the right way. Big data marketing originated from the internet industry and has a profound impact on it. Leveraging the collection of big data from multiple platforms and the analytical and predictive capabilities of big data technology, it enables more precise and effective advertising, delivering higher returns on investment for brands.

[0003] Current tobacco companies' marketing practices are plagued by widespread reliance on manual experience, limited data utilization, and slow response times. Although some companies have deployed point-of-sale systems or customer relationship management platforms, these data types remain poorly integrated, resulting in superficial user segmentation and a lack of dynamic reflection of consumer preferences and market changes. This results in a lack of targeted marketing content and delayed feedback on implementation results, further impacting resource allocation efficiency and market competitiveness.

[0004] Therefore, we proposed a tobacco enterprise marketing plan auxiliary optimization system based on big data to solve the above problems. Summary of the Invention

[0005] The purpose of the present invention is to provide a tobacco enterprise marketing plan auxiliary optimization system based on big data, so as to solve the problems proposed in the above background technology that various types of data cannot be effectively integrated, the user segmentation dimensions are superficial, and it is difficult to dynamically reflect consumer preferences and market changes, resulting in a lack of pertinence in marketing content and untimely feedback on execution results.

[0006] To achieve the above-mentioned purpose, the present invention provides the following technical solutions: a tobacco enterprise marketing plan auxiliary optimization system based on big data, comprising: a training module, an evaluation module, an iteration module and a feedback module;

[0007] The training module is used to collect and preprocess multi-source data, couple the preprocessed data to generate a basic model, and perform calibration analysis on the basic model to determine its practicality;

[0008] The evaluation module is used to collect multi-source data from the enterprise and sales process, and pre-process it, and input the pre-processed data into the model for coupling and optimization;

[0009] The iterative module is used to collect multi-source data from real-time enterprises and sales processes, and pre-process the data, input the pre-processed data into the model for coupling, and optimize the model;

[0010] The feedback module is used to transmit the collected data and model optimization results to the visualization terminal.

[0011] Preferably, the training module includes a feature acquisition unit, a debugging unit and a triggering unit;

[0012] The feature collection unit is used to collect multi-source data, including short-term purchase frequency DQM, single sales volume DCX, sales interval FMJ, daily sales RXL, monthly inventory turnover rate YKC, number of display items CGP, brand sales share change rate PSB, distribution coverage rate PHL and cumulative number of product types LJP, and pre-process and dimensionlessly convert the collected multi-source data;

[0013] The debugging unit is used to couple the collected multi-source data, input the processed multi-source data into the pre-trained deep learning framework, perform feature fusion through a multi-layer neural network, and then calculate the debugging calibration coefficient TSX. The specific calculation formula is as follows:

[0014]

[0015] Where: DQM is the number of short-term purchases, DCX is the single sales volume, FMJ is the sales interval, RXL is the daily sales volume, YKC is the monthly inventory turnover rate, CGP is the number of display items, PSB is the change rate of brand sales share, PHL is the distribution coverage rate, and LJP is the cumulative number of product types;

[0016] The trigger unit is used to perform data analysis on the debugging calibration coefficient and determine whether the current model is practical based on the analysis result.

[0017] Preferably, the trigger unit specifically analyzes the debugging calibration coefficient as follows:

[0018] When TSX≤0.75, it means that the current model cannot be used directly and the weight value needs to be adjusted and the model needs to be recalibrated;

[0019] When TSX>0.75, it means that the current model can be used directly.

[0020] Preferably, the evaluation module includes an information collection unit, a service adaptation unit and a comparison unit;

[0021] The information collection unit is used to collect multi-source data in the enterprise and sales processes, including regional distribution frequency QPS, channel coverage width QDF, brand single-product launch cycle PSZ, average terminal order amount ZPJ, abnormal order frequency of retail terminals YDH, regional sales structure balance index HZS, policy change response time ZXY, regional consumption deviation rate QFP, and product compliance risk score CFK, and preprocess and dimensionlessize the collected data;

[0022] The business adaptation unit is used to couple the collected multi-source data, input the processed multi-source data into a pre-trained deep learning framework, perform feature fusion through a multi-layer neural network, and then calculate and obtain a business optimization reference coefficient YWY. The specific calculation formula is as follows:

[0023]

[0024] In the formula: QPS is the regional distribution frequency, QDF is the channel coverage width, PSZ is the brand single-product launch cycle, ZPJ is the average terminal order amount, YDH is the abnormal order frequency of retail terminals, HZS is the regional sales structure balance index, ZXY is the policy change response time, QFP is the regional consumption deviation rate, and CFK is the product compliance risk score;

[0025] The comparison unit is used to perform data analysis on the calculated business optimization reference coefficient, and judge whether the current enterprise sales need optimization guidance and optimization plan according to the analysis results.

[0026] Preferably, the specific method for the comparison unit to judge whether the current enterprise sales need optimization guidance is as follows:

[0027] When YWY ≤ 0.55, it means that the current enterprise sales need to be optimized;

[0028] When TSX > 0.55, it means that the current enterprise sales do not need to be optimized.

