Classification management method, system and equipment for tobacco retailers and medium

Abstract

The invention provides a tobacco retailer classification management method, system and device and a medium, and belongs to the field of tobacco retail management, and the method comprises the steps: obtaining the multi-source data of tobacco retailers; extracting a feature relationship among different multi-source data, and constructing an association rule base for identifying violation risk types of tobacco retailers according to the feature relationship; based on a sliding window algorithm, according to the multi-source data, constructing an improved Boston matrix with a relative sales contribution degree as a horizontal axis and a cigarette sales growth rate as a longitudinal axis, and distributing each tobacco retailer to a corresponding quadrant to obtain an initial classification result; calculating a credit correction factor according to the initial classification result based on the risk type identified by the association rule base; calibrating the initial classification result according to the credit correction factor to obtain a final classification result; and executing a corresponding supervision strategy according to the quadrant position corresponding to the final classification result. According to the method, the accuracy, the real-time performance and the supervision efficiency of tobacco retailer classification are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of tobacco retail management, specifically to a method, system, equipment, and medium for classifying and managing tobacco retailers. Background Technology

[0002] Under the tobacco monopoly management system, tobacco retailers, as the final link in the industry chain, have their compliance and development potential directly impacting market order and industry efficiency. Traditional tobacco retailer management relies heavily on experience-based judgment or single-dimensional data assessment, making it difficult to address the challenges of integrating multi-source, heterogeneous data, and hindering a comprehensive evaluation of sales performance, market potential, and creditworthiness. With the deepening of digital transformation in the tobacco industry, how to construct a scientific, dynamic, and accurate classification and evaluation system for tobacco retailers to optimize the allocation of regulatory resources and effectively deliver precise services has become a critical issue urgently needing to be addressed by the industry.

[0003] To address these challenges, existing technologies have attempted to incorporate various analytical tools for classifying and managing tobacco retailers. Among these, the Boston Consulting Group (BCG) matrix, a classic strategic analysis tool, has been adopted in some solutions for evaluating tobacco retailers. Traditional methods typically apply this matrix directly, using market growth rate and relative market share as two core indicators to categorize tobacco retailers, thus providing a preliminary reference for resource allocation.

[0004] However, these existing technologies based on the traditional Boston Consulting Group (BCG) matrix have revealed significant shortcomings in practical applications within the tobacco industry. First, their dimensionality is limited, failing to incorporate creditworthiness, a core element of monopoly regulation, resulting in a severe disconnect between assessment results and the compliance-first regulatory orientation. Second, their assessment thresholds are mostly statically set, unable to adapt to the strong seasonal and regional fluctuations in cigarette sales, leading to rigid and outdated classification results. Third, existing technologies lack effective linkage mechanisms with monopoly regulation and logistics systems, making it difficult to directly apply classification results to specific regulatory business scenarios such as supply allocation and inspection frequency, ultimately resulting in low management efficiency and insufficient utilization of regulatory resources. Summary of the Invention

[0005] In view of this, it is necessary to provide a classification management method, system, equipment and medium for tobacco retailers to solve the technical problems of the disconnect between assessment results and regulatory needs and the inaccurate allocation of resources in the existing technology.

[0006] To address the aforementioned technical problems, in a first aspect, the present invention provides a method for classifying and managing tobacco retailers, comprising: Obtain multi-source data on tobacco retailers; Extract the feature relationships between different multi-source data, and construct an association rule base for identifying the violation risk types of the tobacco retailers based on the feature relationships; Based on the sliding window algorithm, an improved Boston Matrix is ​​constructed using the multi-source data, with relative sales contribution as the horizontal axis and cigarette sales growth rate as the vertical axis. Each tobacco retailer is assigned to a quadrant corresponding to the improved Boston Matrix to obtain an initial classification result. The improved Boston Matrix includes four quadrants. Based on the risk types identified by the association rule base, a credit correction factor is calculated according to the initial classification results; The initial classification result is calibrated based on the credit correction factor to obtain the final classification result; The corresponding regulatory strategy will be executed based on the quadrant position corresponding to the final classification result.

[0007] In one possible implementation, prior to acquiring the multi-source data of tobacco retailers, the process includes: Based on the data structure type, the multi-source data is divided into structured data, semi-structured data, and unstructured data; Preprocessing is performed on different types of data; the preprocessing includes outlier handling, data standardization, and format unification.

[0008] In one possible implementation, the sliding window algorithm constructs an improved Boston Matrix based on the multi-source data, with relative sales contribution as the horizontal axis and cigarette sales growth rate as the vertical axis. Each tobacco retailer is assigned to a quadrant corresponding to the improved Boston Matrix to obtain an initial classification result, including: The relative sales contribution and cigarette sales growth rate of each tobacco retailer are calculated based on the multi-source data. The sliding window algorithm is used to calculate the average regional sales contribution and median regional growth rate for the current period. Establish the improved Boston Matrix corresponding to the current period; the improved Boston Matrix is ​​a two-dimensional coordinate system in which the mean of the regional sales contribution in the current period is the horizontal axis threshold and the median of the regional growth rate is the vertical axis threshold; Based on the multi-source data of the current period, the tobacco retailers are classified into the corresponding quadrants of the two-dimensional coordinate system of the current period, and the initial classification result is determined as the quadrant position of the tobacco retailers.

[0009] In one possible implementation, the step of calculating the credit correction factor based on the risk types identified by the association rule base and the initial classification results includes: Based on the multi-source data of the current retailer in the current period, obtain the type of violation risk identified by the association rule base; Based on the pre-established quadrant credit scoring relationship, a basic credit score is determined according to the initial classification results; the quadrant credit scoring relationship includes the correspondence between the quadrant and the basic credit score; The risk-weighted score is obtained by weighting and adjusting the basic credit score based on the aforementioned types of violation risks. The credit correction factor is calculated based on the risk-weighted score and the multi-source data for the current period.

[0010] In one possible implementation, the multi-source data includes the number of violations, credit repair duration, and complaint rate. The calculation of the credit correction factor based on the risk-weighted score and the multi-source data for the current period includes: The penalty amount is calculated based on the number of violations, the credit repair time, and the complaint rate. The credit correction factor is calculated based on the risk-weighted score and the penalty amount.

[0011] In one possible implementation, calibrating the initial classification result according to the credit correction factor to obtain the final classification result includes: When the credit correction factor is less than the first threshold, the current quadrant of the tobacco retailer is adjusted in the direction of increasing credit supervision level, and the adjusted quadrant position is used as the final classification result. When the credit correction factor is greater than or equal to the second threshold, the current quadrant of the tobacco retailer is adjusted in the direction of decreasing credit supervision level, and the adjusted quadrant position is used as the final classification result; the second threshold is greater than the first threshold.

[0012] In one possible implementation, executing the corresponding regulatory strategy based on the quadrant position corresponding to the final classification result includes: If the quadrant position is the first quadrant, the first strategy is executed. If the quadrant position is the second quadrant, the second strategy is triggered. If the quadrant position is the third quadrant, the third strategy is triggered. If the quadrant position is the fourth quadrant, the fourth strategy is triggered. Among them, the relative sales contribution of the first to fourth quadrants increases sequentially, the supply allocation level of the first to fourth strategies decreases sequentially, and the credit supervision level of the first to fourth strategies increases sequentially.

