Regional alliance merchant data processing method and system

Through data mining and cluster analysis technology, high-profit long-tail accessories and potential consumer groups are identified, combined with correlation rule learning and nonlinear analysis, the problem of difficulty in identifying niche accessories and market changes in existing technologies is solved, and real-time market data is achieved Updates and strategy optimization have improved sales and profits.

CN118626959BActive Publication Date: 2024-12-03DECHE CHUANGRONG (BEIJING) TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410736145.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-06-07
Publication Date
2024-12-03
Estimated Expiration
2044-06-07

AI Technical Summary

Technical Problem

Existing data analysis technology is difficult to accurately identify and classify niche vehicle accessories, and the lack of a highly automated data processing mechanism makes it difficult for merchants to update market data in real time and adjust inventory and marketing strategies, and miss the opportunity to utilize the long-tail effect.

Method used

Data mining technology is used to classify vehicle accessories sales data, use cluster analysis and association rule learning algorithms, combine nonlinear analysis and network theory to identify high-profit long-tail accessories and potential value target consumer groups, and analyze different vehicle accessories. The most strategically valuable rules are selected.

Benefits of technology

Through in-depth data insights and market segmentation, companies can adjust their products and services faster, improve market response speed and competitiveness, optimize inventory and marketing strategies, maximize sales growth and profits, and enhance their advantages in dynamic markets.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118626959B_ABST
    Figure CN118626959B_ABST
Patent Text Reader

Abstract

The present invention discloses a regional alliance merchant data processing method and system, which specifically relates to the field of sales data processing, and is used to solve the problem of auto parts sales, and optimizes parts sales and market strategies through data mining technology. First, the sales data of vehicle parts are collected from the sales system of each alliance member, and these data are classified using data mining technology to identify long-tail parts with low sales but high profit potential. After that, the classification results are deeply analyzed, key features are extracted, and cluster analysis algorithms are applied to further segment the market and product characteristics, so as to determine target consumer groups with potential value. Then, the transaction data of these target groups are analyzed through association rule learning algorithms to reveal the correlation between purchase patterns and user behaviors between different vehicle parts. Finally, the obtained association rules are analyzed for time stability and market sensitivity, and the rules with the most strategic value are screened out for further implementation.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of sales data processing, and more specifically, to a method and system for processing regional alliance merchant data. Background Art

[0002] In the modern automotive industry, the sales network of vehicle accessories is complex and large-scale. Merchant alliances have become a key strategy to improve efficiency and market competitiveness by sharing resources and customer data. However, despite this, existing data analysis technologies still face many challenges, especially when dealing with diverse and extensive vehicle accessories sales data. The main shortcoming of the current system is that it over-focuses on mainstream accessories with high sales volume, while ignoring long-tail accessories with low demand but high profit potential. In addition, traditional data analysis tools such as threshold screening and simple classification methods often fail to accurately identify and classify niche accessories that are only popular in specific models or regions. The lack of a highly automated data processing mechanism also makes it impossible for merchants to update market data in real time, making it difficult to quickly adjust inventory and marketing strategies to cope with rapid market changes. In addition, existing technologies fail to effectively implement targeted clustering analysis and association rule learning, thus missing the opportunity to use the long-tail effect to explore niche markets, expand market share, and enhance competitiveness.

[0003] In order to solve the above problems, a technical solution is now provided. Summary of the invention

[0004] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a regional alliance merchant data processing method and system, which optimizes accessory sales and market strategies through data mining technology. First, the sales data of vehicle accessories are collected from the sales system of each alliance member, and these data are classified using data mining technology to identify long-tail accessories with low sales but high profit potential. After that, the classification results are deeply analyzed, key features are extracted, and cluster analysis algorithms are applied to further segment the market and product characteristics, so as to determine target consumer groups with potential value. Then, the transaction data of these target groups are analyzed through association rule learning algorithms to reveal the correlation between purchase patterns and user behaviors between different vehicle accessories. Finally, the obtained association rules are analyzed for time stability and market sensitivity, and the rules with the most strategic value are screened out for further implementation to solve the problems raised in the above background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] S1: Automatically collect vehicle parts sales data provided by each alliance member and classify the data using data mining technology to identify high-profit long-tail auto parts;

[0007] S2: Extract key features from the recognition results and apply cluster analysis algorithms to further segment the market and product characteristics based on accessory sales data and consumer purchase patterns to identify the target purchase groups that are not easily noticed but have potential value in the market;

[0008] S3: Use association rule learning algorithms to extract accessory purchase rule information from the transaction data of the identified purchase target group, analyze the purchase patterns and user behavior associations between different vehicle accessories, and obtain association rules;

[0009] S4: Combining nonlinear analysis and network theory, further analyze the temporal stability and sensitivity of each association rule to market changes, and evaluate and screen the association rules.

[0010] In a preferred embodiment, S2 includes the following contents:

[0011] S2-1, collecting classified vehicle parts data from step S1, i.e., vehicle parts identified as high-profit long-tail parts;

[0012] According to business needs and preliminary analysis results, statistically analyze the characteristics that affect market segmentation;

[0013] S2-2, calculate the feature information value, feature interaction diagram and network clustering coefficient of each feature;

[0014] The feature information value is the ability of a feature to reduce system uncertainty, which is quantified using the concept of conditional entropy: ;

[0015] in is the entropy of the target variable, is a given feature conditions The conditional entropy of the target variable is is the condition observed in the dataset probability;

[0016] Constructing a feature interaction diagram , where nodes represent features and edge weights are determined by the conditional mutual information between features Determined, reflects the amount of information shared by two features when the target variable is known: ;

[0017] in It is a feature and The joint entropy of is the target variable The conditional joint entropy of the known case;

[0018] Calculate the network clustering coefficient of each node (feature) in the feature interaction graph, which reflects the closeness of a feature to other features: ;

[0019] in It is through the characteristics The number of directly connected characteristic pairs, Is with characteristics The number of connected nodes.

[0020] In a preferred embodiment, the comprehensive feature information value, the feature interaction diagram and the network clustering coefficient are combined to obtain the comprehensive feature selection index ;

[0021] Screen out the features whose comprehensive feature selection index is greater than or equal to the impact threshold, and identify these features as key features;

[0022] S2-3, all key features were standardized before clustering;

[0023] Use the DBSCAN algorithm with the epsilon and min_samples parameters to identify natural clusters in the data;

[0024] Use internal and external evaluation indicators to evaluate clustering quality;

[0025] According to the clustering results, the target purchasing group is obtained.

