A method for analyzing and evaluating e-commerce business data

Abstract

The present invention relates to the technical field of e-commerce analysis, and specifically to a method for analyzing and evaluating e-commerce operation data, including the following steps: based on the commodity classification information and commodity characteristic parameters in the e-commerce operation data, extracting the covariance matrix of commodity characteristic vectors, and calculating the eigenvalues and eigenvectors in the matrix. In the present invention, through the linear projection of the user demand vector and the calculation of the Euclidean distance, the matching relationship between the user demand and the commodity characteristics is accurately optimized. Based on the weight adjustment of the contribution rate of the cross-characteristic components, the adaptability of the data decomposition result to dynamic changes is enhanced, ensuring that the synergy between the characteristics is more accurate. By extracting the sparse distribution coefficient of the long-tail commodity characteristics and optimizing the weight allocation, the imbalance problem of long-tail resource allocation is improved, significantly enhancing the resource utilization efficiency. Dynamically adjusting the matching relationship of the key characteristic dimensions and combining with the calculation of the matching ratio, the real-time optimization of the relationship between the commodity and the demand is achieved, providing efficient support for decision-making.

Description

Technical Field

[0001] The present invention relates to the technical field of e-commerce analysis, and particularly to a method for analyzing and evaluating e-commerce operation data. Background Art

[0002] The technical field of e-commerce analysis includes methods and systems for collecting, processing, and analyzing data in e-commerce activities using computer and network technologies. The core content of this technical field is to reveal consumer behavior patterns, market trends, and transaction efficiency, etc. through the mining and analysis of e-commerce data, providing a scientific basis for business decisions. Its overall technical field covers data monitoring in the e-trading process, user behavior analysis, commodity sales forecasting, and data visualization, etc., and is widely used in optimizing operation strategies, enhancing customer experience, and improving resource allocation.

[0003] Among them, the method for analyzing and evaluating e-commerce operation data refers to a method for analyzing and evaluating the operation data on an e-commerce platform. This patent theme targets transaction data, user data, and operation data in the e-commerce platform, adopts specific data collection methods, data cleaning and classification methods, and processes and evaluates the data through quantitative and qualitative analysis methods. Its specific content includes using specific data models to statistically analyze indicators such as sales volume and user conversion rate, and dividing commodities, user groups, or time dimensions based on classification rules to complete multi-dimensional evaluation and induction of the operation status.

[0004] In the prior art, when dealing with high-dimensional characteristic data, there is a lack of efficient dimensionality reduction means, which easily generates computational complexity and redundant information, reducing the efficiency of data analysis. The matching relationship between user needs and commodity characteristics mainly relies on classification methods, lacking the ability to accurately map according to demand characteristics, resulting in a large error in the matching results. The processing method for cross-characteristic weights is relatively static, difficult to cope with dynamic characteristic changes, and limits the applicability of the characteristic analysis results. The weights of long-tail commodities are underestimated in the overall characteristic distribution, and the resource allocation cannot be refined, resulting in a low utilization rate of long-tail resources. The analysis and adjustment of matching degree records lack flexibility, and the impact of real-time data changes on the analysis results has not been effectively solved, limiting the accuracy and reliability of data analysis. Summary of the Invention

[0005] The purpose of the present invention is to solve the deficiencies existing in the prior art, and to propose a method for analyzing and evaluating e-commerce operation data.

[0006] To achieve the above purpose, the present invention adopts the following technical solutions: A method for analyzing and evaluating e-commerce operation data, including the following steps:

[0007] S1: Based on the commodity classification information and commodity characteristic parameters in the e-commerce operation data, extract the covariance matrix of commodity characteristic vectors, calculate the eigenvalues and eigenvectors in the matrix, sort the vectors according to the eigenvalues and complete dimensionality reduction, extract the main characteristic components, and generate the commodity principal component matrix;