[0013] Secondly, the present invention also provides a classification management system for tobacco retailers, comprising: The processing module is used to acquire multi-source data from tobacco retailers; A construction module is used to extract feature relationships between different multi-source data, and to construct an association rule base for identifying the violation risk types of the tobacco retailers based on the feature relationships; The matching module is used to construct an improved Boston Matrix based on the multi-source data using a sliding window algorithm, with relative sales contribution as the horizontal axis and cigarette sales growth rate as the vertical axis, and to assign each tobacco retailer to the quadrant corresponding to the improved Boston Matrix to obtain an initial classification result; the improved Boston Matrix includes four quadrants; The calculation module is used to calculate the credit correction factor based on the risk types identified by the association rule base and the initial classification results. An adjustment module is used to calibrate the initial classification result according to the credit correction factor to obtain the final classification result; The execution module is used to execute the corresponding regulatory strategy based on the quadrant position corresponding to the final classification result.

[0014] Thirdly, the present invention also provides an electronic device, including a memory and a processor, wherein, The memory is used to store programs; The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps in the tobacco retailer classification management method described in any of the above implementations.

[0015] Fourthly, the present invention also provides a computer-readable storage medium for storing a computer-readable program or instruction, which, when executed by a processor, can implement the steps in the tobacco retailer classification management method described in any of the above implementations.

[0016] The beneficial effects of this invention are as follows: The classification and management method for tobacco retailers provided by this invention first integrates multi-dimensional data such as sales and supervision, breaking down information silos and constructing profile data of tobacco retailers to avoid bias in single-dimensional judgments. Furthermore, it mines association rules for hidden violation risk types (such as the combination of "nighttime + school periphery + abnormal inventory") using algorithms such as Apriori and creates an association rule library, improving the foresight and accuracy of risk warnings. Furthermore, it achieves real-time updates and adaptive adjustments to classification results through sliding window algorithms and credit correction mechanisms. Furthermore, the entire process is data- and rule-driven, reducing human intervention, avoiding selective enforcement, and promoting fair and transparent supervision. Through a closed-loop process of "data fusion—rule mining—dynamic classification—factor correction—strategy execution," an intelligent, refined, and sustainable risk classification and supervision system for tobacco retailers is constructed, which not only significantly improves the ability to identify and prevent violations but also has broad application and promotion value. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A schematic flowchart of an embodiment of the tobacco retailer classification management method provided by the present invention; Figure 2 A schematic flowchart of another embodiment of the tobacco retailer classification management method provided by the present invention; Figure 3 For the present invention Figure 1 A schematic diagram of an embodiment of S103; Figure 4 For the present invention Figure 1 A schematic diagram of an embodiment of S104; Figure 5 For the present invention Figure 4 A schematic diagram of an embodiment of S404; Figure 6 A schematic diagram of an embodiment of the tobacco retailer classification management system provided by the present invention; Figure 7 A schematic diagram of an embodiment of the electronic device provided by the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0020] In the description of the embodiments of the present invention, unless otherwise stated, "multiple" means two or more. "And / or" describes the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone.

[0021] The terms "first," "second," etc., used in the embodiments of this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a technical feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature.

[0022] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0023] Before demonstrating the embodiments, the following terms will be explained.

[0024] The 3σ principle is a rule of thumb in statistics for identifying outliers. In a normal distribution, the probability of data falling within the mean (μ) ± 3 standard deviations (σ) is approximately 99.7%. Therefore, data points outside the range of μ ± 3σ are considered outliers with a probability of only about 0.3%.

[0025] This invention provides a method, system, equipment, and medium for the classification and management of tobacco retailers, which will be described below.

[0026] Figure 1 A schematic flowchart of an embodiment of the tobacco retailer classification management method provided by the present invention is shown below. Figure 1 As shown, the classification management methods for tobacco retailers include: S101. Obtain multi-source data of tobacco retailers.

[0027] It should be noted that multi-source data for tobacco retailers is obtained periodically and automatically collected through API interfaces. These systems include the tobacco marketing system, the monopoly supervision system, the logistics tracking system, and the store management system. Data such as retailer purchase amounts, purchased product specifications, and sales rates can be obtained from the tobacco marketing system; violation records (including the number of violations and types), case details, and credit repair records (including credit repair duration and complaint rate) can be obtained from the monopoly supervision system; delivery addresses, delivery times, delivery offset distances, and route information can be obtained from the logistics tracking system; and real-time inventory, real-time sales volume, sales revenue, customer complaint data, inventory turnover days, and cigarette sales growth rate can be obtained from the store management system. This data obtained from the above systems will be considered as multi-source data.

[0028] S102. Extract the feature relationships between different multi-source data, and construct an association rule base for identifying the violation risk types of the tobacco retailers based on the feature relationships.

[0029] It should be noted that risk features are extracted from standardized data, including: sales anomalies (e.g., sales volume > twice the regional average for 3 consecutive weeks), logistics anomalies (e.g., monthly address deviation ≥ 3 times), and case characteristics (e.g., specific violation types and frequency in the past year). The Apriori algorithm or FP-Growth algorithm is used to mine frequent itemsets and association rules from historical data, setting minimum support, confidence, and lift. An example association rule is: "Abnormal sales fluctuations and serial number violations in the past year → Automatic elimination for not sourcing locally." Association rules that meet other thresholds can also be added, thus building an association rule library that includes association rules for various violation risk types. The association rule library is continuously optimized to adapt to new types of violations, providing quantitative risk basis for credit assessment. It automatically identifies hidden violation patterns, improving risk detection capabilities.

[0030] For example, if there is "high frequency of nighttime transactions," "abnormally low inventory turnover," and "a school within 500 meters," then there is a "risk of selling cigarettes to minors" (confidence level: 78%). Another example is if there are "more than 3 past penalties," "a short-term surge in sales," and "not on holidays," then there is a "risk of illegal procurement" (confidence level: 82%). These association rules are categorized by risk type (e.g., selling cigarettes to minors, illegal procurement, unlicensed wholesale, hoarding, etc.) to form a structured association rule library for violation risk types.

[0031] S103. Based on the sliding window algorithm, construct an improved Boston matrix with relative sales contribution as the horizontal axis and cigarette sales growth rate as the vertical axis according to the multi-source data, and assign each tobacco retailer to the quadrant corresponding to the improved Boston matrix to obtain the initial classification result; the improved Boston matrix includes four quadrants.

[0032] It should be noted that the traditional Boston Consulting Group (BCG) matrix, based on market growth rate and relative market share, is not suitable for retail risk assessment. Therefore, an improved BCG matrix is ​​proposed, with two dimensions: a horizontal axis representing a credit risk index (calculated by combining historical violations, complaints, and unusual transactions), and a vertical axis representing an operational activity index (calculated by combining sales volume, order volume, and business hours). A sliding window algorithm is used to dynamically calculate the credit risk index and operational activity index, setting a time window (e.g., the most recent 90 days), and updating the improved BCG matrix weekly. For each tobacco retailer, its two-dimensional score within the current window is calculated. Using the industry mean or quantile within the current window as boundaries, the two-dimensional space is divided into four quadrants: Quadrant 1 (corresponding to high-risk, high-activity key regulatory targets), Quadrant 2 (corresponding to high-risk, low-activity potential risk households), Quadrant 3 (corresponding to low-risk, high-activity high-quality compliant households), and Quadrant 4 (corresponding to low-risk, low-activity stable marginal households). Of course, the risk types corresponding to the above four quadrants are just examples; they can also be set in reverse order: Quadrant 4 corresponds to high-risk, high-activity key regulatory targets, Quadrant 3 to high-risk, low-activity potential risk households, Quadrant 2 to low-risk, high-activity high-quality compliant households, and Quadrant 1 to low-risk, low-activity stable marginal households. Based on the multi-source data of each tobacco retailer falling within the corresponding quadrant, this tobacco retailer is mapped or assigned to that quadrant as the initial classification result.