[0026] In a preferred embodiment, S3 includes the following contents:

[0027] S3-1, for the purchase target group identified in step S2, collecting relevant transaction data from the purchase target group;

[0028] S3-2, in association rule learning, each purchase session is regarded as a transaction, each purchased accessory is regarded as an item in the transaction, and the data is organized into a transaction list according to the accessory ID and the purchase session;

[0029] Use the Apriori algorithm to find frequent item sets from transaction data, that is, accessory combinations that often appear in the same purchase session;

[0030] Generate association rules from frequent itemsets;

[0031] Filter out association rules based on confidence and lift thresholds.

[0032] In a preferred embodiment, S4 includes the following contents:

[0033] Nonlinear analysis and network theory are combined to obtain the time-dependence spectrum index and the market impact network complexity index.

[0034] In a preferred embodiment, the calculation process of the time-dependent spectral index is as follows:

[0035] For each association rule, collect purchase records within a certain period of time and record whether the corresponding rule is triggered at each time point;

[0036] Perform Fourier transform on the time series data of each rule to convert the time series from the time domain to the frequency domain. The result is expressed as the amplitude and phase of different frequency components.

[0037] ;

[0038] in is time series data, is the frequency The complex result of contains amplitude and phase information;

[0039] Compute the time-dependent spectral index: ;

[0040] in is the frequency The amplitude of represents the intensity of the frequency component; is the corresponding phase spectrum, is the highest frequency analyzed.

[0041] In a preferred embodiment, the process of obtaining the market impact network complexity index is as follows:

[0042] Collect market variable data and purchase behavior data related to each association rule;

[0043] Construct a network where nodes include market variables and accessories in association rules. The edge weights between each pair of nodes are calculated by the conditional mutual information between them, indicating the reduction in uncertainty of the state of one node when the state of another node is: ;

[0044] in and are any two nodes in the network, and be it given market conditions or other relevant factors;

[0045] Compute the Laplacian matrix from the adjacency matrix of a network , defined as ,in is the degree matrix, is the adjacency matrix;

[0046] Calculate the Market Impact Network Complexity Index: ;

[0047] in are the non-zero eigenvalues ​​of the Laplacian matrix, is the number of nodes in the network.

[0048] In a preferred embodiment, the time-dependent spectrum index and the market impact network complexity index are comprehensively analyzed and calculated to obtain the market dynamic adaptability coefficient , the calculation formula is: ;

[0049] in:

[0050] is the time-dependent spectral index;

[0051] is the market impact network complexity index;

[0052] and are the weight factors of the time-dependency spectrum index and the market impact network complexity index, respectively;

[0053] is the variance of the time-dependence spectrum index and the market impact network complexity index, indicating the magnitude of changes in these two indices under different market conditions;

[0054] is the covariance of the time-dependency spectrum index and the market impact network complexity index;

[0055] is a small constant used to ensure that the denominator is not zero.

[0056] In a preferred embodiment, the market dynamic adaptability coefficient is compared with a value threshold;

[0057] If the market dynamic adaptability coefficient is greater than or equal to the value threshold, it indicates that the association rule is very suitable for the current market environment, generating a signal for enhanced implementation;

[0058] On the contrary, if the market dynamic adaptability coefficient is less than the value threshold, it indicates that the association rule is not stable enough in time and generates a cautious assessment signal.

[0059] The regional alliance merchant data processing system includes a data collection engine module, a feature insight module, a rule mining module, and a rule selection module;

[0060] The data collection engine module automatically collects the vehicle parts sales data provided by each alliance member, and uses data mining technology to classify the data to identify high-profit long-tail auto parts, and sends the identification results to the feature insight module;

[0061] The feature insight module extracts key features from the recognition results and applies cluster analysis algorithms to further segment the market and product characteristics based on the sales data of accessories and consumer purchase patterns, so as to identify the purchase target groups that are not easily noticed but have potential value in the market, and send the determined purchase target groups to the rule mining module;

[0062] The rule mining module uses the association rule learning algorithm to extract the accessory purchase rule information from the transaction data of the identified purchase target group, analyzes the purchase pattern and user behavior association between different vehicle accessories, obtains the association rules, and sends the association rules to the rule selection module;

[0063] The rule selection module combines nonlinear analysis and network theory to further analyze the temporal stability and sensitivity to market changes of each association rule, and evaluates and screens the association rules.

[0064] The technical effects and advantages of the regional alliance merchant data processing method and system of the present invention are as follows:

[0065] 1. The present invention provides in-depth data insights and market segmentation for the vehicle accessories market by comprehensively applying feature selection and cluster analysis techniques. First, by calculating feature information values, market dynamic response metrics, constructing feature interaction diagrams, and calculating network clustering coefficients, it is ensured that the selected features not only have high information value, but also occupy a core position in the feature network, minimizing information redundancy. This feature selection process ensures the quality and reliability of cluster analysis, so that the clustering results can effectively reveal different consumer groups and product trends in the market. In turn, it can better adjust its products and services to meet market demand, improve market response speed and competitiveness, and ultimately achieve sales growth and profit maximization.

[0066] 2. The present invention provides enterprises with a powerful tool to identify and optimize their market strategies by comprehensively evaluating the time stability and market sensitivity of each association rule. By calculating the time dependency spectrum index and the market impact network complexity index, and integrating these indices into the market dynamic adaptability coefficient, enterprises can clearly see which rules are highly sensitive to market changes while maintaining time consistency, so that they can predict and adapt to rapid changes in the market. This enables enterprises to more accurately locate which marketing and product promotion strategies are most effective, and optimize inventory and pricing strategies accordingly to improve efficiency and reduce waste. In addition, it is possible to identify rules that may require further adjustment or in-depth analysis, thereby avoiding reliance on ineffective or unstable rules in market strategies. This will not only help improve the market response speed and flexibility of enterprises, but also significantly enhance the advantages of enterprises in competition, ensuring the best business performance and customer satisfaction in a dynamic and complex market environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 It is a flow chart of the regional alliance merchant data processing method of the present invention;

[0068] Figure 2 It is a structural schematic diagram of the regional alliance merchant data processing system of the present invention. DETAILED DESCRIPTION

[0069] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Example

[0070] Figure 1 The present invention provides a method for processing regional alliance merchant data, including:

[0071] S1: Automatically collect vehicle parts sales data provided by each alliance member and classify the data using data mining technology to identify high-profit long-tail auto parts;

[0072] S2: Extract key features from the classification results and apply cluster analysis algorithms to further segment the market and product characteristics based on accessory sales data and consumer purchase patterns to identify the target purchase groups that are not easily noticed but have potential value in the market;

[0073] S3: Use association rule learning algorithms to extract accessory purchase rule information from the transaction data of the identified purchase target group, analyze the purchase patterns and user behavior associations between different vehicle accessories, and obtain association rules;

[0074] S4: Further analyze the temporal stability and sensitivity of each association rule to market changes, and evaluate and screen the association rules.