[0029] Preferably, the specific method for the comparison unit to optimize the current enterprise sales plan is as follows:

[0030] When 0.44 < YWY ≤ 0.55, it means that the current enterprise sales need to be optimized at the first level, increase the distribution frequency by 15%, ensure that the channel inventory meets the demand fluctuations, increase the channel layout in the uncovered areas by 10% to improve the sales penetration rate, speed up the launch speed of new products or promotional products by 20%, and improve the market response speed;

[0031] When 0.33 < YWY ≤ 0.44, it indicates that the current enterprise sales need secondary optimization. Analyze whether there is an overly concentrated sales structure or a deviated consumption trend based on the regional sales structure balance index HZS and the regional consumption deviation rate QFP. Adjust the product structure through the regional sales structure balance index HZS to balance the proportion of high-end and low-end products within the region and avoid over-concentration of sales on a certain type of product. Analyze the purchase preferences of consumers through the regional consumption deviation rate QFP and adjust the sales strategy in a timely manner, such as increasing the promotion intensity of products with higher consumption tendencies.

[0032] When YWY ≤ 0.33, it indicates that the current enterprise sales need tertiary optimization. Check whether the product compliance risk score CFK is abnormally high and handle possible compliance risks in a timely manner. Check whether YDH is too high, indicating a very poor match between terminal inventory and demand, and immediate adjustment is required. Increase the regional coverage QDF by 30% by adding new channels or optimizing the existing channel layout to enhance sales penetration. Improve product quality management and compliance inspections for the product compliance risk score CFK to eliminate potential risks. Temporarily adjust the price strategy to quickly respond to market changes and avoid risks brought by large price fluctuations. Completely reconstruct the policy response timeliness ZXY to ensure that policy changes can be communicated to the terminal and executed within the fastest time.

[0033] Preferably, the iterative module includes a data monitoring unit, an update scheduling unit, and a version switching unit;

[0034] The data monitoring unit is used to collect multi-source data in real-time enterprise and sales processes, including the real-time outbound processing duration CKS, the daily order congestion rate YDL, the abnormal transfer request number YDB, the terminal active order reporting frequency BPD, the out-of-stock item ratio DHL, the same-store daily sales fluctuation index BDZ, the product price movement frequency YDZ, the consumption preference deviation trend value QHZ, and the terminal feedback signal density FKZ, and perform preprocessing and dimensionless processing on the collected multi-source data;

[0035] The update scheduling unit is used to couple the processed multi-source data, input the processed multi-source data into a pre-trained deep learning framework, perform feature fusion through a multi-layer neural network, and then calculate and obtain the model debugging coefficient MXT. The specific calculation formula is as follows:

[0036]

[0037] In the formula: real-time outbound processing duration CKS, daily order congestion rate YDL, abnormal transfer request number YDB, terminal active order reporting frequency BPD, out-of-stock item ratio DHL, same-store daily sales fluctuation index BDZ, product price movement frequency YDZ, consumption preference deviation trend value QHZ, and terminal feedback signal density FKZ;

[0038] The version switching unit is used to integrate and analyze the model debugging coefficient obtained by calculation and the debugging calibration coefficient, and judge whether the model needs to be adjusted and the adjustment time according to the analysis result.

[0039] Preferably, the analysis method for the version switching unit to adjust the model is as follows:

[0040] When MXT ≤ (TSX - MXT) × 0.65, it means that the current model needs to be optimized;

[0041] When TSX > (TSX - MXT) × 0.65, it means that the current model does not need to be optimized.

[0042] Preferably, the analysis method for the version switching unit to analyze the model adjustment time is as follows:

[0043] When (TSX - MXT) × 0.45 < YWY ≤ (TSX - MXT) × 0.65, it means that the current model has a first-level error and needs to be optimized within two weeks;

[0044] When (TSX - MXT) × 0.35 < YWY ≤ (TSX - MXT) × 0.45, it means that the current model has a second-level error and needs to be optimized within one week;

[0045] When YWY ≤ (TSX - MXT) × 0.35, it means that the current model has a third-level error and needs to be optimized within three working days.

[0046] Compared with the prior art, the beneficial effects of the present invention are:

[0047] 1. Compared with the marketing management means implemented by traditional tobacco enterprises relying on manual experience and static data analysis, the marketing plan auxiliary optimization system based on big data of the present invention打通了数据采集、模型构建、策略评估、持续迭代及结果反馈的全流程闭环机制。该系统不仅实现了营销数据的自动化处理与建模,还具备策略优化与动态学习能力,显著提升了企业对市场变化的响应速度与策略执行的准确性。特别是在面对政策频繁调整、消费偏好转移、渠道结构复杂等多变情境下,本系统能够有效识别潜在风险,提出精确干预方案,增强销售系统的鲁棒性与稳定性。

[0048] 2. The present invention构建了面向实际运营场景的营销绩效数字化评估机制,不仅提高了销售策略调整的科学性与系统性,也为企业实现营销管理的数字化转型与智能优化提供了技术支撑。 It should be noted that there are some Chinese words or phrases in the original text that seem to be incorrect or incomplete in the English translation part (such as "打通了数据采集、模型构建、策略评估、持续迭代及结果反馈的全流程闭环机制" and "构建了面向实际运营场景的营销绩效数字化评估机制" in the English translation), which may need to be further corrected and refined according to the accurate Chinese content. If you can provide the correct and complete Chinese text, a more accurate English translation can be generated.