[0033] S104. Based on the risk types identified by the association rule base, the credit correction factor is calculated according to the initial classification results.

[0034] It should be noted that: using the established association rule base, the initial classification results are matched for risk types to identify any implicit risks not covered by the matrix dimensions. For each tobacco retailer, the features of their multi-source data are input into the association rule base for forward inference to obtain the risk type, and the credit correction factor is calculated based on the risk type and the initial classification results.

[0035] S105. The initial classification result is calibrated according to the credit correction factor to obtain the final classification result.

[0036] It should be noted that the credit correction factor is applied to the initial classification results to dynamically adjust the position of the tobacco retailer in the quadrant to obtain the final classification result. The final classification result includes the retailer ID, the quadrant to which it belongs, the credit correction factor, etc.

[0037] S106. Execute the corresponding regulatory strategy based on the quadrant position corresponding to the final classification result.

[0038] It should be noted that, based on the final classification results, differentiated and precise regulatory strategies are formulated and implemented for tobacco retailers in different quadrants. The implementation of these regulatory strategies is achieved through automatic task delivery from the regulatory platform to the mobile devices of frontline inspectors, thus realizing closed-loop management.

[0039] In summary, the tobacco retailer classification management method provided in this invention first integrates multi-dimensional data such as sales and supervision to break down information silos and construct profile data of tobacco retailers, avoiding biases from single-dimensional judgments. Furthermore, it uses algorithms such as Apriori to mine association rules for hidden violation risk types (such as the combination of "nighttime + school periphery + abnormal inventory") and creates an association rule library, improving the foresight and accuracy of risk warnings. Furthermore, it achieves real-time updates and adaptive adjustments to classification results through a sliding window algorithm and a credit correction mechanism. Furthermore, the entire process is data- and rule-driven, reducing human intervention, avoiding selective enforcement, and promoting fair and transparent supervision. Through a closed-loop process of "data fusion—rule mining—dynamic classification—factor correction—strategy execution," an intelligent, refined, and sustainable tobacco retailer risk classification and supervision system is constructed, significantly improving the ability to identify and prevent violations and possessing broad application and promotion value.

[0040] The multi-source data related to tobacco retailers is diverse in origin and format. To improve processing efficiency and accuracy, in some embodiments of the present invention, such as... Figure 2 As shown, before step S101, the following steps are included: S201. Based on the data structure type, the multi-source data is divided into structured data, semi-structured data, and unstructured data.

[0041] It should be noted that: Structured data is obtained by identifying data tables with well-organized fields and fixed formats. Structured data generally includes basic information data (such as retailer ID, name, business address, and license information), sales transaction data (such as order number, cigarette code, sales volume, sales amount, and order time), logistics and distribution data (such as delivery tracking number, delivery address, receipt status, and delivery timestamp), and case handling data (such as case number, investigating unit, case cause, and handling result). Semi-structured data is obtained by parsing and extracting data with partial structural features. Semi-structured data generally includes JSON-formatted device log data (such as POS machine transaction logs and barcode scanner records), XML-formatted report data (such as monthly operating reports and audit report templates), and log file data (such as system operation logs and access records). Unstructured data is then processed to mine data without fixed formats. Unstructured data generally includes image data (such as license photos, store photos, and photos of evidence of violations), text data (such as customer complaint content, case description text, and business remarks), and geographic location data (such as latitude and longitude coordinates and delivery trajectory data). Before data collection, the system maintains a data source registry, recording data format examples for each source (e.g., JSON, CSV, XML, images, text, etc.). An automatic sniffing mechanism (e.g., file extensions, MIME types, field regular expressions) is used to determine the data structure type. The identified data is stored in a temporary buffer and tagged with a type label (e.g., struct_type: 1 / 2 / 3) to provide input identifiers for subsequent preprocessing modules. Python's pandas.read_() function is used to attempt to read and determine if the data is structured. For JSON / XML, the json / xml.etree library is used to parse and determine if it is semi-structured. Text / images are identified as unstructured by the file header (e.g., the magic number for PNG / JPG). A data type classifier (based on rules or a lightweight model) is built to achieve automated classification.

[0042] S202. Preprocessing is performed on different types of data; the preprocessing includes outlier handling, data standardization, and format unification.

[0043] It's important to note that typical anomalies in structured data include missing values ​​(e.g., sales revenue is NULL), outliers (e.g., negative daily sales or sales exceeding the limit), inconsistent units (e.g., mixing "yuan" and "ten thousand yuan"), and diverse time formats (e.g., "2025 / 11 / 21" vs. "2025-11-21 00:00:00"). Therefore, outlier detection and processing are necessary for structured data. This involves using the 3σ criterion or IQR (interquartile range) to identify numerical anomalies. Outliers can be deleted (suitable for obvious errors, such as negative inventory), replaced with the median / mean (suitable for slight deviations), or marked as missing and imputed (using KNN or regression interpolation). Then, the anomaly-processed structured data undergoes data standardization. This involves Z-score standardization of numerical fields and one-hot encoding of categorical fields (e.g., "retail format": convenience stores, supermarkets, tobacco and liquor stores). The time data in the anomaly-handled structured data is uniformly converted to ISO8601 format (YYYY-MM-DDHH:MM:SS), and derived features (such as weekday and holiday identifiers) are extracted. Then, the standardized structured data undergoes format unification processing, including standardizing the currency unit to "yuan," removing spaces from strings, converting full-width characters to half-width characters, and standardizing license numbers to uppercase with length validation (e.g., 15-digit number + letter combination). Thus, the cleaned structured data table, resulting from the above anomaly handling, data standardization, and format unification processes, can be directly imported into the database.

[0044] Typical anomalies in semi-structured data include inconsistent field names (e.g., "shop_name" vs. "storeName"), deep nesting levels (e.g., "location.province.city" in JSON), and implicit data types (e.g., numbers stored as strings). Therefore, semi-structured data needs to be parsed and flattened. This involves using `json_normalize()` to convert nested JSON into a wide table, and using `xmltodict` to parse XML and extract key path nodes. A standard field dictionary is established (e.g., unifying "retailer_name" as "retailer_name"), and fuzzy matching (e.g., Levenshtein distance) is used for automatic mapping and type conversion (e.g., converting "123" to the integer 123). Outlier handling is performed on the semi-structured data: records that fail to parse are written to an error log for manual review, and duplicate key-value pairs (e.g., multiple records for one store) are deduplicated (keeping the latest record by ID + timestamp) to output a flattened structured table.

[0045] Typical anomalies of unstructured data include typos, colloquialisms, and irrelevant information in text, and blurry, misaligned, or missing metadata in images, making the data unsuitable for direct analysis. Therefore, unstructured data processing is necessary. For text data (such as complaints and public opinion), HTML tags, emoticons, and URLs are removed for data cleaning. Jieba word segmentation combined with a custom tobacco dictionary (e.g., "Zhonghua cigarettes," "fake cigarettes") is used for word segmentation and error correction. A BERT-BiLSTM-CRF model is used to extract entities (e.g., "store name," "brand," "time") for entity recognition. The intensity of negative sentiment is assessed, and a "complaint sentiment score" (0-1) is generated for sentiment analysis. Finally, structured fields (e.g., {retailer_id, complaint_text, sentiment_score, key_entities}) are output. For image data (such as store photos), OpenCV is used to calculate sharpness (Laplacian variance), and blurry images (variance < 100) are removed for quality inspection. YOLOv8 is used to identify whether "No Selling Cigarettes to Minors" signs are posted for object detection. PaddleOCR is used to extract the sign text, match the license name for consistency, and then output the structured field {has_warning_sign:True / False, sign_text:"..."}. In this way, unstructured data is transformed into computable structured features.