[0075] S1 includes the following:

[0076] The sales data of vehicle accessories are automatically collected from the sales systems of each alliance member through integrated APIs. These data include the accessories’ SKU (stock keeping unit number), sales time, sales location, sales quantity, customer purchase records and price information.

[0077] The collected data also includes customers’ purchase history and behavior data, such as online inquiry history, purchase frequency and purchase preferences, which are obtained through the alliance members’ CRM system or online sales platform.

[0078] Clean the collected raw data, remove duplicate records, correct erroneous or incomplete data entries, and perform data standardization to ensure consistency and accuracy of subsequent analysis.

[0079] Among the three data mining techniques, decision trees, random forests, and support vector machines (SVMs), random forests are more suitable for data classification based on the sales frequency, category, and customer group of vehicle accessories. Decision trees are easy to understand and implement, but they tend to overfit, especially when the data categories are complex or the data volume is large. As an integrated learning method, random forests greatly improve the stability and accuracy of the model and reduce the risk of overfitting by building multiple decision trees and voting to obtain the final result. In addition, random forests can effectively handle large data sets and data with high-dimensional features, and are suitable for processing nonlinear relationships and complex interaction effects in vehicle accessories sales data. In contrast, although SVM performs well in processing small samples, nonlinear and high-dimensional data, its training time may be long when processing large-scale data, and its parameter adjustment is more complicated, which is not as good as random forests in terms of operability and efficiency. Therefore, random forests provide a robust and efficient solution when processing vehicle accessories sales data.

[0080] Random forests are used to intelligently classify data based on the sales frequency, category, and customer group of vehicle accessories. The purpose of classification is to identify long-tail accessories that have low sales but potentially high profit margins. The processing includes the following:

[0081] S1-1. For key data that have been collected, apply automated scripts to remove outliers, process missing values, and standardize data formats to ensure the quality and consistency of input data.

[0082] Extract and construct new features from raw data, such as seasonal sales indicators, customer purchase frequency, regional sales trends, etc. These features are designed to capture subtle differences in sales patterns and consumer behavior;

[0083] S1-2, determine the key parameters of the random forest model, such as the number of trees, the maximum depth of each tree, the minimum number of samples required for each node split, and the random selection method of features;

[0084] in:

[0085] Determine the number of trees: The number of trees is one of the most important parameters in random forests, affecting the performance and degree of overfitting of the model. Generally speaking, the more trees there are, the more stable and accurate the model is, but the computational cost is also greater. Test different numbers of trees through the GridSearch method, for example, starting with 10 trees and gradually increasing to 100, 200, 300, etc., and observe the changes in model performance. Select the number of trees that has the lowest average error rate in cross-validation and an acceptable computational cost.

[0086] Set the maximum depth of each tree: The maximum depth of each tree determines the complexity of the tree. Deeper trees can learn more detailed data details, but the risk is that they may cause overfitting. Using a grid search-like method, test starting with no limit and gradually limiting the depth of the tree (such as 5, 10, 20...), and determine the best depth based on the performance of the model and the overfitting situation.

[0087] Adjust the minimum number of samples required for each node split: This parameter defines the minimum number of samples a node in the tree must have before it can continue to split into child nodes. Setting a larger value can prevent the model from learning noise in the data, but setting it too large may cause the model to fail to capture important patterns in the data. You can start with a value of 2 (the default), and increase this parameter (for example, try 10, 20, 50, etc.) to test the changes in model performance and select the optimal solution.

[0088] Random selection method for features: This parameter determines the number of features that the random forest considers at each split. Common settings include "auto", "sqrt", "log2" or a specific number. "sqrt" randomly selects the square root of the total number of features at each split point, and "log2" selects the logarithm of the number of features. By comparing the effects of these settings through cross-validation, you can find the feature selection strategy that best suits the current dataset.

[0089] The dataset is divided into training and test sets with an 80 / 20 ratio. The training set is used to train the random forest model, and the test set is reserved for the final model evaluation;

[0090] In order to reduce the evaluation error and optimize the parameter selection during the model training process, k-fold cross validation was implemented, and k = 10 was selected. This means that the data set was divided into 10 equal parts, 9 parts were used for training and 1 part was used for testing in turn, and this was repeated 10 times to ensure that each data partition was used as a test set.

[0091] S1-3, use the trained random forest model to test the data set, perform classification prediction, and identify long-tail accessories with low sales but high profit potential;

[0092] Model application: Use the trained random forest model to classify the test dataset. At this stage, the model evaluates each data point (i.e., vehicle accessory) based on the learned pattern and predicts whether it belongs to the long-tail accessory with low sales volume but high profit potential;

[0093] Classification prediction is performed through the probability output generated by the model, where the model outputs the probability of each accessory belonging to the high-profit long-tail category. A threshold (such as 0.5) is set, and when the probability output by the model exceeds this threshold, the accessory is identified as a high-profit long-tail accessory.

[0094] Identify high-profit long-tail accessories: Based on the prediction results, all accessories in the test data will be divided into two categories: ordinary accessories and potential high-profit long-tail accessories. This classification facilitates the subsequent business strategy formulation, such as specific marketing activities or inventory management.

[0095] Analyze and report the importance of each feature in the random forest decision process to identify the most influential factors in classifying high-profit long-tail accessories;

[0096] Calculating feature importance: The random forest model can automatically calculate the importance of each feature in the prediction process. It is evaluated by observing the contribution of each feature to the accuracy of the model when building the decision tree, that is, the frequency of each feature used in the split node and the degree to which the split improves the model performance. Feature importance is usually expressed as a score or percentage, showing the contribution of each feature to the model's predictive ability.

[0097] Report and analyze feature importance: Generate a feature importance report that lists all features and their importance scores. This can be presented visually, such as a bar chart or importance ranking. Analyze this feature importance data to identify the most influential factors for high-margin long-tail accessory categories. For example, if a region or specific customer group has a particularly high feature importance score, then these factors should be given priority consideration in future sales and marketing strategies.

[0098] S1-3, analyze the classification results, identify and mark the auto parts as high-profit long-tail parts;

[0099] Integrate the identified long-tail accessories information into business intelligence tools to help marketing and sales teams develop targeted promotion and pricing strategies and optimize inventory management.

[0100] S2 includes the following:

[0101] S2-1, collecting classified vehicle parts data from step S1, i.e., vehicle parts identified as high-profit long-tail parts;

[0102] Based on business needs and preliminary analysis results, select features that affect market segmentation, such as time series trends in sales data, geographic location information, purchase frequency of customer groups, etc. In addition, include features extracted from consumer behavior analysis, such as purchase cycle, price sensitivity, and brand loyalty.