[0049] 3. The design of the iteration module significantly optimizes real-time data collection, dynamic model correction, and automated version management. Highly intelligent and business-integrated, it not only provides timely and accurate model tuning solutions as the business environment changes, but also continuously maintains the model's adaptability and optimal state in real-world scenarios through automated iteration. This technological innovation provides enterprises with efficient, stable, and scalable technical support in a rapidly changing sales environment, effectively enhancing their overall capabilities in sales forecasting, allocation decision-making, and exception handling. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is a schematic diagram of the overall three-dimensional structure of the present invention.

[0051] In the figure: 1. Training module; 11. Feature collection unit; 12. Debugging unit; 13. Triggering unit; 2. Evaluation module; 21. Information collection unit; 22. Business adaptation unit; 23. Comparison unit; 3. Iteration module; 31. Data monitoring unit; 32. Update scheduling unit; 33. Version switching unit; 4. Feedback module. DETAILED DESCRIPTION

[0052] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0053] Example 1: Please refer to Figure 1 , a tobacco enterprise marketing plan auxiliary optimization system based on big data, including: a training module 1, an evaluation module 2, an iteration module 3 and a feedback module 4;

[0054] Training module 1 is used to collect and preprocess multi-source data, couple the preprocessed data to generate a basic model, and perform calibration analysis on the basic model to determine its practicality;

[0055] Evaluation module 2 is used to collect multi-source data from the enterprise and sales process, pre-process it, and input the pre-processed data into the model for coupling and optimization;

[0056] Iteration module 3 is used to collect multi-source data from real-time enterprises and sales processes, and pre-process the data. The pre-processed data is input into the model for coupling and optimization.

[0057] The feedback module 4 is used to transmit the collected data and model optimization results to the visualization terminal.

[0058] In this embodiment: Training module 1 is used to collect and preprocess multi-source data related to tobacco companies, including but not limited to historical sales data, channel operation data, consumer behavior data, and market environment data. After the data is cleaned, denoised, and normalized according to unified standards, data coupling and feature engineering are further performed to build an initial model. This module also includes a model calibration function, which is used to evaluate the predictive ability and practical value of the basic model through methods such as cross-validation and residual analysis, thereby screening out a model structure that can be adapted to the company's actual operating scenarios. The implementation of this module helps to improve the scientific nature and adaptability of model construction, and provides data support and algorithmic foundation for the system's subsequent strategy optimization and decision support.

[0059] Evaluation Module 2 is used to conduct real-time strategy evaluation based on multi-source dynamic data collected during the enterprise's marketing process. The collected data includes product data on sale, channel coverage, delivery frequency, policy implementation, and terminal order information. After pre-processing, it is input into the basic model constructed by Training Module 1 for calculation, and the effectiveness and matching degree of the current sales strategy are inferred from the model output results. This module can also quantify the contribution of each key marketing parameter to the overall sales performance, thereby identifying strategic shortcomings and assisting enterprises to adjust key strategic dimensions such as brand single product launch cycle PSZ, regional delivery frequency QPS or channel width QDF, to realize a model-driven strategy evaluation system. Through the setting of this module, quantitative feedback and scientific optimization of the execution effect of sales strategy can be achieved.

[0060] Iterative Module 3 aims to continuously optimize the model by dynamically collecting and processing real-time enterprise operational data. This module continuously receives real-time data from enterprises' sales processes, such as feedback on policy adjustments, shifts in consumer trends, and unusual ordering behavior, and feeds this data into the existing model for parameter updates and path corrections. This module integrates online learning and incremental modeling mechanisms, enabling rapid adaptation to market disruptions. Through continuous model iteration, the weighting of key indicators such as the regional consumption deviation rate (QFP) and the number of unusual retail ordering times (YDH) can be dynamically adjusted, enabling the system to maintain predictive accuracy and responsiveness in complex and volatile market environments. The implementation of this module effectively enhances the system's self-learning capabilities and the timeliness of marketing strategies.

[0061] Feedback Module 4 integrates the analysis results, model optimization status, and related data indicators from each module and outputs them to a visualization terminal. Through visual charts, data panels, and dynamic curves, it displays the changing trends and optimization suggestions for key indicators such as the Regional Sales Structure Balance Index (HZS), Policy Change Response Time (ZXY), and Product Compliance Risk Score (CFK), helping managers quickly understand the current operating status of the sales system and potential risk points. Furthermore, this module synchronizes optimization results in real time to the company's marketing execution system and terminal devices, forming a closed-loop process from strategy formulation to execution feedback. By establishing a visual feedback mechanism, the company's strategic transparency and execution efficiency are improved, driving the transformation of its marketing model towards a more refined and intelligent one.

