Cross-border e-commerce data processing method

Abstract

The invention relates to the field of e-commerce, in particular to a cross-border e-commerce data processing method. The method comprises the following steps: collecting and preprocessing an original data source, constructing a multilayer data source matrix, dynamically adjusting the weight of the data source, carrying out nonlinear conversion on the data source based on the optimized weight of the data source, and integrating to generate a hierarchical aggregation matrix; normalizing the hierarchical aggregation matrix to obtain a standard normalized matrix; and on the basis of the standard normalization matrix, clustering processing is carried out on the e-commerce data through a multi-level and multi-region constraint K-means clustering method, and a clustering result is obtained. The problems that a traditional cross-border e-commerce data processing method lacks comprehensive consideration of multi-level and multi-region data sources, and the overall analysis result is not comprehensive enough are solved; the adopted clustering algorithm cannot fully consider the feature difference of different regions, so that the unique demand mode of each country or region is difficult to accurately identify; the response to the dynamic change of real-time data is weak, and the market change and fluctuation are difficult to adapt in real time.

Description

Technical Field

[0001] The present invention relates to the field of e-commerce, and particularly to a method for processing cross-border e-commerce data. Background Art

[0002] With the development of cross-border e-commerce, e-commerce platforms need to process and analyze a large amount of data from different countries, regions and levels. The data includes multi-dimensional data sources such as orders, inventory, and user behavior. Currently, many cross-border e-commerce platforms optimize inventory management and marketing strategies through data analysis, and analyze market demand trends by combining user behavior characteristics. The processing of the multi-source data can help cross-border e-commerce platforms more accurately grasp the consumption dynamics in different regions and time periods in the international market, improve operation efficiency, and provide support for the global layout of cross-border e-commerce platforms.

[0003] However, there are still some problems in the application of the above cross-border e-commerce data processing method: First, there is a lack of comprehensive consideration of multi-level and multi-region data sources, and it is unable to effectively integrate information at different levels such as orders, inventory, and user behavior, resulting in an incomplete overall analysis result; Second, the clustering algorithm adopted fails to fully consider the characteristic differences of different regions, and it is difficult to accurately identify the unique demand patterns of each country or region, resulting in a lag in resource allocation and market response; In addition, the response to the dynamic changes of real-time data is weak, and it is unable to flexibly adjust data weights and feature consistency, thus making it difficult to adapt to market changes and fluctuations in real time. Summary of the Invention

[0004] The present invention provides a method for processing cross-border e-commerce data to solve the problems that the traditional cross-border e-commerce data processing method lacks comprehensive consideration of multi-level and multi-region data sources, is unable to effectively integrate information at different levels such as orders, inventory, and user behavior, resulting in an incomplete overall analysis result; the clustering algorithm adopted fails to fully consider the characteristic differences of different regions, and it is difficult to accurately identify the unique demand patterns of each country or region, resulting in a lag in resource allocation and market response; the response to the dynamic changes of real-time data is weak, and it is unable to flexibly adjust data weights and feature consistency, thus making it difficult to adapt to market changes and fluctuations in real time.

[0005] A method for processing cross-border e-commerce data according to the present invention specifically includes the following technical solutions:

[0006] A method for processing cross-border e-commerce data includes the following steps:

[0007] S1: Collect the original data sources and preprocess them to obtain the data sources; based on the data sources, construct a multi-layer data source matrix; through a weight optimization method based on the volatility and time series changes of the data sources, dynamically adjust the weights of the data sources to obtain the optimized data source weights; based on the optimized data source weights, perform a non-linear transformation on the data sources and integrate them to generate a hierarchical aggregation matrix; perform a normalization process on the hierarchical aggregation matrix to obtain a standard normalized matrix;

[0008] S2: Based on the standard normalized matrix, perform clustering on the e-commerce data through a K-means clustering method with multi-level and multi-region constraints to obtain a clustering result.