[0103] S2-2, statistical features, calculate the feature information value, feature interaction diagram and network clustering coefficient of each feature;

[0104] The feature information value is the ability of a feature to reduce system uncertainty, which is quantified using the concept of conditional entropy: ;

[0105] in is the entropy of the target variable, is a given feature conditions The conditional entropy of the target variable is is the condition observed in the dataset probability.

[0106] Market dynamic response metrics are metrics designed to capture and quantify the sensitivity and responsiveness of individual features to market changes. In the automotive parts market, specific product characteristics such as type, sales frequency, and seasonality may be particularly sensitive to market changes. By calculating market dynamic response metrics, those features that have the greatest impact on sales and profits are identified, which can indicate changes in market demand or trends in consumer behavior. For example, sales of a certain accessory may increase sharply in the summer, and the identification of this pattern is crucial for predicting future sales trends, optimizing inventory management, and adjusting marketing strategies. Therefore, market dynamic response metrics not only help companies understand the specific contribution of each feature to business goals, but also enable companies to respond more quickly in a highly competitive market environment.

[0107] Constructing a feature interaction diagram , where nodes represent features and edge weights are determined by the conditional mutual information between features Determined, reflects the amount of information shared by two features when the target variable is known: ;

[0108] in It is a feature and The joint entropy of is the target variable The conditional joint entropy of the known case.

[0109] The calculation of the feature interaction diagram aims to reveal the interactions and dependencies between different features, forming a visual network diagram of the relationship between features. In the analysis of the vehicle accessories market, it is extremely important to understand how the various features are related to each other, because these relationships may reveal potential market opportunities or risks. For example, consumers in a certain geographical location may prefer a specific brand of accessories, and this preference may have complex interactions with other features such as price sensitivity and purchase frequency. By calculating the feature interaction diagram, analysts can identify core feature clusters and how these clusters influence each other, thereby providing support for the formulation of more effective target market strategies and product positioning. In addition, this topological mapping helps to optimize the feature selection process, ensuring that the selected feature set can comprehensively cover important market dynamics, avoid over-reliance on a single feature, and improve the robustness and explanatory power of the model.

[0110] Calculate the network clustering coefficient of each node (feature) in the feature interaction graph, which reflects the closeness of a feature to other features: ;

[0111] in It is through the characteristics The number of directly connected characteristic pairs, Is with characteristics The number of connected nodes (degree).

[0112] The calculation of the network clustering coefficient is to evaluate the closeness of nodes (i.e., features) in the network structure formed between features. In the application of vehicle accessories market analysis, calculating the network clustering coefficient helps to identify whether there is a strong clustering phenomenon between features, which usually indicates that certain feature combinations may have a mutually reinforcing effect in influencing market behavior. For example, if sales data of a certain region, a specific type of vehicle accessories, and certain consumer group characteristics often appear together in a group with a high clustering coefficient, this may indicate that these features are interdependent and jointly affect the sales performance of accessories. Such information is crucial for understanding market distribution, optimizing product promotion strategies, and adjusting supply chain management. By integrating the network clustering coefficient in the feature selection process, companies can ensure that the selected feature set not only performs well when analyzed separately, but also provides the best predictive performance and business insights in a multi-feature interaction environment, thereby achieving more refined market segmentation and more effective marketing resource allocation. In addition, this network analysis method can also help avoid over-reliance on highly correlated redundant features in the model, thereby improving the efficiency and effectiveness of analysis.

[0113] The comprehensive feature selection index is obtained by combining the feature information value, feature interaction diagram and network clustering coefficient ;

[0114] The calculation formula design of the comprehensive feature selection index combines the concepts of information theory and network science. Its core idea is to evaluate not only the influence of a single feature on the target variable, but also the interaction between features and their position in the network structure. First, the feature information value quantifies the ability of each feature to reduce the uncertainty of the target variable alone, which directly reflects the value of the feature in prediction. Then, by calculating the conditional mutual information between features, we can evaluate the amount of information shared between feature pairs when the target variable is known, which helps to identify redundant features in the data and ensure that the selected feature set is complementary in information provision. Finally, the network clustering coefficient describes the clustering tendency of features in the entire feature network. A higher network clustering coefficient value means that the feature is highly clustered with other features, which may indicate that the feature is part of an important feature set that may represent more complex patterns in the data. The comprehensive feature selection index provides a comprehensive perspective to evaluate and select features by combining the measurements of these three dimensions - the information contribution of a single feature, the information sharing between features, and the structural position of features in the network, aiming to optimize the feature set to maximize prediction performance and reduce data redundancy. This approach is particularly suitable for complex data environments, such as the vehicle parts market, where the interrelationships and influences between features may have a significant impact on the prediction results.

[0115] The value of the comprehensive feature selection index is used to measure the comprehensive utility of each feature in the prediction model, which specifically reflects the information content, independence and cluster importance of the feature in the network. The larger the value of the comprehensive feature selection index, the higher the individual contribution of the feature in reducing the uncertainty of the target variable, the less information overlap with other features, and the core or important position in the feature network. This means that features with high comprehensive feature selection index values ​​not only have strong predictive power for the target variable, but also provide unique information in the feature set, which helps to improve the overall performance and explanatory power of the model. On the contrary, features with low comprehensive feature selection index values ​​may have weak information contribution, be highly redundant with other features, or have a marginal position in the feature network. These features may not be suitable for selection into the final model because they may increase the complexity of the model without sufficiently improving the prediction accuracy of the model. Therefore, the comprehensive feature selection index, as a metric, helps data scientists and analysts optimize feature selection and ensure that the built model is both accurate and highly explanatory. It is particularly suitable for scenarios with high data dimensions and complex relationships between features, such as vehicle parts sales analysis.

[0116] Calculate the comprehensive feature selection index for each feature, and compare the comprehensive feature selection index with the corresponding threshold. If the comprehensive feature selection index is greater than or equal to the impact threshold, it means that the feature has significant value in predicting consumer purchasing patterns. Specifically, such features show high utility in reducing the uncertainty of target variables (such as purchase volume, purchase frequency, or customer preferences), while occupying a core position in the feature network, providing unique and important information, and having little information overlap with other features. In the analysis of the vehicle accessories market, features with high comprehensive feature selection index values ​​may include the customer's geographic location, the frequency of purchasing a specific type of accessories, or historical purchasing patterns. These features can help companies more accurately identify potential market opportunities, develop targeted marketing strategies, and optimize inventory and pricing strategies;

[0117] On the contrary, if the comprehensive feature selection index is less than the impact threshold, this indicates that the feature has a low contribution to predicting consumer purchase patterns, perhaps because the feature information is not highly correlated with the target variable, or is relatively marginal in the feature network, or is highly redundant with other features. In practical applications, features with low comprehensive feature selection index values ​​may not be selected into the final analysis model because they may not significantly improve the model's predictive ability, while increasing the model's complexity, leading to overfitting or reducing the model's interpretability. For example, some broad customer base characteristics in vehicle parts sales (such as age or gender) may have little impact on the purchase decision of a specific part, while more detailed features (such as driving habits or vehicle type) may be more critical.