[0008] S2: Extract the user purchase preferences in the e-commerce operation data, combine with the commodity principal component matrix, perform linear projection on the user demand vector according to the commodity characteristic dimension, calculate the Euclidean distance between the demand and the principal component vector, sort by the distance size and complete characteristic alignment, and generate the user demand alignment matrix;

[0009] S3: Statistically analyze the transaction cross-characteristic table in the e-commerce operation data, combine with the user demand alignment matrix, perform decomposition operation on the characteristic cross matrix, extract the component contribution rate and adjust the matrix weight, re-integrate the component matrix according to the adjustment result, complete the characteristic reconstruction operation, and generate the reconstructed characteristic matching matrix;

[0010] S4: Analyze the commodity long-tail distribution in the e-commerce operation data, combine with the reconstructed characteristic matching matrix, extract the characteristic weight sparse distribution coefficient in the matrix, analyze the long-tail commodity characteristic weight and increase the weight of the sparse characteristic dimension, optimize the long-tail characteristic distribution, and generate the long-tail characteristic optimization matrix;

[0011] S5: Statistically analyze the user-commodity matching degree records in the e-commerce operation data, combine with the long-tail characteristic optimization matrix, perform dynamic adjustment operation on the key characteristic dimension matching components in the matrix, calculate the matching ratio between the commodity and the demand and integrate the matching values, and generate the commodity demand characteristic matching ratio value.

[0012] As a further solution of the present invention, the commodity principal component matrix includes main characteristic components, eigenvalue vectors, and eigenvalue sorting; the user demand alignment matrix includes user demand vectors, linear projection results, and Euclidean distances; the reconstructed characteristic matching matrix includes characteristic cross matrix components, component contribution rates, and weight adjustment results; the long-tail characteristic optimization matrix includes characteristic weight sparse distribution coefficients, long-tail commodity characteristic weights, and sparse characteristic dimension gain weights; the commodity demand characteristic matching ratio value includes key characteristic dimension matching components, dynamic adjustment results, and commodity demand matching ratios.

[0013] As a further solution of the present invention, the specific steps for obtaining the commodity principal component matrix are as follows:

[0014] S101: Based on the product classification information and product characteristic parameters in e-commerce operation data, construct a product characteristic vector space, call the hierarchical structure parameters in the product classification information, screen the classification nodes with independent characteristic descriptions within the product classification hierarchy, match the product characteristic parameters under the classification nodes, eliminate the redundant characteristic items caused by classification intersections, establish a set of characteristic parameters, adjust the data scale according to the value range, data type and comparability of the product characteristic parameters, and use normalization transformation to unify the data scale to construct a product characteristic covariance matrix;

[0015] S102: Perform eigenvalue decomposition on the product characteristic covariance matrix, calculate the eigenvalues and eigenvectors of the covariance matrix, call the matrix decomposition result, sort them in descending order according to the eigenvalue size, calculate the eigenvalue contribution rate based on the eigenvalue distribution, set a contribution rate threshold, screen the eigenvalues whose contribution rate exceeds the threshold, and extract the corresponding eigenvectors to form an eigenvalue sorting vector matrix;

[0016] S103: Perform dimensionality reduction processing on the eigenvalue sorting vector matrix, set a cumulative contribution rate threshold, retain the first k eigenvectors whose contribution rate reaches the set value, and use the formula:

[0017] ;

[0018] Calculate the main components of product characteristics and establish a product main component matrix;

[0019] Among them, represents the product main component matrix, represents the i-th eigenvalue of the covariance matrix, represents the c-th eigenvector of the covariance matrix, represents the number of selected main components, is a smoothing adjustment parameter to ensure the numerical stability of the dimensionality reduction result.