[0046] After all types of data have been preprocessed, a global format unification is performed, with all data uniformly encoded in UTF-8, time fields uniformly encoded in UTC+8 time zone, and primary keys uniformly encoded in retailer license number (18-digit national standard format). The output is a standard Parquet or CSV file, supporting subsequent ETL loading.

[0047] In this embodiment, a dedicated cleaning strategy is employed for data with different structures to effectively eliminate noise, anomalies, and inconsistencies, eliminate differences in multi-source data formats, improve data quality, provide accurate input for subsequent analysis, reduce classification bias caused by data inconsistencies, and improve assessment reliability. Furthermore, by standardizing fields, unifying time, and unifying coding, the semantic gap between multi-source data is bridged, ensuring data consistency and fusionability, and achieving seamless integration across systems and platforms. Before acquiring multi-source data from tobacco retailers, a structure-aware preprocessing mechanism transforms the raw, messy data into high-quality, computable, and fusion-compatible standardized data. This provides clean, consistent, and high-dimensional input for subsequent core analysis steps such as building an association rule base and improving the Boston Matrix, and significantly improves the robustness and accuracy of the entire tobacco retailer risk identification system.

[0048] This invention provides horizontal and vertical axis indicators for constructing an improved Boston Consulting Group (BCG) matrix, replacing the traditional BCG matrix's "market share" and "market growth rate," making it more aligned with the regulatory context of the tobacco industry. In some embodiments of this invention, such as... Figure 3 As shown, step S103 includes: S301. Calculate the relative sales contribution and cigarette sales growth rate of each of the tobacco retailers based on the multi-source data.

[0049] It should be noted that the formula for calculating relative sales contribution is: (Purchase amount × α1 + Number of purchased product specifications × β1 + Sales rate × γ1) / Regional average × 100. Here, α1, β1, and γ1 are the respective weighting coefficients for purchase amount, number of purchased product specifications, and sales rate. The regional average is calculated based on retailer groups within the same administrative region and of similar business scale, ensuring comparability across different regions and retailer sizes.

[0050] The formula for calculating the cigarette sales growth rate is: (Current time window sales - Previous time window monthly sales) / Previous time window monthly sales × 100%. A 3-month time window can be set, which calculates the cigarette sales growth rate for the past 3 months by comparing the sales of the last 3 months with the sales of the previous 3 months. This avoids misjudgments caused by monthly fluctuations and reflects a continuous growth trend. Considering the cyclical impact of peak sales seasons such as Spring Festival and National Day, the year-on-year growth rate is used as a supplementary reference indicator.

[0051] S302. Use the sliding window algorithm to calculate the average regional sales contribution and median regional growth rate for the current period.

[0052] It should be noted that the window period is set to a preset duration (e.g., 1 month), and is automatically updated at the beginning of each window period (e.g., automatically updated on the 1st of each month). Real-time data collection is performed on the metrics of all valid retailers within the window period, excluding newly opened accounts, retailers under suspension or rectification, and other special circumstances. The arithmetic mean of the relative sales contribution of all retailers is calculated as the horizontal axis threshold (regional average sales contribution). This arithmetic mean of relative sales contribution reflects the overall sales contribution level of the region. The median of the cigarette sales growth rate of all retailers is calculated as the vertical axis threshold (regional median growth rate). The median cigarette sales growth rate helps to resist extreme values ​​and better reflects the central level of regional growth.

[0053] S303. Establish the improved Boston matrix corresponding to the current period; the improved Boston matrix is ​​a two-dimensional coordinate system in which the mean of the regional sales contribution in the current period is the horizontal axis threshold and the median of the regional growth rate is the vertical axis threshold.

[0054] It should be noted that the improved Boston Matrix for the current period is constructed by using the average regional sales contribution of the current period as the horizontal axis threshold and the median regional growth rate of the current period as the vertical axis threshold.

[0055] S304. Based on the multi-source data of the current period, classify the tobacco retailers into the corresponding quadrant of the two-dimensional coordinate system of the current period, and determine the initial classification result as the quadrant position of the tobacco retailers.

[0056] It should be noted that: Based on the quadrant mapping algorithm, each retailer is located by coordinate according to the multi-source data of the current period. The horizontal axis is the relative sales contribution of the retailer in the current period, and the vertical axis is the cigarette sales growth rate of the retailer in the current period. The relationship between the horizontal and vertical axes of the retailer and the horizontal (X-axis) threshold and vertical (Y-axis) threshold of the current period is compared to determine the quadrant to which the retailer belongs and thus obtain the initial classification result.

[0057] In this embodiment, a regionally adaptive improved Boston Matrix is ​​constructed to achieve a scientific, dynamic, and interpretable initial classification of tobacco retailers. A sliding window is introduced to dynamically update the horizontal and vertical thresholds corresponding to different window periods, improving classification fairness. Using the regional mean and median as boundaries eliminates biases caused by differences in economic levels and population density between different regions, avoiding unfairness and achieving fair competition within the region. Furthermore, the sliding window approach enables dynamic updates and trend awareness; the horizontal and vertical thresholds are updated continuously with historical data, capturing seasonal market fluctuations (such as the Spring Festival sales peak) and policy impacts (such as the implementation of tobacco control regulations), ensuring the classification results are up-to-date.

[0058] In some embodiments of the present invention, such as Figure 4 As shown, step S104 includes: S401. Based on the multi-source data of the current retailer in the current period, obtain the type of violation risk identified by the association rule base.

[0059] It should be noted that real-time feature vectors are extracted from multi-source data of current retailers. These feature vectors include sales characteristics (such as recent sales volatility, order frequency changes, and product concentration), logistics characteristics (such as the number of delivery anomalies, address offset distance, and regularity of delivery time), and operational characteristics (such as abnormal business hours, changes in payment methods, and frequency of customer complaints). The feature vectors are then compared with the association rule base for similarity calculation. The matching threshold is set to association rules with a confidence level ≥ 0.7. The successfully matched violation risk type and its corresponding confidence level are output. A risk level classification system is established, including high-risk types (violations involving core provisions of the Monopoly Law), medium-risk types (general operational violations), and low-risk types (minor operational irregularities). Taking into account rule confidence, support, and lift, the risk intensity is calculated as: Risk Intensity = Rule Confidence × Rule Lift.

[0060] S402. Based on the pre-established quadrant credit scoring relationship, determine the basic credit score according to the initial classification result; the quadrant credit scoring relationship includes the correspondence between the quadrant and the basic credit score.

[0061] It should be noted that the quadrant credit scoring relationship is established based on historical data statistical analysis. This involves collecting historical data (e.g., the actual credit performance of retailers in each quadrant over the past three years) and analyzing the correlation between different quadrants and credit scores to establish the quadrant credit scoring relationship. A baseline score of 90 points can be set for the first quadrant (excellent business performance, good credit expectations), 80 points for the second quadrant (stable contribution, growth needs improvement), 70 points for the third quadrant (potential but requires attention), and 60 points for the fourth quadrant (both business performance and credit need improvement). Considering regional differences, adjustment coefficients can be set according to different levels of economic development in different regions, incorporating urban-rural differences into the scoring calibration. Of course, considering the time factor, the baseline score for peak / off-peak seasons can be slightly adjusted.