[0118] The features whose comprehensive feature selection index is greater than or equal to the impact threshold are screened and identified as key features.

[0119] S2-3, before clustering, the selected features are standardized (such as Z-score standardization or minimum-maximum standardization) to ensure the dimensional consistency of different features and avoid the algorithm being overly sensitive to certain features with a large numerical range.

[0120] Use the DBSCAN algorithm with the epsilon and min_samples parameters to identify natural clusters in the data.

[0121] The quality of clustering is assessed using internal evaluation metrics such as the silhouette coefficient and external metrics such as the adjusted Rand index. These metrics can help verify whether the clusters are meaningful and discriminative.

[0122] Based on the clustering results, we can identify the common features in each cluster, understand the market segments or consumer groups represented by different clusters, and obtain the target purchase groups. For example, we may find that a cluster mainly includes high-value but low-frequency accessories, which may target a specific consumer group.

[0123] Suppose there is a company that sells vehicle parts and needs to perform market segmentation analysis on its product line to better understand which parts have high potential for profit despite low purchase frequency. The following are the specific steps and hypothetical examples of the company's market segmentation using cluster analysis:

[0124] The following features were selected for cluster analysis:

[0125] Sales frequency (monthly sales)

[0126] Average Sales Price

[0127] Product category (e.g. engine parts, body parts, electronic equipment, etc.)

[0128] Geographic Sales Area

[0129] The Z-score standardization method is used to process the above features to ensure the dimensional consistency of different features and eliminate the influence of the numerical range on the clustering results.

[0130] The DBSCAN algorithm is selected for clustering because it can effectively handle clusters of different shapes and sizes and has good robustness to noise.

[0131] The epsilon is set to 0.5 and the min_samples is set to 10. These parameters are determined through multiple experiments to ensure that obvious and meaningful clusters can be found.

[0132] The silhouette coefficient is used to evaluate the cohesion and separation of clusters. A high silhouette coefficient indicates that the clustering results are more reasonable. The adjusted Rand index is used to evaluate the stability and accuracy of clustering.

[0133] The data clustering results revealed three main cluster groups:

[0134] Cluster A: Contains high-value but low-frequency electronic devices, mainly sold in urban areas.

[0135] Group B: Contains vehicle body parts of medium value and medium frequency, mainly sold in suburban areas.

[0136] Group C: Contains low-value, high-frequency purchased engine parts with a wide sales area.

[0137]

[0138] Through cluster analysis, a group of electronic devices with potential high profits (Group A) was identified. Although these accessories are not sold frequently, they have great market potential in urban areas due to their high unit price and specialization. The company can increase marketing investment and optimize inventory strategies for this market segment to improve overall profits.

[0139] The present invention provides in-depth data insights and market segmentation for the vehicle accessories market by comprehensively applying feature selection and cluster analysis techniques. First, by calculating feature information values, market dynamic response metrics, constructing feature interaction diagrams, and calculating network clustering coefficients, it is ensured that the selected features not only have high information value, but also occupy a core position in the feature network, minimizing information redundancy. This feature selection process ensures the quality and reliability of cluster analysis, so that the clustering results can effectively reveal different consumer groups and product trends in the market. In turn, it can better adjust its products and services to meet market demand, improve market response speed and competitiveness, and ultimately achieve sales growth and profit maximization.

[0140] S3 includes the following:

[0141] S3-1, for the target purchase group (high-profit long-tail accessories group) identified in step S2, collect relevant transaction data from the target purchase group, including accessory purchase records, such as purchase date, purchase quantity, accessory ID, customer ID, etc.

[0142] Clean transaction data, including removing duplicate records, filling or removing missing values, and converting the data format to suit the needs of association rule algorithms.

[0143] S3-2, in association rule learning, each purchase session is considered as a transaction, and each purchased accessory is considered as an item in the transaction. The data is organized into a transaction list based on the accessory ID and purchase session;

[0144] Use the Apriori algorithm to find frequent item sets from transaction data, that is, those accessory combinations that often appear in the same purchase session. Based on a predetermined support threshold, filter out sufficiently frequent item sets.

[0145] Generate association rules from frequent itemsets. Association rules are expressed in the form of "if accessory A is purchased, then accessory B is also purchased". Use statistical metrics such as confidence and lift to evaluate the strength and relevance of these rules.

[0146] Confidence: Measures the conditional probability of purchasing accessory B given that accessory A has been purchased.

[0147] Lift: reflects the ratio of the probability of accessory A and accessory B being purchased together to the probability of them being purchased independently. A lift greater than 1 indicates a positive correlation between the two accessories.

[0148] Filter out association rules (which accessories are often purchased together) based on confidence and lift thresholds. Make sure the rules selected are not only frequent but also highly predictive.

[0149] To specifically illustrate how to use confidence and lift thresholds to filter out meaningful association rules, this article assumes that there is a set of transaction data for vehicle accessories and shows how to extract association rules from this data and how to filter them based on confidence and lift.

[0150] Assume that there is the following transaction data, which records the vehicle accessories purchased by consumers in a month:

[0151]

[0152] First, use the Apriori algorithm and other methods to find frequent item sets from the above data. Assuming that the support threshold is set to 50%, the following are the calculated frequent item sets:

[0153] {A}: Support = 50% (4 occurrences / 8 transactions)

[0154] {B}: Support = 62.5% (5 occurrences / 8 transactions)

[0155] {C}: Support = 62.5% (5 occurrences / 8 transactions)

[0156] {D}: Support = 50% (4 occurrences / 8 transactions)

[0157] {B,C}: Support = 37.5% (3 occurrences / 8 transactions)

[0158] Generate possible association rules from these frequent itemsets, such as:

[0159] Rule 1: If you buy B (wipers), you also buy C (tires)

[0160] Rule 2: If C (tires) is purchased, B (wipers) is also purchased

[0161] Compute the confidence and lift of these rules:

[0162] Rule 1:

[0163] Confidence = P(C|B) = 60% (3 times B and C appear together / 5 times B appears)

[0164] Lift = P(C|B) / P(C) = 0.6 / 0.625 = 0.96

[0165] Rule 2:

[0166] Confidence = P(B|C) = 60% (3 times B and C appear together / 5 times C appears)

[0167] Lift = P(B|C) / P(B) = 0.6 / 0.625 = 0.96

[0168] Assume that the confidence threshold is set to 60% and the lift threshold is set to 1. According to these thresholds, although the confidence of both rule 1 and rule 2 meet the conditions, the lift does not reach the threshold, indicating that although the purchases of B and C are related, this correlation may be due to their respective high independent purchase probabilities, and they are not strongly related to each other.