[0062] Compared to traditional tobacco companies' marketing management methods that rely on manual experience and static data analysis, the present invention's big data-based marketing program auxiliary optimization system has opened up a closed-loop mechanism for the entire process of data collection, model building, strategy evaluation, continuous iteration, and result feedback. This system not only realizes the automated processing and modeling of marketing data, but also has the ability to optimize strategies and dynamically learn, significantly improving the company's response speed to market changes and the accuracy of strategy execution. Especially in the face of changing situations such as frequent policy adjustments, shifting consumer preferences, and complex channel structures, this system can effectively identify potential risks, propose precise intervention plans, and enhance the robustness and stability of the sales system.

[0063] Example 2: Please refer to Figure 1 , the training module 1 includes a feature acquisition unit 11, a debugging unit 12 and a triggering unit 13;

[0064] The feature collection unit 11 is used to collect multi-source data, including short-term purchase frequency DQM, single sales volume DCX, sales interval FMJ, daily sales RXL, monthly inventory turnover rate YKC, number of display items CGP, brand sales share change rate PSB, distribution coverage rate PHL, and cumulative number of product types LJP, and pre-process and dimensionlessly convert the collected multi-source data;

[0065] The debugging unit 12 is used to couple the collected multi-source data, input the processed multi-source data into the pre-trained deep learning framework, perform feature fusion through a multi-layer neural network, and then calculate the debugging calibration coefficient TSX. The specific calculation formula is as follows:

[0066]

[0067] Where: DQM is the number of short-term purchases, DCX is the single sales volume, FMJ is the sales interval, RXL is the daily sales volume, YKC is the monthly inventory turnover rate, CGP is the number of display items, PSB is the change rate of brand sales share, PHL is the distribution coverage rate, and LJP is the cumulative number of product types;

[0068] The trigger unit 13 is used to perform data analysis on the debugging calibration coefficients and determine whether the current model is practical based on the analysis results.

[0069] The specific analysis method of the trigger unit 13 on the debugging calibration coefficient is as follows:

[0070] When TSX≤0.75, it means that the current model cannot be used directly and the weight value needs to be adjusted and the model needs to be recalibrated;

[0071] When TSX>0.75, it means that the current model can be used directly.

[0072] In this embodiment, feature acquisition unit 11 automatically collects business data across multiple key dimensions, including sales, inventory, display, and distribution. This includes data on short-term purchasing behavior (DQM), single-sale performance (DCX), inventory turnover (YKC), brand sales volatility (PSB), and product richness (LJP). This comprehensive approach to collecting multi-dimensional, multi-source data significantly enhances the model's ability to perceive actual business scenarios and avoids the accuracy deviations associated with previous single-dimensional modeling. Furthermore, by preprocessing and dimensionlessizing the data, the integration and comparability of metrics across different dimensions are enhanced, providing a unified data foundation for subsequent modeling.

[0073] Debugging unit 12 uses a well-designed coupling calculation formula to quantitatively model the interactions between various business indicators, enabling initial behavior-driven model training. This formula mathematically couples factors such as the relationship between short-term purchase frequency and sales volume, the relationship between sales intervals, the linkage between daily sales and inventory turnover, and the combined performance of brand sales changes and product distribution coverage. This generates a debugging calibration coefficient (TSX) representing the model's reliability and adaptability. This approach effectively overcomes the inadequate identification of interactions between variables in traditional static training methods.

[0074] The trigger unit 13 automatically analyzes and makes decisions on the debugging calibration coefficient TSX output by the debugging unit 12 by setting a clear numerical judgment mechanism, further improving the objectivity and response efficiency of the model practicality judgment. When the TSX value is less than or equal to 0.75TSX≤0.75, the system automatically identifies that there is a deviation in the prediction structure of the current model, indicating that it does not yet have the ability to be directly invested in actual marketing decisions. At this time, the trigger unit 13 automatically sends an adjustment signal to guide the model to reset weights and recalibrate parameters to ensure that the model meets the established accuracy requirements before use; and when the TSX value is greater than 0.75TSX>0.75, the system determines that the model is already usable and can be directly invested in the subsequent evaluation and optimization module for strategy recommendation and simulation analysis. By introducing the TSX threshold judgment mechanism in the trigger unit 13, not only the intelligence level of the model training process is improved, but also the dynamic adaptability and deployment efficiency of the marketing assistance optimization system in different sales scenarios are significantly enhanced, laying a solid foundation for the subsequent online optimization and real-time decision-making of the model.

[0075] Training Module 1, by constructing a training process that includes multi-source acquisition, coupled calculation, and feedback control, not only significantly improves the model's stability and business matching, but also enhances the system's training adaptability under complex market conditions, thereby providing a more effective intelligent model foundation for the entire marketing support optimization system.

[0076] Example 3: Please refer to Figure 1 , the evaluation module 2 includes an information collection unit 21, a service adaptation unit 22 and a comparison unit 23;

[0077] The information collection unit 21 is used to collect multi-source data in the enterprise and sales process, including regional delivery frequency QPS, channel coverage width QDF, brand single product launch cycle PSZ, average terminal order amount ZPJ, retail terminal abnormal order number YDH, regional sales structure balance index HZS, policy change response time ZXY, regional consumption deviation rate QFP and product compliance risk score CFK, and pre-process and dimensionlessly convert the collected data;

[0078] The service adaptation unit 22 is used to couple the collected multi-source data, input the processed multi-source data into the pre-trained deep learning framework, perform feature fusion through a multi-layer neural network, and then calculate the service optimization reference coefficient YWY. The specific calculation formula is as follows:

[0079]

[0080] Where: QPS is the regional distribution frequency, QDF is the channel coverage width, PSZ is the brand single product launch cycle, ZPJ is the average amount of terminal orders, YDH is the number of abnormal orders of retail terminals, HZS is the regional sales structure balance index, ZXY is the response time limit of policy changes, QFP is the regional consumption deviation rate, and CFK is the product compliance risk score;

[0081] The comparison unit 23 is used to perform data analysis on the calculated business optimization reference coefficient, and based on the analysis results, determine whether the current enterprise sales need optimization guidance and optimization solutions.