[0062] S403. The basic credit score is weighted and adjusted based on the type of violation risk to obtain a risk-weighted score.

[0063] It should be noted that differentiated weights are assigned based on the type of violation. High-risk, medium-risk, and low-risk violation types each have different risk weight coefficients, and the risk level is negatively correlated with the risk weight coefficient; that is, the higher the risk level, the lower the risk weight coefficient. For example, the risk weight coefficient for high-risk violations is -0.15 to -0.20, for medium-risk violations it is -0.08 to -0.12, and for low-risk violations it is -0.03 to -0.05. The risk-weighted score is calculated by multiplying the risk weight coefficient corresponding to the violation risk type by the basic credit score.

[0064] S404. Based on the risk-weighted score and the multi-source data of the current period, the credit correction factor is calculated.

[0065] It should be noted that: based on the analysis results of multi-source data in the current period, combined with the risk-weighted score calculated by S403, the credit correction factor is calculated.

[0066] In this embodiment, each violation pattern that is matched by the associated rules will be reflected in the credit score in real time, enabling retailers to exercise self-discipline, reducing the motivation for intentional violations, and encouraging retailers to actively maintain a good credit record. By combining real-time operating data with the associated rule library, comprehensive risk scanning and multi-dimensional risk identification can be achieved, which can promptly detect new violation patterns and prevent potential credit risks. Furthermore, the credit correction factor, with its output range uniformly within (0, 1], can be directly used to determine quadrant upgrade / downgrade thresholds, exhibiting good compatibility and scalability. The credit correction factor can be directly mapped to factors such as supply volume, inspection frequency, and new product application eligibility, promoting a virtuous cycle in the industry's credit ecosystem. Furthermore, it enables the rational allocation of regulatory resources based on credit status, achieving refined management. Through the early warning function of the credit correction factor, it achieves pre-emptive risk prevention, with fully automated calculation reducing manual intervention and subjective judgment. Furthermore, through the complete chain of "rule identification of violation risk types → determination of basic credit scores → risk-weighted correction → calculation of credit correction factors," qualitative risks are transformed into quantitative scores, improving the objectivity and comparability of assessments. This provides a quantitative basis for differentiated regulatory measures and offers a unified, quantitative, and interpretable core basis for subsequent differentiated regulation, supply control, and credit repair, promoting precise regulation, digital governance, and credit supervision in the tobacco industry. It possesses high practical, innovative, and promotional value.

[0067] In some embodiments of the present invention, such as Figure 5 As shown, the multi-source data includes the number of violations, credit repair time, and complaint rate. Step S404 includes: S501. The penalty amount is calculated based on the number of violations, the credit repair time, and the complaint rate.

[0068] It should be noted that the cumulative number of violations within the past 12 months is extracted, and a logarithmic function is used to calculate log(number of violations + 1) to standardize the number of violations, avoiding excessive influence from extreme values. A maximum threshold for the number of violations is set; if a single month has more than 5 violations, it is counted as 5. The actual number of days from the occurrence of a violation to the completion of credit repair is calculated to obtain the credit repair time. The complaint rate = number of complaints / total number of transactions × 10,000 (per ten thousand). Considering differences in business scale, an adjustment coefficient is set for small-scale retailers, and a mechanism for identifying and filtering malicious complaints is established.

[0069] The penalty amount is calculated based on the number of violations, the credit repair time, and the complaint rate: Penalty=(V×W_v+T×W_t+R×W_r) / N; Where V represents the number of violations, T represents the credit repair time (in days), and R represents the complaint rate (percentage). W_v is the weight for the number of violations, W_t is the weight for the credit repair time, W_r is the weight for the complaint rate, and N is the standard threshold (e.g., 10, used to adjust the range of penalties). The standard threshold is determined based on the 85th percentile of historical data for the region and is adjusted quarterly according to the overall credit situation of the region. Temporary adjustment coefficients are set for special periods (such as during the pandemic).

[0070] S502. Calculate the credit correction factor based on the risk-weighted score and the penalty amount.

[0071] It should be noted that the risk-weighted score (0-100 points) is converted to a 0-1 range using the formula: Normalized Score = Risk-Weighted Score / 100. A minimum guarantee value is set to ensure a basic score even for higher-risk individuals. Non-linear mapping optimization can be performed: a convex function mapping is used for the high-score range (>80 points) to highlight good credit, while a concave function mapping is used for the low-score range (<60 points) to strengthen risk warnings. The credit correction factor is calculated based on the risk-weighted score and the penalty amount: Credit Correction Factor = Normalized Score - Penalty Amount.

[0072] In this embodiment, abstract credit risk is transformed into a concrete numerical credit correction factor. Classification management based on this credit correction factor improves the efficiency of regulatory resource utilization. The entire process, from data input to result output, is automated, reducing manual intervention and establishing a precise, fair, and efficient credit assessment system. This provides reliable technical support for the accurate classification and scientific supervision of tobacco retailers. Calculating the credit correction factor accurately reflects the credit status of retailers, providing a basis for differentiated supervision, promoting credit repair, and incentivizing retailers to proactively improve their credit behavior. This optimized resource allocation not only improves the accuracy and timeliness of credit assessment but also promotes self-improvement in retailers' credit behavior through a clear incentive mechanism, achieving a win-win situation for both regulatory effectiveness and retailer development.

[0073] In some embodiments of the present invention, step S105 includes: When the credit correction factor is less than the first threshold, the current quadrant of the tobacco retailer is adjusted in the direction of increasing credit supervision level, and the adjusted quadrant position is used as the final classification result. When the credit correction factor is greater than or equal to the second threshold, the current quadrant of the tobacco retailer is adjusted in the direction of decreasing credit supervision level, and the adjusted quadrant position is used as the final classification result; the second threshold is greater than the first threshold.

[0074] It should be noted that the first threshold (supervision strengthening threshold, e.g., 0.8) and the second threshold (incentive and support threshold, e.g., 1.2) are set. The first and second thresholds can be adjusted quarterly based on historical data of regional credit status. Considering the differences in economic development levels, different regions can set different adjustments for the first and second thresholds. The credit supervision level in the first quadrant of the improved Boston Matrix can be set as level 1 (the lowest supervision intensity), the second quadrant as level 2 (lower supervision intensity), the third quadrant as level 3 (medium supervision intensity), and the fourth quadrant as level 4 (the highest supervision intensity). The direction of increasing credit supervision level is: level 1 → level 2 → level 3 → level 4, and the direction of decreasing credit supervision level is: level 4 → level 3 → level 2 → level 1.

[0075] According to the credit correction factor and the initial classification result, the adjustment range of the final classification result is at most across one supervision level. For example, if the initial classification result shows that the current quadrant of a tobacco retailer is in the second quadrant, and the credit correction factor < 0.8, the classification of the tobacco retailer is adjusted to the third quadrant. If the credit correction factor ≥ 1.2, the classification of the tobacco retailer is adjusted to the first quadrant.