[0169] The following table shows the analysis results of these two rules:

[0170]

[0171] This example shows that by calculating the confidence and lift, we can effectively screen out meaningful association rules. Although B and C are often purchased together, the lift is less than 1, indicating that there is not enough positive correlation between them, which may guide the marketing strategy to not emphasize bundling them.

[0172] After obtaining the association rules, further analysis of these rules, combined with nonlinear analysis and network theory, to evaluate the stability of each association rule over time and its sensitivity to market changes is a critical step, because such in-depth analysis can help enterprises more accurately understand the long-term effectiveness of the rules and the reliability of practical applications. First, the analysis of temporal stability can reveal whether the association rules remain consistent over time, or whether their effectiveness is affected by specific seasons or market cycles. The high temporal stability of the rules means that enterprises can rely on these rules for long-term business decisions and strategic planning. Second, the dynamics of the market change very quickly, especially in highly competitive industries, so analyzing the sensitivity of association rules to market changes through network theory can help enterprises understand the performance of these rules under different market conditions. This analysis can show the interdependence and influence between rules and market variables, so that enterprises can adjust their strategies to adapt to market changes, such as economic recession, changes in consumer behavior, or the entry of new competitors. Combining nonlinear analysis and network theory not only improves the depth and breadth of the analysis, but also makes the analysis of association rules more comprehensive and accurate, so that enterprises can effectively deal with the uncertainty and complexity of the market. Help enterprises grasp market trends, optimize product promotion, and adjust market strategies in a timely manner, ultimately achieving the goal of improving market competitiveness and business results.

[0173] S4 includes the following:

[0174] First, organize the historical transaction records for each association rule and record whether the rule is triggered (i.e., whether the product combination in the rule is purchased together) at each time point (e.g., every month). Combine nonlinear analysis and network theory to obtain the time-dependent spectrum index and market impact network complexity index;

[0175] The calculation process of the time-dependent spectral index is as follows:

[0176] For each association rule, purchase records within a certain period of time are collected, and whether the rule is triggered at each time point (such as every month) (that is, whether the product combination in the rule is purchased together).

[0177] Perform Fourier transform on the time series data of each rule to convert the time series from the time domain to the frequency domain. The result is expressed as the amplitude and phase of different frequency components.

[0178] ;

[0179] in is time series data, is the frequency The complex result of contains amplitude and phase information;

[0180] Compute the time-dependent spectral index: ;

[0181] in is the frequency The amplitude of represents the intensity of the frequency component; is the corresponding phase spectrum, is the highest frequency analyzed;

[0182] Amplitude Indicates the frequency The strength of the buying pattern, the phase Represents how these patterns change over time.

[0183] The time-dependent spectral index is used to quantify the consistency and predictiveness of association rules for time series data in the frequency domain. It considers not only the intensity of the frequency components, but also the relative changes of these components over time (phase information).

[0184] A time-dependent spectrum index value close to 1 indicates that the purchase pattern of the association rule is highly consistent and predictable in time, indicating that this purchase behavior changes little in time and is highly predictable. This shows that the rule is very stable, and the purchase behavior repeats in a certain time pattern, which is suitable for long-term strategic planning; a lower time-dependent spectrum index value indicates that the purchase pattern is highly inconsistent and random in time, and the purchase behavior is difficult to predict and may be affected by a variety of incidental factors. This rule may require a more flexible response strategy or in-depth analysis of the reasons behind it.

[0185] The process of obtaining the market impact network complexity index is as follows:

[0186] First, collect market variable data (such as price changes, economic indicators, consumer behavior data, etc.) and purchasing behavior data related to each association rule.

[0187] Construct a network where nodes include market variables and accessories in association rules. The edge weight between each pair of nodes is calculated by the conditional mutual information between them, which represents the reduction in uncertainty of the state of one node when the state of another node is: ;

[0188] in and are any two nodes in the network, and be given market conditions or other relevant factors.

[0189] Compute the Laplacian matrix from the adjacency matrix of a network , defined as ,in is the degree matrix (the elements on the diagonal are the degree of the node, i.e. the sum of the weights of the edges connected to the node), is the adjacency matrix.

[0190] Laplacian Matrix The eigenvalues ​​of the network reflect the complexity and connectivity of the network structure.

[0191] Calculate the Market Impact Network Complexity Index: ;

[0192] in are the non-zero eigenvalues ​​of the Laplacian matrix, is the number of nodes in the network.

[0193] The market influence network complexity index is used to quantify the structural complexity of the network of interactions between market variables and purchasing behavior. It reflects the connectivity and stability of the network, as well as the complexity of interactions between nodes.

[0194] The larger the value of the market influence network complexity index, the more complex the interactions between nodes in the network, and the stronger the dependence and responsiveness between market variables and purchasing behavior. This shows that the association rules are very sensitive to market changes and can capture subtle market dynamics; the smaller the value of the market influence network complexity index, the simpler the interactions in the network, the lower the dependence between nodes, and the smaller the direct impact of market factors on purchasing behavior. This may mean that the association rules are less adaptable in the market and are insensitive to market changes.

[0195] The market dynamic adaptability coefficient is obtained by comprehensively analyzing the time-dependent spectrum index and the market impact network complexity index. , the calculation formula is: ;

[0196] in:

[0197] is the time-dependent spectral index, which quantifies the temporal stability of association rules;

[0198] is the market impact network complexity index, which measures the sensitivity of rules to market changes;

[0199] and are the weight factors of the time-dependence spectrum index and the market impact network complexity index, respectively, which are used to adjust the relative importance of time stability and market sensitivity in the index;

[0200] is the variance of the time-dependence spectrum index and the market impact network complexity index, indicating the magnitude of changes in these two indices under different market conditions;

[0201] is the covariance between the time-dependent spectrum index and the market impact network complexity index, which is used to assess the correlation of changes between these two indices;

[0202] is a small constant used to ensure that the denominator is not zero and increase the stability of the calculation.

[0203] This formula enhances the nonlinear characteristics of the original indicator through an exponential function, making the calculation result more sensitive to changes in the indicator value.