[0020] As a further solution of the present invention, the steps for obtaining the user demand alignment matrix are specifically as follows:

[0021] S201: Extract user purchase records and preferences from e-commerce operation data, screen the product categories of user purchase records, call the product characteristic parameters of the product main component matrix, match the user-purchased products according to the product classification labels, and extract user demand vectors consistent with the characteristic dimensions of the product main component matrix;

[0022] S202: Use the commodity principal component matrix to perform a linear projection on the user demand vector, call the principal component weights of the commodity characteristic parameters, and according to the linear mapping rule of the projection matrix, transform the user demand vector on the commodity principal component matrix, calculate the projection coordinates, screen the eigenvectors that deviate from the central distribution in the projection results, adjust the user demand weights, and obtain the user demand principal component vector;

[0023] S203: Calculate the Euclidean distance between the user demand principal component vector and the commodity principal component matrix, using the formula:

[0024] ;

[0025] Obtain the distance sorting result;

[0026] where the distance between the user demand principal component vector and the a-th principal component vector of the commodity principal component matrix is The component of the user demand principal component vector on the j-th characteristic dimension is The component of the commodity principal component matrix on the j-th characteristic dimension of the a-th principal component is , represents the weight coefficient of the j-th characteristic dimension, adjusts the influence degree of the differential characteristic dimension, and corrects the parameter is used to avoid calculation errors caused by too small values. After calculating all distances, sort them in ascending order. n represents the total number of characteristic dimensions;

[0027] S204: Call the distance sorting result, perform characteristic alignment according to the sorting order, screen the principal component vector closest to the user demand preference, adjust the proportion between characteristic dimensions, and generate a user demand alignment matrix.

[0028] As a further solution of the present invention, the steps for obtaining the reconstructed characteristic matching matrix are specifically as follows:

[0029] S301: Statistically analyze the transaction cross-characteristic table in the e-commerce operation data, extract the correlation parameters between differential transaction characteristics, call the user demand alignment matrix, and construct a characteristic correlation matrix according to the co-occurrence relationship of commodities in the transaction records. Screen the commodity characteristic items according to the co-occurrence frequency, calculate the correlation weights between transaction characteristics, and establish a characteristic cross matrix;

[0030] S302: Perform a decomposition operation on the characteristic cross matrix, extract the principal components of multiple characteristic dimensions, calculate the component contribution rate, call the principal component weight vector, sort according to the transaction characteristic contribution rate, screen the characteristic components with a contribution rate lower than the threshold, and recalculate the contribution weights of the remaining characteristic items to obtain a weight adjustment matrix;

[0031] S303: Reconstruct the feature cross matrix based on the weight adjustment matrix, calculate the weight of the adjusted feature matrix, using the formula:

[0032] ;

[0033] Calculate the adjusted feature matrix to obtain the reconstructed feature matrix;

[0034] Among them, represents the feature weight value of the j-th feature dimension in the z-th row of the adjusted feature matrix, represents the feature value at the corresponding position in the original feature cross matrix, Calculate the normalized proportion of the current feature dimension in the adjusted matrix, is a smoothing adjustment parameter to avoid a zero denominator;

[0035] S304: Re-integrate the matrix structure according to the feature components of the reconstructed feature matrix, call the correlation parameters between trading features, calculate the interaction intensity of the feature dimensions, screen the feature components with high interaction intensity, optimize the matching rules of the feature components, adjust the feature alignment method, and generate a reconstructed feature matching matrix.

[0036] As a further solution of the present invention, the steps for obtaining the long-tail feature optimization matrix are specifically as follows:

[0037] S401: Analyze the long-tail distribution of commodities in e-commerce operation data, call the sales frequency of commodities in the transaction records, calculate the distribution of commodity sales volume, screen the commodities with long-term low-frequency sales volume, extract the feature components of the low-frequency commodities in the reconstructed feature matching matrix, calculate the weight value of the long-tail commodities in the feature dimension, and establish a long-tail commodity feature weight matrix;

[0038] S402: Calculate the sparse distribution coefficient of the long-tail commodity feature weight matrix, call the feature weight vector, calculate the weight proportion of multiple feature dimensions, screen the feature dimensions with low weight proportion according to the discreteness of the feature weights, extract the feature components with the lowest weight proportion, calculate the average weight between the feature dimensions, and call the feature dimension weight data below the average value to construct a sparse feature weight matrix;