[0082] The specific method for the comparison unit 23 to determine whether the current enterprise sales need optimization guidance is as follows:

[0083] When YWY ≤ 0.55, it means that the current enterprise sales need to be optimized;

[0084] When TSX > 0.55, it means that the current enterprise sales do not need to be optimized.

[0085] The specific method for the comparison unit 23 to determine the optimization solution for the current enterprise sales is as follows:

[0086] When 0.44 < YWY ≤ 0.55, it means that the current enterprise sales need to be optimized at the first level, increase the distribution frequency by 15%, ensure that the channel inventory meets the demand fluctuations, increase the channel layout in the uncovered areas by 10% to improve the sales penetration rate, speed up the new product or promotional product launch speed by 20%, and improve the market response speed;

[0087] When 0.33 < YWY ≤ 0.44, it means that the current enterprise sales need to be optimized at the second level. Analyze whether there is an overly concentrated sales structure or a deviated consumption trend based on the regional sales structure balance index HZS and the regional consumption deviation rate QFP. Adjust the product structure through the regional sales structure balance index HZS to balance the ratio of high-end and low-end products within the region and avoid over-concentration of sales on a certain type of product. Analyze the purchase preferences of consumers through the regional consumption deviation rate QFP and adjust the sales strategy in a timely manner, such as increasing the promotion intensity of products with higher consumption tendencies;

[0088] When YWY≤0.33, it means that the current enterprise sales need to enter the third level of optimization, check whether the product compliance risk score CFK is abnormally high, deal with possible compliance risks in a timely manner, check whether YDH is too high, indicating that the terminal inventory and demand are extremely poorly matched and need to be adjusted immediately, increase the regional coverage QDF by 30%, enhance sales penetration by adding new channels or optimizing the existing channel layout, improve product quality management and compliance inspection based on the product compliance risk score CFK, eliminate potential risks, temporarily adjust the pricing strategy, respond quickly to market changes, avoid the risks brought by large price fluctuations, and completely reconstruct the policy response time ZXY to ensure that policy changes can be transmitted to the terminal and implemented in the fastest time.

[0089] In this embodiment, the information collection unit 21 supports structured collection of key business parameters in the enterprise's sales process, including but not limited to regional delivery frequency (QPS), channel coverage width (QDF), brand single product launch cycle (PSZ), average terminal order amount (ZPJ), abnormal order number (YDH), regional sales structure balance index (HZS), policy change response time (ZXY), regional consumption deviation rate (QFP), and product compliance risk score (CFK). After preprocessing and dimensionless conversion, the collected data ensures unified processing of different types of indicators in the model, improving the stability of data fusion and model adaptability.

[0090] Based on this data, the business adaptation unit 22 constructs a composite coupling calculation model for business performance, generating a business optimization reference coefficient YWY. This coefficient comprehensively reflects the performance of multiple core dimensions, including delivery efficiency, channel breadth, product launch efficiency, terminal activity, structural balance, policy execution capabilities, and product compliance risks. Through formulaic quantitative modeling, marketing analysis, which previously relied on empirical judgment, is structured and standardized, enhancing the system's ability to accurately diagnose a company's marketing status.

[0091] Comparison unit 23 performs a judgment analysis on the calculated YWY coefficient. When the YWY value is less than or equal to 0.55 (YWY≤0.55), the system automatically determines that the company's current sales performance has a certain degree of structural deviation and needs to initiate a corresponding level of optimization guidance process to improve its matching efficiency in terms of channel layout, product launch rhythm, terminal order health, etc.; when the YWY value is higher than 0.55 (YWY>0.55), it indicates that the company's current sales strategy is generally performing well. The model does not need to trigger the optimization process and can maintain the current configuration and continue to run. Based on its interval distribution results, it automatically determines the optimization level of the company's current sales status and outputs a decision signal on whether to initiate optimization guidance and recommend specific solutions. Compared with traditional static reconciliation or lagged report analysis methods, this mechanism has greater real-time and forward-looking capabilities and can effectively identify sales bottlenecks and business risks in advance.