[0009] Preferably, S1 specifically includes:

[0010] In the implementation process of the weight optimization method based on the volatility and time series changes of the data sources, based on the volatility and the amplitude of time series changes of the data sources, dynamically adjust the weights of the data sources to obtain the optimized data source weights. The specific formula is as follows:

[0011]

[0012] where w′ i is the optimized weight of the i-th data source; i is the data source index; w i is the initial weight of the i-th data source; α is a regulation coefficient used to determine the influence intensity of volatility and the amplitude of time series changes in weight adjustment; σ(D i ) is the standard deviation of the i-th data source D i ; is the standard deviation of the -th data source ; ΔD i is the amplitude of time series change of the i-th data source Di; is the amplitude of time series change of the -th data source ; n is the total number of data sources; is the data source index.

[0013] Preferably, S1 specifically includes:

[0014] Based on the optimized data source weights, use a kernel function to perform a non-linear transformation on the data sources to obtain the kernel function output of the data sources.

[0015] Preferably, S1 specifically includes:

[0016] Integrate the kernel function outputs of each data source according to the hierarchical structure to generate a hierarchical aggregation matrix.

[0017] Preferably, S2 specifically includes:

[0018] In the implementation process of the multi-level and multi-region constrained K-means clustering method, each column of the standard normalization matrix is used as a sample, and initial clustering centers are selected. Each initial clustering center includes global features, hierarchical features, and regional features. When allocating samples, calculate the distances from the samples to each clustering center, introduce a cross-layer region consistent clustering function, calculate the clustering objective function value to measure the matching degree between the samples and the clustering centers, and allocate the samples to the cluster represented by the clustering center with the minimum clustering objective function value. After the allocation is completed, update the means of the global features, hierarchical features, and regional features of the samples in the cluster.

[0019] Preferably, S2 specifically includes:

[0020] The cross-layer region consistent clustering function clusters multi-level data by optimizing the global feature similarity, intra-level consistency, and cross-region consistency of the samples.

[0021] Preferably, S2 specifically includes:

[0022] In the cross-layer region consistent clustering function, introduce a global feature similarity term, a hierarchical feature term, and a regional feature term, and calculate the clustering objective function value. The specific formula is as follows:

[0023]

[0024] Where, is the clustering objective function value; K is the number of clusters after partitioning; k is the cluster index; j ∈ C k represents the j-th sample assigned to the k-th cluster; C k is the sample set of the k-th cluster; j is the sample index; ||M j - μ k || 2 is the global feature similarity term; M j is the j-th sample, representing the j-th column in the standard normalization matrix; μk is the central feature vector of the k-th cluster; is the hierarchical feature term; λ is the hierarchical difference adjustment coefficient; L is the number of levels; l is the level index; represents the index of the data source belonging to the l-th level; L l is the data source set of the l-th level; is the standard normalization feature value of the j-th sample on the -th data source; is the central feature value of the samples in the k-th cluster on the -th data source; is the regional feature term; is the regional difference adjustment coefficient; M is the total number of different countries or regions in the cross-border e-commerce platform; m is the country or region index; is an indicator function; R m is the sample set of the m-th country or region; μ k (m) is the central feature vector of the samples in the k-th cluster in the m-th country or region.

[0025] Preferably, the S2 specifically includes:

[0026] During the clustering process, set the convergence threshold and the maximum number of iterations. When the change in the Euclidean distance of all cluster centers between two consecutive iterations is less than the set convergence threshold, or when the maximum number of iterations is reached, stop the iteration to obtain the clustering result; analyze the clustering result to optimize the inventory and marketing strategies.

[0027] The beneficial effects of the technical solution of the present invention are:

[0028] 1. By introducing a cross-layer region-consistent clustering function, the present invention realizes the clustering analysis of multi-level and multi-region data sources, enabling the cross-border e-commerce platform to more comprehensively grasp the demand characteristics of different markets and time periods; in the clustering of multi-level data such as orders, inventory, and user behavior, by optimizing the overall characteristics, hierarchical characteristics, and regional characteristics, the cross-border e-commerce platform is more accurate in resource allocation, can quickly identify high-demand regions and time periods, and arrange replenishment in a timely manner.

[0029] 2. By introducing a weight optimization method based on the volatility of data sources and the amplitude of temporal changes, the present invention can dynamically adjust the weights of different data sources, enabling data sources with greater timeliness and volatility to occupy higher weights in the analysis results, significantly improving the accuracy and real-time nature of data analysis, and helping the cross-border e-commerce platform to respond in a timely manner to changes in market demand. Especially during peak hours or promotional activities, it can quickly adapt to the dynamic market demand.