[0204] The logarithmic function was used to scale the range of the results to make them more suitable for further analysis and comparison.

[0205] By calculating the variance and covariance and incorporating them into the final index calculation, the formula not only takes into account the independent contributions of the time-dependence spectrum index and the market impact network complexity index, but also takes into account the mutual influence between them, enhancing the model's ability to capture the intrinsic connections of the data.

[0206] The calculation formula of the market dynamic adaptability coefficient combines exponential and logarithmic functions, as well as statistical measures (variance and covariance). The significance and rationality of its construction mainly lie in providing a measure that comprehensively reflects the time stability and market sensitivity of the association rules, while taking into account the interaction between the two. By using the exponential function, the formula can amplify the differences in the two indexes (the time-dependent spectrum index and the market impact network complexity index), so that even small changes can be significantly reflected in the results, increasing the sensitivity of the indicator. The use of the logarithmic function is to adjust this amplification effect and ensure that the numerical range of the results is within a practical range, which is convenient for further analysis and interpretation. At the same time, by incorporating variance and covariance into the calculation, the formula not only focuses on the independent impact of a single index, but also considers the relationship between the time-dependent spectrum index and the market impact network complexity index, thereby more comprehensively evaluating the performance of the rules in the market. The variance part reflects the fluctuation range of the two indexes under different market conditions, while the covariance part evaluates the correlation between their changes. This method can reveal the adaptability and coordination of rules under different market dynamics. Therefore, a high market dynamic adaptability coefficient means that the rule is not only stable in itself, but also highly sensitive to market changes and there is a synergy between the two, which is suitable for application in a rapidly changing market environment. A low market dynamic adaptability coefficient may indicate that the rule needs to be improved in terms of adaptability or coordination. This measurement method makes the market dynamic adaptability coefficient a highly comprehensive and practical tool to help companies optimize strategies in complex market environments and improve the efficiency and effectiveness of decision-making.

[0207] The larger the value of the market dynamic adaptability coefficient is, the more stable the association rule is in time and the more sensitive it is to the market, and the better the coordination between the two. This shows that the rule is very suitable for a dynamically changing market environment and can provide enterprises with stable and sensitive market insights;

[0208] The smaller the value of the market dynamic adaptability coefficient, the more likely it is that the rule is less stable over time, is more sluggish in responding to the market, or lacks synergy between the two, indicating that the rule has lower value in practical applications and may require further adjustment or optimization.

[0209] Compare the market dynamics adaptability coefficient with the value threshold;

[0210] If the market dynamic adaptability coefficient is greater than or equal to the value threshold, it means that the association rule performs well in terms of time stability and market sensitivity, can effectively adapt to market changes, and has good predictability and adaptability. This shows that the association rule is very suitable for the current market environment and can be used as a reliable basis for formulating strategies;

[0211] Companies should use these rules to optimize their marketing strategies and inventory management, such as by promoting product combinations that are often purchased together, implementing bundling, or using these rules in sales forecasting and pricing strategies. This can increase sales, improve customer satisfaction, and optimize inventory;

[0212] Generate reinforcement implementation signals to encourage management and marketing teams to apply these rules more broadly, increase marketing efforts for related products, or adjust promotion strategies for related products.

[0213] On the contrary, if the market dynamic adaptability coefficient is less than the value threshold, it indicates that the association rule is not stable enough in time or is not sensitive enough to market changes and cannot fully capture the dynamic changes of the market, which may lead to poor results or high risks in practical applications;

[0214] The applicability of these rules needs to be reassessed, and further data analysis may be needed to adjust or improve the rules, or develop new rules to better suit the market environment. At the same time, companies may need to update and monitor these rules more frequently to ensure that they can reflect the latest changes in the market in a timely manner;

[0215] Generates a prudent assessment signal that prompts decision makers to conduct in-depth analysis and possible revisions of these rules. This includes reducing the number of decisions that rely on these rules and avoiding over-application of these poorly performing rules in key business operations.

[0216] The present invention provides enterprises with a powerful tool to identify and optimize their market strategies by comprehensively evaluating the time stability and market sensitivity of each association rule. By calculating the time dependency spectrum index and the market impact network complexity index, and integrating these indices into the market dynamic adaptability coefficient, enterprises can clearly see which rules are highly sensitive to market changes while maintaining time consistency, so that they can predict and adapt to rapid changes in the market. This enables enterprises to more accurately locate which marketing and product promotion strategies are most effective, and optimize inventory and pricing strategies accordingly to improve efficiency and reduce waste. In addition, it is possible to identify those rules that may need further adjustment or in-depth analysis, thereby avoiding reliance on ineffective or unstable rules in market strategies. This not only helps to improve the market response speed and flexibility of enterprises, but also significantly enhances the advantages of enterprises in competition, ensuring the best business performance and customer satisfaction in a dynamic and complex market environment. Example

[0217] Figure 2 The regional alliance merchant data processing system of the present invention is provided, including a data collection engine module, a feature insight module, a rule mining module and a rule selection module;

[0218] The data collection engine module automatically collects the vehicle parts sales data provided by each alliance member, and uses data mining technology to classify the data to identify high-profit long-tail auto parts, and sends the identification results to the feature insight module;

[0219] The feature insight module extracts key features from the recognition results and applies cluster analysis algorithms to further segment the market and product characteristics based on the sales data of accessories and consumer purchase patterns, so as to identify the purchase target groups that are not easily noticed but have potential value in the market, and send the determined purchase target groups to the rule mining module;

[0220] The rule mining module uses the association rule learning algorithm to extract the accessory purchase rule information from the transaction data of the identified purchase target group, analyzes the purchase pattern and user behavior association between different vehicle accessories, obtains the association rules, and sends the association rules to the rule selection module;

[0221] The rule selection module combines nonlinear analysis and network theory to further analyze the temporal stability and sensitivity to market changes of each association rule, and evaluates and screens the association rules.

[0222] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.

[0223] The above technical solution was originally designed for the vehicle parts sales network in the modern automobile industry, but its principles and methods have wide applicability and can be extended to other different industries. This method is particularly suitable for implementation in industries that require regional merchant alliances and traffic sharing, such as retail, consumer electronics, home decoration, etc. By sharing resources and customer data, merchant alliances in various industries can similarly improve efficiency and market competitiveness, and achieve real-time adjustment of inventory optimization and marketing strategies. Therefore, this technical solution is not limited to the automotive industry, and its core concepts and processing mechanisms can be adjusted and applied according to the characteristics of specific industries to meet the dynamic changes of a wide range of market needs.

[0224] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or may be transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from one website, computer, server or data center to another website, computer, server or data center by wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more available media sets. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium may be a solid-state hard disk.