[0039] S403: Perform gain weight adjustment on the feature dimensions in the sparse feature weight matrix, using the formula:

[0040] ;

[0041] Calculate to obtain the gain-adjusted feature matrix;

[0042] Among them, represents the feature weight after gain adjustment, represents the original sparse feature weight, is the gain adjustment factor for adjusting the proportion of low-weight characteristics Calculate the normalized total weight of the current characteristic dimension is the smoothing parameter to avoid the interference of extreme data in the calculation;

[0043] S404: Optimize the long-tail characteristic distribution according to the gain adjustment characteristic matrix, call the characteristic weight values of long-tail products, screen the characteristic weights after gain adjustment, extract the characteristic components with the adjustment amplitude exceeding the set threshold, adjust the weight proportion of low-weight characteristics in long-tail products, optimize the characteristic structure of long-tail products, and generate a long-tail characteristic optimization matrix.

[0044] As a further solution of the present invention, the steps for obtaining the commodity demand characteristic matching ratio value are specifically as follows:

[0045] S501: Statistically record the user-commodity matching degree in the e-commerce operation data, call the user transaction data, select the matching items for the number of successful transactions and evaluations, extract the characteristic vectors of the corresponding commodities, call the long-tail characteristic optimization matrix, screen the optimized characteristic matching parameters, calculate the matching degree between the user demand vector and the commodity characteristic vector, and establish a commodity characteristic matching matrix;

[0046] S502: Call the key characteristic dimensions of the commodity characteristic matching matrix, dynamically adjust the matching components in the matrix, calculate the contribution weights of user demands in multiple characteristic dimensions, adjust the matching characteristic weights according to user preferences, screen the characteristic dimensions with high contribution weights, call the adjusted matching matrix, and perform a normalization transformation on the matching weights to obtain a matching characteristic adjustment matrix;

[0047] S503: Calculate the matching ratio between the commodity and the demand, call the characteristic matching components in the matching characteristic adjustment matrix, calculate the contribution proportion of the commodity in multiple characteristic dimensions based on the commodity demand matching weight, screen the characteristic components with the proportion higher than the set threshold, and use the formula:

[0048] ;

[0049] Calculate the commodity matching degree parameter to obtain a commodity matching ratio matrix;

[0050] wherein, represents the commodity matching ratio, represents the characteristic matching weight, represents the matching degree of the commodity in the u-th characteristic dimension, Calculate the normalized proportion of all matching characteristics, is the smoothing adjustment parameter, represents the demand expectation value of the commodity in the u-th characteristic dimension, represents the actually matched characteristic value is an adjustment factor, calculate the normalized deviation of the expected demand value;

[0051] S504: Call the commodity matching ratio matrix, integrate multiple matching characteristic components, calculate the weighted matching ratio of the commodity in the dimension of differentiated user needs, call the weighted matching ratio data, screen the commodities whose matching degree exceeds the set threshold, adjust the weights of the commodities with lower matching ratios, optimize the matching rules, and generate the commodity demand characteristic matching ratio value.

[0052] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0053] In the present invention, through the covariance matrix extraction and dimensionality reduction analysis of commodity characteristic parameters, the interference of redundant information caused by high-dimensional characteristics is effectively avoided, and the data processing efficiency and analysis accuracy are improved. Through the linear projection of the user demand vector and the calculation of the Euclidean distance, the matching relationship between the user demand and the commodity characteristics is accurately optimized. Based on the weight adjustment of the contribution rate of the cross-characteristic components, the adaptability of the data decomposition result to dynamic changes is enhanced, ensuring that the synergistic effect between the characteristics is more accurate. By extracting the sparse distribution coefficient of the long-tail commodity characteristics and optimizing the weight allocation, the imbalance problem of long-tail resource allocation is improved, and the resource utilization efficiency is significantly improved. Dynamically adjust the matching relationship of the key characteristic dimensions, combined with the calculation of the matching ratio, to realize the real-time optimization of the relationship between the commodity and the demand, providing efficient support for decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 is a schematic diagram of the main steps of the present invention;