[0030] 3. The present invention introduces a K-means clustering method with multi-level and multi-region constraints, incorporating the characteristic differences of different countries and regions into the clustering analysis to ensure the consistency of characteristics within the same cluster at different levels and regions, helping the cross-border e-commerce platform to more accurately identify the demand patterns of different global markets, optimize the supply chain management and market promotion strategies, achieve the optimal allocation of resources in the global market, and improve the operation efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 is a flowchart of a cross-border e-commerce data processing method according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0032] In order to further elaborate on the technical means and effects adopted by the present invention to achieve the intended invention purpose, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0033] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0034] The following specifically describes the specific solution of a cross-border e-commerce data processing method provided by the present invention in conjunction with the accompanying drawings.

[0035] Refer to the attached Figure 1 , which shows a flowchart of a cross-border e-commerce data processing method provided by an embodiment of the present invention. The method includes the following steps:

[0036] S1: Collect the original data source and perform preprocessing to obtain the data source; based on the data source, construct a multi-layer data source matrix, and dynamically adjust the data source weight to obtain the optimized data source weight; based on the optimized data source weight, perform non-linear transformation on the data source, and integrate and generate a hierarchical aggregation matrix; perform normalization processing on the hierarchical aggregation matrix to obtain a standard normalization matrix;

[0037] The cross-border e-commerce platform collects the original data source through channels such as the background system, user front-end interface, and inventory management system, and performs preprocessing on the original data source to ensure the accuracy and consistency of the data, and obtains the data source; the preprocessing includes cleaning, deduplication, and formatting, which are well-known means to those skilled in the art and will not be elaborated here.

[0038] Furthermore, the data source is stratified according to the actual use and characteristics to form a multi-layer data source matrix; the stratification process is to divide the data source into different layers. For example, the data source is divided into order data source, inventory data source, and user behavior data source; the order data source includes information such as order volume and sales amount, which directly reflects the sales demand of the platform; the inventory data source includes information such as the quantity of goods in stock and replenishment time; the user behavior data source includes information such as the browsing record, add-to-cart record, favorite record, and search preference of the user; the data sources belonging to the same layer are arranged adjacent to each other to form a multi-layer data source matrix; the multi-layer data source matrix is composed of data sources at different layers, such as the data source storing the order volume, the data source storing the destination country, the data source storing the quantity of goods in stock, etc.;

[0039] The data source expression is D i(i = 1, 2, ..., n), where D i represents the i-th data source, represented in the form of a vector. The same column of different data source vectors represents the sample data collected from the same time point, that is, the specific data such as the order volume, sales amount, and commodity inventory quantity corresponding to this data point; i is the data source index and also the row vector of the multi-layer data source matrix; n is the total number of data sources, and the total number of data sources is set according to the specific implementation scenario.

[0040] To optimize the utilization effect of each data source in the multi-layer data source matrix and reflect the dynamic changes of each data source at different times, the present invention adopts a weight optimization method based on the volatility and time series changes of the data source. According to the real-time fluctuation situation and change trend of each data source, the weight of the corresponding data source is dynamically adjusted, so that in subsequent data analysis and prediction, the data source information with higher timeliness and volatility occupies a greater weight, effectively improving the accuracy of subsequent processing.

[0041] Based on the volatility and the amplitude of time series changes of the data source, the weight of the data source is dynamically adjusted to obtain the optimized weight of the data source, so that the weight of each data source can more accurately reflect its real-time characteristics and the impact on platform operation, and the data source with significant changes can obtain a higher weight in analysis and prediction. The weight optimization formula is as follows:

[0042]

[0043] where, w′ i is the optimized weight of the i-th data source; i is the data source index; w i is the initial weight of the i-th data source, which is set according to the business requirements in the specific implementation scenario and the importance of the data source; α is the adjustment coefficient, which is used to determine the influence intensity of volatility and the amplitude of time series changes in weight adjustment, and is obtained through experimental optimization; σ(D i ) is the standard deviation of the i-th data source D i ; is the standard deviation of the -th data i ; ΔD i is the amplitude of time series change of the i-th data source Di, which represents the total change amount of the data source within the current time period, and is obtained by calculating the difference between the first data point (i.e., the starting value) and the last data point (i.e., the latest value) of the i-th data source D is the amplitude of time series change of the -th data source; n is the total number of data sources; is the data source index;

[0044] The weight optimization formula, through a dynamic adjustment mechanism, makes the weight of each data source not only depend on the initial weight but also be affected by the volatility and temporal variation amplitude of the data source, so as to better reflect the characteristics of real-time data and is suitable for dynamic data analysis and prediction scenarios.