[0076] It is also possible to calculate the deviation degree and the advantage degree based on the credit correction factor. According to the deviation degree, it is determined that the adjustment range of correcting the initial classification result to the final classification result is at least one supervision level. For example, the deviation degree = (0.8 - credit correction factor) / 0.8. If the deviation degree < PLD1 (e.g., 25%), the supervision level is set to increase by one level. If the deviation degree is between [PLD1, PLD2) (e.g., between 25% - 50%), the supervision level is set to increase by one level. If the deviation degree > PLD2, the supervision level is set to increase by two levels. Perform quadrant adjustment and record the reasons for adjustment. When the credit correction factor ≥ 1.2, calculate the advantage degree = (credit correction factor - )

[0077] Generate the adjusted quadrant position as the final classification result, record the adjustment track: original quadrant → new quadrant, adjustment time, adjustment reason, update the classification status and expiration date in the retailer file, conduct a sample review of the adjustment result to ensure the rationality of the adjustment, set up an objection appeal channel, where retailers can raise objections to the adjustment result, regularly evaluate the effectiveness of the adjustment rules, and continuously optimize the algorithm parameters.

[0078] In this embodiment, by setting a first threshold and a second threshold, retailers with different credit statuses are accurately distinguished. The specific adjustment range is determined based on a credit correction factor, achieving refined supervision and improving the efficiency of regulatory resource utilization. Furthermore, when the credit correction factor is less than the first threshold, the current quadrant of the tobacco retailer is adjusted towards an increasing credit regulatory level; when the credit correction factor is greater than or equal to the second threshold, the current quadrant is adjusted towards a decreasing credit regulatory level. This clear promotion mechanism incentivizes retailers to proactively maintain and improve their credit status. Combined with strict downgrade rules, credit risks are promptly identified and isolated to prevent systemic risks. The entire set of adjustment rules is automated, reducing manual intervention, improving work efficiency, and ensuring traceability. This establishes a scientific, fair, and efficient credit calibration system, not only achieving a precise correlation between credit status and classification results but also promoting the healthy development of the credit ecosystem through a clear incentive mechanism.

[0079] In some embodiments of the present invention, step S106 includes: If the quadrant position is the first quadrant, the first strategy is executed. If the quadrant position is the second quadrant, the second strategy is triggered. If the quadrant position is the third quadrant, the third strategy is triggered. If the quadrant position is the fourth quadrant, the fourth strategy is triggered. Among them, the relative sales contribution of the first to fourth quadrants increases sequentially, the supply allocation level of the first to fourth strategies decreases sequentially, and the credit supervision level of the first to fourth strategies increases sequentially.

[0080] It should be noted that a precise mapping relationship is established between quadrant positions and strategy execution. Strategy 1 is an incentive-based strategy combination specifically for retailers in the first quadrant; Strategy 2 is a stability-based strategy combination for retailers in the second quadrant; Strategy 3 is a corrective strategy combination for retailers in the third quadrant; and Strategy 4 is a constraint-based strategy combination for retailers in the fourth quadrant. The supply allocation level can adopt a 10-level quantitative system, with level 1 being the highest priority. The credit supervision level can adopt a 5-level quantitative system, with level 5 being the most stringent supervision. A strategy parameter library is established to support flexible adjustments based on the characteristics of different regions.

[0081] In this embodiment, differentiated management strategies are implemented based on the characteristics of retailers in different quadrants. This improves the targeting of supervision and services, prioritizing the allocation of limited, scarce goods to retailers with high contributions and good credit, thereby enhancing overall operational efficiency. Through strategy differentiation, retailers are incentivized to improve their operational level and creditworthiness to obtain better goods and more lenient supervision. Furthermore, stronger supervision is strengthened for retailers with poor credit and operational performance to effectively prevent the risk of violations. Moreover, high-quality resources are concentrated on high-performing retailers, while regulatory resources are focused on high-risk retailers, maximizing resource utilization efficiency. Precise control over retailers with different risk levels effectively prevents systemic risks. Fair, transparent, and predictable management methods enhance retailer acceptance and improve overall satisfaction. Furthermore, through a four-level strategy system, full-spectrum management from incentives to constraints is achieved. Resource allocation is precisely matched with the actual value of retailers, ensuring that management measures respond promptly to changes in retailers' status. This provides operable and quantifiable standards for refined management, optimizes the allocation of management resources and comprehensively improves regulatory efficiency. In addition, through a clear incentive and constraint mechanism, it promotes the overall improvement of the retailer group and the healthy development of the market, demonstrating significant business value and social benefits.

[0082] For example, based on the TRIZ "segmentation principle" and "pre-processing principle," multi-source heterogeneous data is classified and preprocessed. Specifically, this includes classifying multi-source retail customer credit supervision data according to data type and business attributes, clarifying the characteristics of each data source. For instance, structured data includes basic data (retailer ID, name, business address, license number, license validity period, contact information, longitude, latitude), sales data (order number, retailer ID, cigarette code, sales volume, sales amount, order time), logistics data (delivery order number, order number, delivery time, delivery address, recipient, receipt time, delivery status), and structured historical case data (case number, seizing unit, seizing department, retailer license number, parties involved, time of incident, cause of action, nature of case). This type of data has a standardized format and can be directly processed at the field level. Differentiated cleaning strategies are adopted to address quality issues of different data types. Outlier Handling: A dual mechanism of "3σ principle + business rule verification" is adopted. For example, in sales data, if a retailer's sales volume of a certain type of cigarettes exceeds three times the average sales volume of similar retailers in the region (i.e., exceeding the μ+3σ range), it is initially identified as an outlier. Further analysis, combined with historical case data, shows that if the retailer has a record of "not purchasing goods from the local tobacco monopoly wholesale enterprise" within the past year, it is marked as a "violation-related outlier."

[0083] Data standardization: unify data formats and units, such as converting delivery time from the format "2024-05-20 09:30" to the ISO8601 standard format "2024-05-20T09:30:00+08:00", unifying the sales unit to "yuan" (if the original data is "ten thousand yuan", multiply by 10000 to convert), and unifying the delivery address to the format "province-city-district-detailed address" (such as "Hubei Province-Huangshi City-Xisai District-Yanhu Road No. 123").

[0084] Then, combining the principles of integration, versatility, and pre-emptive reaction in TRIZ theory, a multi-dimensional data fusion scheme is designed to fully explore the correlations and complementarities between data. TRIZ principles emphasize combining related objects or functions to improve system efficiency. By mining the correlations between multi-source data, a rule base for identifying violations and risks is constructed. Specifically, this includes filtering potentially correlated features from preprocessed multi-source data, such as: Sales data characteristics: abnormal sales fluctuations (sales exceeding the regional average by more than 2 times for 3 consecutive weeks), abnormal order frequency (daily order count exceeding the historical average by more than 3 times). Logistics data characteristics: delivery address does not match the registered address (signature offset exceeds 50 meters, etc.), number of offsets (monthly offset ≥ 2 times, 3 times, 4 times), and route signatures (signature interval within 10 seconds, 20 seconds, 30 seconds). Characteristics of historical case data: Number of violations in the past year (≥1 time), type of violation (not purchasing from local tobacco monopoly wholesale enterprises / selling illegally produced tobacco monopoly products / operating tobacco product retail business without a tobacco monopoly retail license); After identifying the aforementioned features, an association rule mining algorithm is used, employing "support-confidence-lift" as the selection criteria to mine association rules between features. The minimum support can be set to 0.04 (meaning the rule appears at least 4% of the historical data), the minimum confidence to 0.8 (meaning the probability of the consequent occurring when the antecedent occurs is at least 80%), and the minimum lift to 1.2 (meaning the rule's association is higher than random association). For example, using historical data from Huangshi Tobacco in December 2024 as the training set, the algorithm generates initial association rules, such as "abnormal sales fluctuations ∧ violations of serial number sales in the past year → not purchasing from local tobacco monopoly wholesale enterprises." The support (0.06), confidence (0.85), and lift (1.3) are calculated, all meeting the threshold requirements, and are included in the rule base. Rules with a confidence below 0.8 are removed (e.g., "receipt offset ∧ number of offsets → operating a tobacco product retail business without a tobacco monopoly retail license," with a confidence of 0.65). The system includes a quarterly evaluation mechanism for rule effectiveness. If the confidence level of a certain association rule in new data is below 0.7 for two consecutive quarters, it will be deleted from the association rule library. If a newly discovered association rule meets the threshold requirements, it will be added to the association rule library. The final association rule library contains core rules covering major violation types such as not purchasing goods from local tobacco monopoly wholesale enterprises, operating a tobacco product retail business without a tobacco monopoly retail license, selling tobacco to minors, and mailing or carrying tobacco leaves or tobacco products in excess of the limit.