[0055] Figure 2 is a flowchart of the steps for obtaining the main component matrix of the commodity of the present invention;

[0056] Figure 3 is a flowchart of the steps for obtaining the user demand alignment matrix of the present invention;

[0057] Figure 4 is a flowchart of the steps for obtaining the reconstructed characteristic matching matrix of the present invention;

[0058] Figure 5 is a flowchart of the steps for obtaining the long-tail characteristic optimization matrix of the present invention;

[0059] Figure 6 is a flowchart of the steps for obtaining the commodity demand characteristic matching ratio value of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0060] In order to make the objectives, technical solutions and advantages of the present invention more clear and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0061] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the accompanying drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation to the present invention. In addition, in the description of the present invention, the meaning of "a plurality of" is two or more, unless otherwise specifically defined. Embodiment 1

[0062] Please refer to Figure 1 , the present invention provides a technical solution: an e-commerce operation data analysis and evaluation method, including the following steps:

[0063] S1: Based on the product classification information and product characteristic parameters in the e-commerce operation data, extract the product characteristic vector covariance matrix, calculate the eigenvalues and eigenvectors in the matrix, sort the vectors according to the eigenvalues and complete the dimensionality reduction, extract the main characteristic components, and generate the product principal component matrix;

[0064] S2: Extract the user purchase preferences in the e-commerce operation data, combine with the product principal component matrix, perform a linear projection on the user demand vector according to the product characteristic dimension, calculate the Euclidean distance between the demand and the principal component vector, sort according to the distance size and complete the characteristic alignment, and generate the user demand alignment matrix;

[0065] S3: Statistically analyze the transaction cross-characteristic table in the e-commerce operation data, combine with the user demand alignment matrix, perform a decomposition operation on the characteristic cross-matrix, extract the component contribution rate and adjust the matrix weight, re-integrate the component matrix according to the adjustment result, complete the characteristic reconstruction operation, and generate the reconstructed characteristic matching matrix;

[0066] S4: Analyze the product long-tail distribution in the e-commerce operation data, combine with the reconstructed characteristic matching matrix, extract the characteristic weight sparse distribution coefficient in the matrix, analyze the long-tail product characteristic weight and increase the weight of the sparse characteristic dimension, optimize the long-tail characteristic distribution, and generate the long-tail characteristic optimization matrix;

[0067] S5: Statistically record the user-product matching degree in the e-commerce operation data, optimize the matrix in combination with the long-tail characteristics, perform dynamic adjustment operations on the matching components of the key characteristic dimensions in the matrix, calculate the matching ratio between the product and the demand, integrate the matching values, and generate the product-demand characteristic matching ratio value.

[0068] The product principal component matrix includes main characteristic components, eigenvalue vectors, and eigenvalue rankings. The user demand alignment matrix includes user demand vectors, linear projection results, and Euclidean distances. The reconstructed characteristic matching matrix includes characteristic cross matrix components, component contribution rates, and weight adjustment results. The long-tail characteristic optimization matrix includes characteristic weight sparse distribution coefficients, long-tail product characteristic weights, and sparse characteristic dimension gain weights. The product-demand characteristic matching ratio value includes key characteristic dimension matching components, dynamic adjustment results, and product-demand matching ratios.

[0069] Please refer to Figure 2 , and the specific steps for obtaining the product principal component matrix are as follows:

[0070] S101: Based on the product classification information and product characteristic parameters in the e-commerce operation data, construct a product characteristic vector space, call the hierarchical structure parameters in the product classification information, screen the classification nodes with independent characteristic descriptions within the product classification hierarchy, match the product characteristic parameters under the classification nodes, eliminate the redundant characteristic items caused by classification intersections, establish a set of characteristic parameters, adjust the data scale according to the numerical range, data type, and comparability of the product characteristic parameters, and use normalization transformation to unify the data scale to construct a product characteristic covariance matrix;