[0045] To further improve the integration effect of data features, after weight optimization, a kernel function is used to perform a non-linear transformation on each data source to enhance the flexibility and accuracy of feature expression;

[0046] The kernel function formula is as follows:

[0047]

[0048] Among them, K(D i , w′ i ) is the output of the kernel function, represented in vector form; i is the data source index; D i represents the i-th data source; w′ i is the optimized weight of the i-th data source; exp is the exponential function, used for non-linearly scaling the data source, so that data sources with a large deviation from the mean are attenuated to enhance the expressiveness of data features; μ(D i ) is the mean of the i-th data source D i , which is used as the central value of the kernel function to calculate the deviation of the data source from the center; ||D i - μ(D i )|| 2 is the squared Euclidean distance between the i-th data source Di and the mean μ(D i ) of the i-th data source D i , calculated by summing the squares of the deviations of each data point of the i-th data source D i from the mean μ(D i ) of the i-th data source D i to measure the overall deviation of all elements in the data source from the data source mean; σ 2 (Di) is the variance of the i-th data source Di;

[0049] The kernel function introduces the optimized data source weight, dynamically adjusts the influence of the data source in the hierarchical aggregation matrix, so that the hierarchical aggregation matrix can more effectively retain the features of multi-layer data sources and meet the requirements for data feature integration in cross-border e-commerce scenarios.

[0050] The hierarchical aggregation matrix is generated by integrating the outputs of the kernel functions of each data source according to a hierarchical structure, and the expression is M layered = [K(D 1 , w′ 1 ), K(D 2 , w′ 2),...,K(D n ,w′ n )] T , where T is the transpose symbol.

[0051] Normalize the hierarchical aggregation matrix to generate a standard normalized matrix to ensure data consistency. The specific formula is:

[0052]

[0053] where M is the standard normalized matrix, representing the standardized aggregation result of the multi-layer data source. The row vectors of the standard normalized matrix are standard normalized eigenvectors; ln is the natural logarithm function; M layered is the hierarchical aggregation matrix; μ(M layered ) is the mean of the hierarchical aggregation matrix; σ 2 (M layered ) is the variance of the hierarchical aggregation matrix, which is used to scale the deviation during the normalization process to suppress the influence brought by data volatility; ∈ is a constant used to prevent the denominator from being zero and is set according to the specific implementation scenario;

[0054] Through the normalization formula, the features of each layer of data source are made consistent, and the features of the multi-layer data source are scaled to the same scale, thus ensuring data consistency, laying a unified data foundation for the subsequent clustering and prediction steps, and helping to improve the accuracy of data analysis.

[0055] S2: Based on the standard normalized matrix, use the K-means clustering method with multi-level and multi-region constraints to cluster the e-commerce data to obtain the clustering result.

[0056] Take each column of the standard normalized matrix as a sample and input it into the clustering algorithm. The clustering algorithm uses the K-means clustering method with multi-level and multi-region constraints to implement cross-level and multi-region data clustering analysis through the cross-layer region consistent clustering function.

[0057] The K-means clustering method with multi-level and multi-region constraints is a data clustering algorithm suitable for cross-border e-commerce. On the basis of the traditional K-means algorithm, the cross-layer region consistent clustering function (CRCC) is introduced to optimize the similarity of samples in overall features, ensure the feature consistency within each level (such as orders, inventory, user behavior), and the clustering rationality between different regions. The cross-layer region consistent clustering function (CRCC) is a clustering objective function designed specifically for the cross-border e-commerce scenario. By optimizing the overall feature similarity, intra-level consistency, and cross-region consistency, it clusters multi-level data such as orders, inventory, and user behavior, so that the features within the same cluster are highly consistent at different levels and regions, thus meeting the multi-dimensional needs of cross-border e-commerce.