[0085] An improved Boston Consulting Group (BCG) matrix is ​​constructed based on the "relative sales contribution - cigarette sales growth rate" formula. The classification results are then adjusted using a credit correction factor, ultimately outputting four types of retailer profiles and regulatory strategies. The system architecture includes a data layer, processing layer, model layer, and application layer, achieving an end-to-end closed loop for classification and evaluation. As shown in Table 1 below, a multi-dimensional indicator system is constructed:

[0086] Table 1. Mapping Table of Multi-Dimensional Indicator System The improved Boston Matrix uses a two-dimensional quadrant comprising a horizontal axis and a vertical axis. The horizontal axis represents "relative sales contribution" (replacing the traditional "relative market share"), reflecting the retailer's actual contribution to regional sales. The vertical axis represents "cigarette sales growth rate" (replacing the traditional "market growth rate"), reflecting operational growth. A sliding window algorithm (window period = 1 month) is used to update the horizontal axis threshold (average regional sales contribution) and the vertical axis threshold (median regional growth rate) in real time, addressing the issue of seasonal fluctuations in cigarette sales. When the retailer's credit correction factor is <0.8, a quadrant adjustment is triggered; when the correction factor is >1.2, a quadrant adjustment is triggered, strengthening the orientation towards compliant operations. The classification results and regulatory strategies are shown in Table 2 below.

[0087] Table 2. Classification Results and Regulatory Strategies This application system integrates with the tobacco marketing system (sales data), the monopoly supervision system (credit / case data), the logistics tracking system (delivery data), and the store management system (real-time inventory data). It supports API interfaces and batch import, with a data update frequency of at least once per day. It employs a combination of "missing value imputation (mean filling) + outlier removal (3σ principle) + data standardization (Z-score)," calculating sales performance indicator weights using the entropy weight method to avoid subjective assignment bias. It incorporates an improved Boston Matrix algorithm, supports visualization (scatter plots + quadrant partitioning), and provides a threshold adjustment interface to adapt to different regional market characteristics (e.g., urban vs. rural retailers). It generates retailer classification reports (including credit scores and potential predictions) and triggers regulatory action pushes (e.g., sending "problem customer visit reminders" to account manager apps). Based on TRIZ theory, it constructs a unified retail customer credit supervision data model through a systematic data fusion method, significantly improving data quality and utilization value. It uses analytical tools such as the Boston Matrix to classify and evaluate retail customers. Based on factors such as sales performance, credit status, and market potential, customers are categorized into different groups, providing a foundation for precise customer profiling. This constructs a three-dimensional evaluation system integrating "sales performance, credit status, and market potential," addressing the problem of traditional single-dimensional classification. It creates an improved Boston Consulting Group (BCG) matrix adapted to the tobacco industry, establishes a dynamic threshold adjustment mechanism, enables real-time updates of retailer classifications, and connects classification results with the application link of tobacco monopoly supervision, providing precise basis for sourcing, inspection focus, and credit repair. Credit status is used as a correction factor in the BCG matrix to address the "compliance-first" regulatory requirement of tobacco monopoly, filling industry gaps. The sliding window threshold mechanism adapts to the seasonal fluctuations in cigarette sales, increasing the update frequency of classification results from quarterly to monthly, improving response speed. Classification results are provided for application reference in sourcing and inspection frequency, improving the utilization rate of regulatory resources.

[0088] To better implement the tobacco retailer classification management method in this invention embodiment, based on the tobacco retailer classification management method, correspondingly, as follows: Figure 6 As shown, this embodiment of the invention also provides a classification management system 600 for tobacco retailers, which includes: Processing module 601 is used to acquire multi-source data of tobacco retailers; The construction module 602 is used to extract the feature relationships between different multi-source data and construct an association rule base for identifying the violation risk types of the tobacco retailers based on the feature relationships; The matching module 603 is used to construct an improved Boston Matrix based on the multi-source data using a sliding window algorithm, with relative sales contribution as the horizontal axis and cigarette sales growth rate as the vertical axis, and to assign each tobacco retailer to the quadrant corresponding to the improved Boston Matrix to obtain an initial classification result; the improved Boston Matrix includes four quadrants. Calculation module 604 is used to calculate a credit correction factor based on the risk types identified by the association rule base and the initial classification results. The adjustment module 605 is used to calibrate the initial classification result according to the credit correction factor to obtain the final classification result; The execution module 606 is used to execute the corresponding regulatory strategy based on the quadrant position corresponding to the final classification result.

[0089] The tobacco retailer classification management system 600 provided in the above embodiments can realize the technical solutions described in the above tobacco retailer classification management method embodiments. The specific implementation principles of each module or unit can be found in the corresponding content in the above tobacco retailer classification management method embodiments, and will not be repeated here.

[0090] like Figure 7 As shown, the present invention also provides an electronic device 700. The electronic device 700 includes a processor 701, a memory 702, and a display 703. Figure 7 Only some components of the electronic device 700 are shown, but it should be understood that it is not required to implement all the components shown, and more or fewer components may be implemented instead.

[0091] In some embodiments, processor 701 may be a central processing unit (CPU), microprocessor, or other data processing chip, used to run program code stored in memory 702 or process data, such as the tobacco retailer classification management method of the present invention.

[0092] In some embodiments, processor 701 may be a single server or a group of servers. The server group may be centralized or distributed. In some embodiments, processor 701 may be local or remote. In some embodiments, processor 701 may be implemented on a cloud platform. In one embodiment, the cloud platform may include a private cloud, public cloud, hybrid cloud, community cloud, distributed cloud, intranet, multi-cloud, etc., or any combination thereof.

[0093] In some embodiments, memory 702 may be an internal storage unit of electronic device 700, such as a hard disk or memory of electronic device 700. In other embodiments, memory 702 may also be an external storage device of electronic device 700, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. equipped on electronic device 700.

[0094] Furthermore, the memory 702 may include both internal storage units of the electronic device 700 and external storage devices. The memory 702 is used to store application software and various types of data installed on the electronic device 700.

[0095] In some embodiments, display 703 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. Display 703 is used to display information from electronic device 700 and to display a visual user interface. Components 701-703 of electronic device 700 communicate with each other via a system bus.