[0225] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0226] In the several embodiments provided in the present application, it should be understood that the disclosed systems and devices can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0227] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0228] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0229] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program codes.

[0230] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0231] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for processing regional alliance merchant data, characterized in that: S1: Automatically collect vehicle parts sales data provided by each alliance member and classify the data using data mining technology to identify high-profit long-tail auto parts; S2: Extract key features from the recognition results and apply cluster analysis algorithms to further segment the market and product characteristics based on accessory sales data and consumer purchase patterns to identify the target purchase groups that are not easily noticed but have potential value in the market; S3: Use association rule learning algorithms to extract accessory purchase rule information from the transaction data of the identified purchase target group, analyze the purchase patterns and user behavior associations between different vehicle accessories, and obtain association rules; S4: Combining nonlinear analysis and network theory, further analyze the temporal stability and sensitivity of each association rule to market changes, and evaluate and screen the association rules; S2 includes the following: S2-1, collecting classified vehicle parts data from step S1, i.e., vehicle parts identified as high-profit long-tail parts; According to business needs and preliminary analysis results, statistically analyze the characteristics that affect market segmentation; S2-2, calculate the feature information value, feature interaction diagram and network clustering coefficient of each feature; The feature information value is the ability of a feature to reduce system uncertainty, which is quantified using the concept of conditional entropy: ; in is the entropy of the target variable, is a given feature conditions The conditional entropy of the target variable is is the condition observed in the dataset probability; Constructing a feature interaction diagram , where nodes represent features and edge weights are determined by the conditional mutual information between features Determined, reflects the amount of information shared by two features when the target variable is known: ; in It is a feature and The joint entropy of is the target variable The conditional joint entropy of the known case; Calculate the network clustering coefficient of each node feature in the feature interaction graph, which reflects the closeness of a feature to other features: ; in It is through the characteristics The number of directly connected characteristic pairs, Is with characteristics The number of connected nodes.

2. The regional alliance merchant data processing method according to claim 1, characterized in that: The comprehensive feature selection index is obtained by combining the feature information value, feature interaction diagram and network clustering coefficient ; Screen out the features whose comprehensive feature selection index is greater than or equal to the impact threshold, and identify these features as key features; S2-3, all key features were standardized before clustering; Use the DBSCAN algorithm with the epsilon and min_samples parameters to identify natural clusters in the data; Use internal and external evaluation indicators to evaluate clustering quality; According to the clustering results, the target purchasing group is obtained.

3. The method for processing regional alliance merchant data according to claim 2, characterized in that: S3 includes the following: S3-1, for the purchase target group identified in step S2, collecting relevant transaction data from the purchase target group; S3-2, in association rule learning, each purchase session is regarded as a transaction, each purchased accessory is regarded as an item in the transaction, and the data is organized into a transaction list according to the accessory ID and the purchase session; Use the Apriori algorithm to find frequent item sets from transaction data, that is, accessory combinations that often appear in the same purchase session; Generate association rules from frequent itemsets; Filter out association rules based on confidence and lift thresholds.

4. The method for processing regional alliance merchant data according to claim 3, characterized in that: S4 includes the following: Nonlinear analysis and network theory are combined to obtain the time-dependence spectrum index and the market impact network complexity index.

5. The regional alliance merchant data processing method according to claim 4 is characterized by: The calculation process of the time-dependent spectral index is as follows: For each association rule, collect purchase records within a certain period of time and record whether the corresponding rule is triggered at each time point; Perform Fourier transform on the time series data of each rule to convert the time series from the time domain to the frequency domain. The result is expressed as the amplitude and phase of different frequency components. ; in is time series data, is the frequency The complex result of contains amplitude and phase information; Compute the time-dependent spectral index: ; in is the frequency The amplitude of represents the intensity of the frequency component; is the corresponding phase spectrum, is the highest frequency analyzed.

6. The method for processing regional alliance merchant data according to claim 5, characterized in that: The process of obtaining the market impact network complexity index is as follows: Collect market variable data and purchase behavior data related to each association rule; Construct a network where nodes include market variables and accessories in association rules. The edge weights between each pair of nodes are calculated by the conditional mutual information between them, indicating the reduction in uncertainty of the state of one node when the state of another node is: ; in and are any two nodes in the network, and be it given market conditions or other relevant factors; Compute the Laplacian matrix from the adjacency matrix of a network , defined as ,in is the degree matrix, is the adjacency matrix; Calculate the Market Impact Network Complexity Index: ; in are the non-zero eigenvalues ​​of the Laplacian matrix, is the number of nodes in the network.

7. The method for processing regional alliance merchant data according to claim 6, characterized in that: The market dynamic adaptability coefficient is obtained by comprehensively analyzing the time-dependent spectrum index and the market impact network complexity index. , the calculation formula is: ; in: is the time-dependent spectral index; is the market impact network complexity index; and are the weight factors of the time-dependency spectrum index and the market impact network complexity index, respectively; is the variance of the time-dependence spectrum index and the market impact network complexity index, indicating the magnitude of changes in these two indices under different market conditions; is the covariance of the time-dependency spectrum index and the market impact network complexity index; is a small constant used to ensure that the denominator is not zero.

8. The method for processing regional alliance merchant data according to claim 7, characterized in that: Compare the market dynamics adaptability coefficient with the value threshold; If the market dynamic adaptability coefficient is greater than or equal to the value threshold, it indicates that the association rule is very suitable for the current market environment, generating a signal for enhanced implementation; On the contrary, if the market dynamic adaptability coefficient is less than the value threshold, it indicates that the association rule is not stable enough in time and generates a cautious assessment signal.

9. A regional alliance merchant data processing system, used to implement the regional alliance merchant data processing method according to any one of claims 1 to 8, comprising a data collection engine module, a feature insight module, a rule mining module and a rule selection module; The data collection engine module automatically collects the vehicle parts sales data provided by each alliance member, and uses data mining technology to classify the data to identify high-profit long-tail auto parts, and sends the identification results to the feature insight module; The feature insight module extracts key features from the recognition results and applies cluster analysis algorithms to further segment the market and product characteristics based on the sales data of accessories and consumer purchase patterns, so as to identify the purchase target groups that are not easily noticed but have potential value in the market, and send the determined purchase target groups to the rule mining module; The rule mining module uses the association rule learning algorithm to extract the accessory purchase rule information from the transaction data of the identified purchase target group, analyzes the purchase pattern and user behavior association between different vehicle accessories, obtains the association rules, and sends the association rules to the rule selection module; The rule selection module combines nonlinear analysis and network theory to further analyze the temporal stability and sensitivity to market changes of each association rule, and evaluates and screens the association rules.