[0092] Comparative Unit 23 not only quantitatively assesses a company's sales status through the business optimization reference coefficient YWY, but also further establishes corresponding tiered optimization strategies based on different YWY value ranges, building a quantifiable, executable, and closed-loop intelligent sales optimization mechanism. This significantly enhances the company's dynamic response capabilities and refined management capabilities in complex and volatile market environments. When 0.44 < YWY ≤ 0.55, the system determines that the company's current sales status is in the first-level optimization stage, requiring light intervention. It proposes recommendations such as increasing regional delivery frequency (QPS) by 15%, expanding channel width (QDF) in uncovered areas by 10%, and accelerating the launch cycle (PSZ) of brand single products by 20%. These recommendations ensure that channel inventory matches market fluctuations. By shortening the product launch cycle and increasing reach density, the company's rapid response capabilities and sales penetration at the end of the channel are enhanced. This stage's optimization plan is lightweight, fast, and efficient, allowing for strategic fine-tuning without impacting the existing operational rhythm.

[0093] The present invention constructs a digital evaluation mechanism for marketing performance for actual operation scenarios by setting up a complete logical process of information collection, coupling analysis and judgment feedback in the evaluation module 2. It not only improves the scientificity and systematicity of sales strategy adjustment, but also provides technical support for enterprises to achieve digital transformation and intelligent optimization of marketing management.

[0094] Example 4: Please refer to Figure 1 , the iteration module 3 includes a data monitoring unit 31, an update scheduling unit 32 and a version switching unit 33;

[0095] The data monitoring unit 31 is used to collect multi-source data in real-time enterprise and sales processes, including real-time outbound processing time CKS, daily order congestion rate YDL, number of abnormal transfer requests YDB, terminal active order frequency BPD, out-of-stock product ratio DHL, same-store daily sales fluctuation index BDZ, product price fluctuation number YDZ, consumer preference anomaly trend value QHZ, and terminal feedback signal density FKZ, and pre-process and dimensionless the collected multi-source data;

[0096] The update scheduling unit 32 is used to couple the processed multi-source data, input the processed multi-source data into the pre-trained deep learning framework, perform feature fusion through the multi-layer neural network, and then calculate the model debugging coefficient MXT. The specific calculation formula is as follows:

[0097]

[0098] Where: CKS is the real-time outbound processing duration, YDL is the same-day order congestion rate, YDB is the number of abnormal transfer requests, BPD is the terminal active reporting frequency, DHL is the ratio of out-of-stock items, BDZ is the same-store daily sales fluctuation index, YDZ is the number of product price fluctuations, QHZ is the consumption preference deviation trend value, and FKZ is the terminal feedback signal density;

[0099] The version switching unit 33 is used to integrate and analyze the calculated model debugging coefficient and the debugging calibration coefficient, and based on the analysis result, determine whether the model needs to be adjusted and the adjustment time.

[0100] The analysis method of the version switching unit 33 for adjusting the model is as follows:

[0101] When MXT ≤ (TSX - MXT) × 0.65, it means that the current model needs to be optimized;

[0102] When TSX > (TSX - MXT) × 0.65, it means that the current model does not need to be optimized.

[0103] The analysis method of the version switching unit 33 for the model adjustment time is as follows: [[ID=!7]]

[0104] When (TSX - MXT) × 0.45 < YWY ≤ (TSX - MXT) × 0.65, it means that the current model has a first-level error and needs to be optimized within two weeks;

[0105] When (TSX - MXT) × 0.35 < YWY ≤ (TSX - MXT) × 0.45, it means that the current model has a second-level error and needs to be optimized within one week; <!

[0106] When YWY ≤ (TSX - MXT) × 0.35, it means that the current model has a third-level error and needs to be optimized within three working days.

[0107] In this embodiment: The data monitoring unit 31 constructs a real-time monitoring system based on multi-dimensional business operation indicators, covering operation efficiency and business stability indicators including real-time outbound processing duration CKS, same-day order congestion rate YDL, number of abnormal transfer requests YDB, terminal active reporting frequency BPD, etc., as well as important data dimensions reflecting market dynamics and terminal feedback such as the ratio of out-of-stock items DHL, number of price fluctuations YDZ, and consumption preference deviation trend value QHZ. Through preprocessing and dimensionless operation of the above multi-source data, the system ensures the comparability and integration of indicators with different dimensions, and realizes a comprehensive perception of the enterprise operation status and a high-precision modeling input basis.

[0108] The update scheduling unit 32 constructs a comprehensive model debugging coefficient (MXT) by coupling data from multiple sources. This calculation formula rationally groups indicators by business attributes, enabling the system to not only identify single abnormal indicators but also comprehensively assess the cumulative effects of multiple abnormal signals. This mechanism significantly improves the model's responsiveness to sudden issues such as logistics congestion, price fluctuations, and sudden shifts in consumer trends. It enables enterprises to quickly identify and track the root causes of business fluctuations at the model level, enhancing the system's proactive perception and debugging capabilities.

[0109] Version switching unit 33 integrates and analyzes the current model debugging coefficients MXT with the original debugging calibration coefficients TSX to dynamically determine whether the model exhibits performance degradation, poor adaptation, or delayed response. If the analysis result exceeds a set threshold, the model adjustment and version switching mechanism is automatically triggered. The system selectively retrieves and deploys the new model version and schedules the optimal adjustment time for smooth switching, avoiding business interruption. This mechanism overcomes the limitations of traditional models that rely on manual intervention or periodic manual updates, truly realizing closed-loop model iteration capabilities that are adaptive to business scenarios.