[0096] In one embodiment, when processor 701 executes the tobacco retailer classification management program in memory 702, the following steps can be implemented: Obtain multi-source data on tobacco retailers; Extract the feature relationships between different multi-source data, and construct an association rule base for identifying the violation risk types of the tobacco retailers based on the feature relationships; Based on the sliding window algorithm, an improved Boston Matrix is ​​constructed using the multi-source data, with relative sales contribution as the horizontal axis and cigarette sales growth rate as the vertical axis. Each tobacco retailer is assigned to a quadrant corresponding to the improved Boston Matrix to obtain an initial classification result. The improved Boston Matrix includes four quadrants. Based on the risk types identified by the association rule base, a credit correction factor is calculated according to the initial classification results; The initial classification result is calibrated based on the credit correction factor to obtain the final classification result; The corresponding regulatory strategy will be executed based on the quadrant position corresponding to the final classification result.

[0097] It should be understood that when the processor 701 executes the tobacco retailer classification management program in the memory 702, in addition to the functions mentioned above, it can also perform other functions, as detailed in the description of the corresponding method embodiments above.

[0098] Furthermore, the embodiments of the present invention do not specifically limit the type of electronic device 700 mentioned. Electronic device 700 can be a mobile phone, tablet computer, personal digital assistant (PDA), wearable device, laptop computer, or other portable electronic device. Exemplary embodiments of portable electronic devices include, but are not limited to, portable electronic devices running iOS, Android, Microsoft, or other operating systems. The aforementioned portable electronic device can also be other portable electronic devices, such as a laptop computer with a touch-sensitive surface (e.g., a touch panel). It should also be understood that in some other embodiments of the present invention, electronic device 700 may not be a portable electronic device, but rather a desktop computer with a touch-sensitive surface (e.g., a touch panel).

[0099] Accordingly, this application also provides a computer-readable storage medium for storing computer-readable programs or instructions. When the programs or instructions are executed by a processor, they can implement the steps or functions of the tobacco retailer classification management method provided in the above-described method embodiments.

[0100] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.), and the computer program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0101] The above provides a detailed description of the classification management method, apparatus, equipment, and medium for tobacco retailers provided by this invention. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, those skilled in the art will recognize that there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A method for classifying and managing tobacco retailers, characterized in that, include: Obtain multi-source data on tobacco retailers; Extract the feature relationships between different multi-source data, and construct an association rule base for identifying the violation risk types of the tobacco retailers based on the feature relationships; Based on the sliding window algorithm, an improved Boston Matrix is ​​constructed using the multi-source data, with relative sales contribution as the horizontal axis and cigarette sales growth rate as the vertical axis. Each tobacco retailer is assigned to the quadrant corresponding to the improved Boston Matrix to obtain the initial classification result. The improved Boston matrix comprises four quadrants; Based on the risk types identified by the association rule base, a credit correction factor is calculated according to the initial classification results; The initial classification result is calibrated based on the credit correction factor to obtain the final classification result; The corresponding regulatory strategy will be executed based on the quadrant position corresponding to the final classification result.

2. The method for classifying and managing tobacco retailers according to claim 1, characterized in that, Before obtaining multi-source data on tobacco retailers, the process includes: Based on the data structure type, the multi-source data is divided into structured data, semi-structured data, and unstructured data; Preprocessing is performed on different types of data; the preprocessing includes outlier handling, data standardization, and format unification.

3. The method for classifying and managing tobacco retailers according to claim 1, characterized in that, The sliding window algorithm constructs an improved Boston Matrix based on the multi-source data, with relative sales contribution as the horizontal axis and cigarette sales growth rate as the vertical axis. Each tobacco retailer is assigned to a quadrant corresponding to the improved Boston Matrix to obtain initial classification results, including: The relative sales contribution and cigarette sales growth rate of each tobacco retailer are calculated based on the multi-source data. The sliding window algorithm is used to calculate the average regional sales contribution and median regional growth rate for the current period. Establish the improved Boston Matrix corresponding to the current period; the improved Boston Matrix is ​​a two-dimensional coordinate system in which the mean of the regional sales contribution in the current period is the horizontal axis threshold and the median of the regional growth rate is the vertical axis threshold; Based on the multi-source data of the current period, the tobacco retailers are classified into the corresponding quadrants of the two-dimensional coordinate system of the current period, and the initial classification result is determined as the quadrant position of the tobacco retailers.

4. The method for classifying and managing tobacco retailers according to claim 1, characterized in that, The process of calculating the credit correction factor based on the risk types identified by the association rule base and the initial classification results includes: Based on the multi-source data of the current retailer in the current period, obtain the type of violation risk identified by the association rule base; Based on the pre-established quadrant credit scoring relationship, a basic credit score is determined according to the initial classification results; the quadrant credit scoring relationship includes the correspondence between the quadrant and the basic credit score; The risk-weighted score is obtained by weighting and adjusting the basic credit score based on the aforementioned types of violation risks. The credit correction factor is calculated based on the risk-weighted score and the multi-source data for the current period.

5. The method for classifying and managing tobacco retailers according to claim 4, characterized in that, The multi-source data includes the number of violations, credit repair duration, and complaint rate. The credit correction factor, calculated based on the risk-weighted score and the multi-source data for the current period, includes: The penalty amount is calculated based on the number of violations, the credit repair time, and the complaint rate. The credit correction factor is calculated based on the risk-weighted score and the penalty amount.

6. The method for classifying and managing tobacco retailers according to claim 1, characterized in that, The step of calibrating the initial classification result according to the credit correction factor to obtain the final classification result includes: When the credit correction factor is less than the first threshold, the current quadrant of the tobacco retailer is adjusted in the direction of increasing credit supervision level, and the adjusted quadrant position is used as the final classification result. When the credit correction factor is greater than or equal to the second threshold, the current quadrant of the tobacco retailer is adjusted in the direction of decreasing credit supervision level, and the adjusted quadrant position is used as the final classification result; the second threshold is greater than the first threshold.

7. The method for classifying and managing tobacco retailers according to claim 1, characterized in that, The step of executing the corresponding regulatory strategy based on the quadrant position corresponding to the final classification result includes: If the quadrant position is the first quadrant, the first strategy is executed. If the quadrant position is the second quadrant, the second strategy is triggered. If the quadrant position is the third quadrant, the third strategy is triggered. If the quadrant position is the fourth quadrant, the fourth strategy is triggered. Among them, the relative sales contribution of the first to fourth quadrants increases sequentially, the supply allocation level of the first to fourth strategies decreases sequentially, and the credit supervision level of the first to fourth strategies increases sequentially.

8. A classification management system for tobacco retailers, characterized in that, include: The processing module is used to acquire multi-source data from tobacco retailers; A construction module is used to extract feature relationships between different multi-source data, and to construct an association rule base for identifying the violation risk types of the tobacco retailers based on the feature relationships; The matching module is used to construct an improved Boston Matrix based on the multi-source data using the sliding window algorithm, with relative sales contribution as the horizontal axis and cigarette sales growth rate as the vertical axis, and to assign each tobacco retailer to the quadrant corresponding to the improved Boston Matrix to obtain the initial classification result. The improved Boston matrix comprises four quadrants; The calculation module is used to calculate the credit correction factor based on the risk types identified by the association rule base and the initial classification results. An adjustment module is used to calibrate the initial classification result according to the credit correction factor to obtain the final classification result; The execution module is used to execute the corresponding regulatory strategy based on the quadrant position corresponding to the final classification result.

9. An electronic device, characterized in that, Including memory and processor, among which, The memory is used to store programs; The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps in the tobacco retailer classification management method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Used to store computer-readable programs or instructions, which, when executed by a processor, can implement the steps in the classification management method for tobacco retailers as described in any one of claims 1 to 7.

Images