[0058] Specifically, K samples are selected as the initial clustering centers according to the expert experience method. Each initial clustering center includes overall features, hierarchical features, and regional features. When allocating samples, calculate the distance from each sample to each clustering center, and use the cross-layer region consistent clustering function to measure the matching degree between the sample and the clustering center through the clustering objective function value. Finally, allocate the sample to the cluster represented by the clustering center with the smallest clustering objective function value; after the allocation is completed, update the mean values of the overall features, hierarchical features, and regional features of the samples in the cluster.

[0059] The cross-layer region consistent clustering function is as follows:

[0060]

[0061] Where, is the clustering objective function value; K is the number of clusters after partitioning, which is set according to specific business requirements or experience; k is the clustering index; j ∈ C k represents the j-th sample assigned to the k-th cluster; C k is the sample set of the k-th cluster, which contains all samples assigned to the k-th cluster; j is the sample index; ||M j -μ k || 2 is the overall feature similarity term, which is used to ensure that the samples within the same cluster are similar in overall features; M j is the j-th sample, that is, the j-th column in the standard normalization matrix; μ k is the central feature vector of the k-th cluster, which is the average feature vector of all samples in the sample set of the k-th cluster; is the hierarchical feature term, which is used to ensure the feature consistency of each level (such as order, inventory, user behavior) within the cluster; λ is the hierarchical difference adjustment coefficient, which is used to adjust the weight of the hierarchical feature term in the cross-layer region consistent clustering function to balance the consistency between different levels, and is obtained through experimental tuning; L is the number of levels, such as order, inventory, and user behavior levels, which is set according to the data source levels in the specific implementation scenario; l is the level index; represents the index of the data source belonging to the l-th level; L l is the data source set of the l-th layer, such as the order level includes two data sources: order quantity and sales amount; is the standard normalized feature value of the j-th sample on the -th data source; is the central feature value of the samples in the k-th cluster on the -th data source, that is, the average feature vector; It is a regional feature item, which is used to ensure that the sample features of each country or region within a cluster are similar, so as to ensure the rationality of data distribution in different countries or regions, and help to identify demand patterns globally; It is a regional difference adjustment coefficient, which is used to adjust the weight of the consistency of features in different countries or regions in the cross-layer regional consistent clustering function, and is obtained through experimental optimization; M is the total number of different countries or regions in the cross-border e-commerce platform, which is set according to the number of countries or regions in the specific implementation scenario market or platform; m is the country or region index, which is used to identify a specific country or region; It is an indicator function, which takes the value of 1 only when j ∈ R m and 0 otherwise; R m is the sample set of the m-th country or region; μ k (m) is the central feature vector of the samples of the k-th cluster in the m-th country or region. In the m-th country or region, the mean value of the samples belonging to the k-th cluster is taken.

[0062] Set the convergence threshold and the maximum number of iterations according to the expert experience method. When the change amount of the Euclidean distance of all cluster centers between two consecutive iterations is less than the set convergence threshold, or when the maximum number of iterations is reached, stop the iteration to obtain the clustering result.

[0063] Next, analyze each clustering result to clarify the demand characteristics of different markets and time periods, and further optimize the inventory and marketing strategies: replenishment can be arranged in advance for regions with high order volume and tight inventory; precise promotion plans and advertising placements can be formulated in countries where users are active to ensure efficient allocation of resources and meet market demands. At the same time, the clustering results help cross-border e-commerce to identify demand patterns in different regions of the world, and improve the accuracy and pertinence of supply chain management and market promotion. For example, the European and American markets with high order volume and sales, and active orders and user behaviors are suitable for holiday promotions and require key stockpiling; the Japanese and Korean markets with stable order volume and inventory have stable demands and are suitable for regular inventory maintenance; the Southeast Asian markets with low order volume but high page views, low inventory and frequent replenishment are suitable for increasing promotional activities and inventory investment.

[0064] In summary, a cross-border e-commerce data processing method is completed.

[0065] The sequence of the invention embodiments is only for description and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0066] Each embodiment in this specification is described in a progressive manner. For the same or similar parts among the embodiments, reference can be made to each other, and the differences between each embodiment and other embodiments are emphasized.