[0110] When the model debugging coefficient MXT≤(TSX-MXT)×0.65, it means that the performance output of the current model has obviously deviated from its original debugging and calibration state, and it is necessary to judge it as a model failure trend and trigger the optimization and adjustment process. When the debugging and calibration coefficient TSX>(TSX-MXT)×0.65, it means that the model is still within a reasonable performance range and no optimization operation is required for the time being.

[0111] This judgment method introduces a dynamic relative deviation judgment rule, which is based on the difference between the current state of the model and the initial debugging benchmark. It further defines the optimization threshold by setting a proportional coefficient of 0.65, thereby achieving a more flexible and adaptive judgment mechanism.

[0112] When (TSX - MXT) × 0.45 < YWY ≤ (TSX - MXT) × 0.65, it indicates that the current model deviation is in the first - level error range, and the model optimization needs to be completed within two weeks to prevent the performance from deteriorating further. When (TSX - MXT) × 0.35 < YWY ≤ (TSX - MXT) × 0.45, it means that the current model has entered the second - level error range, and the optimization process needs to be completed within one week to ensure the model stability. When YWY ≤ (TSX - MXT) × 0.35, it shows that the model is in the third - level error state, with significant performance anomalies, and the model needs to be optimized or reconstructed immediately within three working days to prevent serious impact on the business judgment results. This solution constructs a multi - level error judgment logic based on the dynamic relationship among the debugging calibration coefficient TSX, the model debugging coefficient MXT, and the business optimization reference coefficient YWY, achieving an accurate mapping and closed - loop control between the degree of model performance deviation and the optimization time, improving the timeliness and controllability of model operation and maintenance. Compared with the fixed - threshold strategy, this method calculates the error - level boundary through dynamic benchmarks, more sensitively reflecting the adaptability and deviation degree of the model during actual operation, making the optimization scheduling have scenario awareness and self - adaptability. By setting corresponding optimization time limits for different levels of errors, it not only avoids resource waste caused by frequent triggering of adjustments due to small - scale deviations but also can quickly respond to major performance anomalies, balancing the relationship between system stability and business continuity and improving the overall operation efficiency. This error - time - effect linkage mechanism cooperates with the data monitoring, debugging calculation, and version switching unit 33 in the iterative module 3 to construct a complete dynamic management system for the full life cycle of the model, enabling a closed - loop process of "deviation warning → hierarchical judgment → time - limited optimization → automatic update", significantly enhancing the intelligence and automation level of the system.

[0113] The design of the iterative module 3 has achieved significant optimizations in aspects of real - time data acquisition, dynamic model correction, and automated version management, with high intelligence and business coupling capabilities. It can not only provide timely and accurate model optimization solutions when the business environment changes but also continuously maintain the adaptability and optimal state of the model in the actual scenario through an automatic iteration mechanism. This technological innovation provides efficient, stable, and evolvable technical support for enterprises in a rapidly changing sales environment, effectively improving the overall capabilities of sales forecasting, allocation decision - making, and exception handling.

[0114] Content not described in detail in this specification belongs to the prior art well - known to those skilled in the art.

[0115] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A tobacco enterprise marketing plan auxiliary optimization system based on big data, characterized by: include: Training module (1), evaluation module (2), iteration module (3) and feedback module (4); The training module (1) is used to collect and preprocess multi-source data, and couple the preprocessed data to generate a basic model, and perform calibration analysis on the basic model to determine its practicality; The evaluation module (2) is used to collect multi-source data from the enterprise and sales process, and pre-process the data, and input the pre-processed data into the model for coupling and optimization; The iterative module (3) is used to collect multi-source data from real-time enterprises and sales processes, and pre-process the data, input the pre-processed data into the model for coupling, and optimize the model; The feedback module (4) is used to transmit the collected data and model optimization results to the visualization terminal.

2. The tobacco enterprise marketing plan auxiliary optimization system based on big data according to claim 1, characterized in that: The training module (1) includes a feature acquisition unit (11), a debugging unit (12) and a triggering unit (13); The feature collection unit (11) is used to collect multi-source data, including short-term purchase frequency DQM, single sales volume DCX, sales interval FMJ, daily sales RXL, monthly inventory turnover rate YKC, display product specifications CGP, brand sales share change rate PSB, distribution coverage rate PHL and cumulative product type number LJP, and pre-process and dimensionlessly convert the collected multi-source data; The debugging unit (12) is used to couple the collected multi-source data, input the processed multi-source data into a pre-trained deep learning framework, perform feature fusion through a multi-layer neural network, and then calculate and obtain the debugging calibration coefficient TSX; The trigger unit (13) is used to perform data analysis on the debugging calibration coefficient and judge whether the current model is practical based on the analysis result.

3. The tobacco enterprise marketing plan auxiliary optimization system based on big data according to claim 2 is characterized by: The specific analysis method of the trigger unit (13) on the debugging calibration coefficient is as follows: When TSX≤0.75, it means that the current model cannot be used directly and the weight value needs to be adjusted and the model needs to be recalibrated; When TSX>0.75, it means that the current model can be used directly.

4. The tobacco enterprise marketing plan auxiliary optimization system based on big data according to claim 3 is characterized by: The evaluation module (2) includes an information collection unit (21), a service adaptation unit (22) and a comparison unit (23); The information collection unit (21) is used to collect multi-source data in the enterprise and sales process, including regional delivery frequency QPS, channel coverage width QDF, brand single product launch cycle PSZ, terminal order average amount ZPJ, retail terminal abnormal order number YDH, regional sales structure balance index HZS, policy change response time ZXY, regional consumption deviation rate QFP and product compliance risk score CFK, and pre-process and dimensionless the collected data; The service adaptation unit (22) is used to couple the collected multi-source data, input the processed multi-source data into a pre-trained deep learning framework, perform feature fusion through a multi-layer neural network, and then calculate and obtain a service optimization reference coefficient YWY; The comparison unit (23) is used to perform data analysis on the business optimization reference coefficient obtained by calculation, and based on the analysis results, determine whether the current enterprise sales need optimization guidance and optimization solutions.

5. The tobacco enterprise marketing plan auxiliary optimization system based on big data according to claim 4 is characterized by: The specific method for the comparison unit (23) to determine whether the current enterprise sales need optimization guidance is as follows: When YWY ≤ 0.55, it means that the current enterprise sales need to be optimized; When TSX > 0.55, it means that the current enterprise sales do not need to be optimized.

6. The tobacco enterprise marketing plan auxiliary optimization system based on big data according to claim 5 is characterized by: The specific method for the comparison unit (23) to determine the optimization solution for the current enterprise sales is as follows: When 0.44 < YWY ≤ 0.55, it means that the current enterprise sales need primary optimization, increase the distribution frequency by 15%, ensure that the channel inventory meets the demand fluctuations, increase the channel layout in the uncovered areas by 10% to improve the sales penetration rate, speed up the launch speed of new products or promotional products by 20%, and improve the market response speed; When 0.33 < YWY ≤ 0.44, it means that the current enterprise sales need secondary optimization. Analyze whether there is an overly concentrated sales structure or a deviated consumption trend based on the regional sales structure balance index HZS and the regional consumption deviation rate QFP. Adjust the product structure through the regional sales structure balance index HZS to balance the proportion of high-end and low-end products within the region and avoid over-concentration of sales on a certain type of product. Analyze the purchase preferences of consumers through the regional consumption deviation rate QFP and timely adjust the sales strategy, such as increasing the promotion intensity of products with higher consumption tendencies; When YWY ≤ 0.33, it means that the current enterprise sales need tertiary optimization. Check whether the product compliance risk score CFK is abnormally high and promptly handle possible compliance risks. Check whether YDH is too high, indicating a very poor match between the terminal inventory and demand, and immediate adjustment is required. Increase the regional coverage QDF by 30% by adding new channels or optimizing the existing channel layout to enhance the sales penetration. Improve product quality management and compliance inspections for the product compliance risk score CFK to eliminate potential risks. Temporarily adjust the price strategy to quickly respond to market changes and avoid risks brought by large price fluctuations. Completely reconstruct the policy response timeliness ZXY to ensure that policy changes can be communicated to the terminal and executed in the shortest possible time.

7. The tobacco enterprise marketing plan auxiliary optimization system based on big data according to claim 6 is characterized by: The iteration module (3) includes a data monitoring unit (31), an update scheduling unit (32), and a version switching unit (33); The data monitoring unit (31) is used to collect multi-source data in the real-time enterprise and sales process, including the real-time outbound processing duration CKS, the daily order congestion rate YDL, the abnormal transfer request number YDB, the terminal active order reporting frequency BPD, the out-of-stock item ratio DHL, the same-store daily sales fluctuation index BDZ, the product price movement frequency YDZ, the consumption preference deviation trend value QHZ, and the terminal feedback signal density FKZ, and perform preprocessing and dimensionless processing on the collected multi-source data; The update scheduling unit (32) is used to couple the processed multi-source data, input the processed multi-source data into a pre-trained deep learning framework, perform feature fusion through a multi-layer neural network, and then calculate and obtain the model debugging coefficient MXT; The version switching unit (33) is used to integrate and analyze the calculated model debugging coefficient and the debugging calibration coefficient, and judge whether the model needs to be adjusted and the adjustment time according to the analysis results.

8. The tobacco enterprise marketing plan auxiliary optimization system based on big data according to claim 7 is characterized by: The analysis method for the version switching unit (33) to adjust the model is specifically as follows: When MXT ≤ (TSX - MXT) × 0.65, it means that the current model needs to be optimized; When TSX > (TSX - MXT) × 0.65, it means that the current model does not need to be optimized.

9. The tobacco enterprise marketing plan auxiliary optimization system based on big data according to claim 8, characterized in that: The analysis method for the version switching unit (33) to analyze the model adjustment time is as follows: When (TSX - MXT) × 0.45 < YWY ≤ (TSX - MXT) × 0.65, it means that the current model has a first-level error and needs to be optimized within two weeks; When (TSX - MXT) × 0.35 < YWY ≤ (TSX - MXT) × 0.45, it means that the current model has a second-level error and needs to be optimized within one week; When YWY ≤ (TSX - MXT) × 0.35, it means that the current model has a third-level error and needs to be optimized within three working days.

Images