[0067] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features. These modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of each embodiment of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A cross-border e-commerce data processing method, characterized in that: The following steps are involved: S1: collect the original data source and preprocess it to obtain the data source; Based on the data source, build a multi-layer data source matrix; By using a weight optimization method based on data source volatility and time series changes, the data source weight is dynamically adjusted to obtain the optimized data source weight; Based on the optimized data source weights, the data sources are nonlinearly transformed and integrated to generate a hierarchical aggregation matrix; Normalize the hierarchical aggregation matrix to obtain a standard normalized matrix; S2: Based on the standard normalized matrix, the e-commerce data is clustered through the multi-level and multi-region constrained K-means clustering method to obtain the clustering results.

2. A cross-border e-commerce data processing method according to claim 1, characterized in that: The S1 specifically includes: In the process of implementing the weight optimization method based on data source volatility and time series changes, the data source weight is dynamically adjusted based on the volatility and time series change amplitude of the data source to obtain the optimized data source weight. The specific formula is as follows: Among them, w′ i is the optimized weight of the i-th data source; i is the data source index; w i is the initial weight of the ith data source; α is the adjustment coefficient used to determine the impact of volatility and time series variation in weight adjustment; σ(D i ) is the i-th data source D i The standard deviation of It is Data sources The standard deviation of i is the temporal variation amplitude of the i-th data source Di; It is Data sources The temporal variation of; n is the total number of data sources; Is the data source index.

3. A cross-border e-commerce data processing method according to claim 2, characterized in that: The S1 specifically includes: Based on the optimized data source weight, the kernel function is used to perform nonlinear transformation on the data source to obtain the kernel function output of the data source.

4. A cross-border e-commerce data processing method according to claim 3, characterized in that: The S1 specifically includes: The kernel function output of each data source is integrated hierarchically to generate a hierarchical aggregation matrix.

5. A cross-border e-commerce data processing method according to claim 1, characterized in that: The S2 specifically includes: In the implementation process of the multi-level and multi-region constrained K-means clustering method, each column of the standard normalized matrix is ​​taken as a sample, and an initial cluster center is selected. Each initial cluster center includes overall features, hierarchical features and regional features. When allocating samples, the distance from the sample to each cluster center is calculated, and a cross-layer regional consistent clustering function is introduced to calculate the clustering objective function value, measure the matching degree between the sample and the cluster center, and allocate the sample to the cluster represented by the cluster center with the smallest clustering objective function value. After the allocation is completed, the overall features, hierarchical features and regional features of the samples in the cluster are updated by taking the mean.

6. A cross-border e-commerce data processing method according to claim 5, characterized in that: The S2 specifically includes: The cross-layer regional consistent clustering function clusters multi-layer data by optimizing the overall feature similarity, internal consistency of the layer, and cross-region consistency of the samples.

7. A cross-border e-commerce data processing method according to claim 6, characterized in that: The S2 specifically includes: In the cross-layer regional consistent clustering function, the overall feature similarity term, the hierarchical feature term and the regional feature term are introduced to calculate the clustering objective function value. The specific formula is as follows: in, is the clustering objective function value; K is the number of clusters after division; k is the cluster index; j∈C k is the jth sample assigned to the kth cluster; C k is the sample set of the kth cluster; j is the sample index; ||M j -μ k || 2 is the overall feature similarity term; M j is the jth sample, representing the jth column in the standard normalized matrix; μ k is the central eigenvector of the kth cluster; is the hierarchical feature term; λ is the hierarchical difference adjustment coefficient; L is the number of levels; l is the hierarchical index; Indicates the index of the data source in the lth level; L l is the data source set of layer l; is the jth sample in Standard normalized feature values ​​on data sources; is the sample in the kth cluster in The central eigenvalue of each data source; is a regional characteristic item; is the regional difference adjustment coefficient; M is the total number of different countries or regions in the cross-border e-commerce platform; m is the country or region index; is the indicator function; R m is the sample set of the mth country or region; μ k (m) is the central feature vector of the samples of the kth cluster in the mth country or region.

8. A cross-border e-commerce data processing method according to claim 7, characterized in that: The S2 specifically includes: During the clustering process, the convergence threshold and the maximum number of iterations are set. When the change in the Euclidean distance of all cluster centers between two consecutive iterations is less than the set convergence threshold, or when the maximum number of iterations is reached, the iteration is stopped to obtain the clustering result. The clustering results are analyzed to optimize inventory and marketing strategies.

Images