[0071] Based on the product classification information and product characteristic parameters in the e-commerce operation data, call the hierarchical structure parameters in the product classification information, parse the product classification data, split the classification nodes according to the product category hierarchical structure, extract the classification nodes with independent characteristic descriptions, establish a classification mapping relationship, extract the product characteristic parameters under the corresponding product categories, call the historical data in the product database, perform data preprocessing on each product characteristic parameter, eliminate the data missing values and outliers, use the mean filling or interpolation algorithm to complete the missing items, and perform normalization transformation on the product characteristic parameters according to their data types respectively. For continuous numerical data, use the min-max normalization method to map all data to the [0,1] interval. For discrete numerical data, use the Z-score normalization method to adjust the data scale with the standard deviation to ensure data comparability. According to the covariance calculation formula of the product characteristic parameters: Among them, represents the product characteristic parameter and the product characteristic parameter The covariance between them, represents the th sample corresponding to the product characteristic parameter The value of represents the commodity characteristic parameter corresponding to the th sample, the value of and respectively represent the mean values of the commodity characteristic parameters and the commodity characteristic parameter . represents the total number of commodity samples, is the index number of the commodity sample. Calculate the covariance relationship between all characteristic parameters, eliminate redundant characteristic items and parameters affected by multicollinearity, and finally obtain the commodity characteristic covariance matrix.

[0072] S102: Perform eigenvalue decomposition on the commodity characteristic covariance matrix, calculate the eigenvalues and eigenvectors of the covariance matrix, call the matrix decomposition result, sort in descending order according to the eigenvalue size, calculate the eigenvalue contribution rate based on the eigenvalue distribution, set the contribution rate threshold, screen the eigenvalues whose contribution rate exceeds the threshold, and extract the corresponding eigenvectors to form an eigenvalue sorting vector matrix;

[0073] Perform eigenvalue decomposition on the commodity characteristic covariance matrix, call the matrix decomposition algorithm to calculate the eigenvalues and eigenvectors of the covariance matrix, sort the eigenvalues, sort in descending order according to the eigenvalue size, and set the eigenvalue contribution rate calculation formula: Among them, represents the contribution rate of the eigenvalue to the total sum of all eigenvalues, represents the th eigenvalue of the covariance matrix, represents the sum of all eigenvalues of the covariance matrix, represents the total number of eigenvalues of the covariance matrix. Call the cumulative contribution rate calculation formula: Among them, represents the cumulative contribution rate of the first eigenvalues, represents the number of selected eigenvalues. Set the cumulative contribution rate threshold, screen the eigenvalues whose cumulative contribution rate exceeds the set threshold, and extract the corresponding eigenvectors to construct an eigenvalue sorting vector matrix.

[0074] S103: Perform dimensionality reduction processing on the eigenvalue sorting vector matrix, set the cumulative contribution rate threshold, retain the first k eigenvectors whose contribution rate reaches the set value, and use the formula:

[0075] ;

[0076] Calculate the main components of commodity characteristics and establish a commodity main component matrix;

[0077] Among them, represents the commodity main component matrix, represents the i-th eigenvalue of the covariance matrix, represents the c-th eigenvector of the covariance matrix, represents the number of selected principal components, is the smoothing adjustment parameter to ensure the numerical stability of the dimensionality reduction result

[0078] Formula:

[0079] ;

[0080] The advantage of the formula is that by introducing the cumulative term of the absolute value eigenvalues in the denominator part, the numerical stability of the calculation is ensured, and the smoothing adjustment parameter is added to the denominator , effectively reducing the numerical fluctuation problem caused by small eigenvalues. At the same time, the weighted sum form of eigenvalues and eigenvectors is adopted to ensure that the calculation of the principal component matrix can maintain the optimal feature representation ability.

[0081] Detailed explanation of the formula and the derivation process of the formula calculation:

[0082] Among them, represents the commodity principal component matrix, represents the -th eigenvalue of the covariance matrix, , represents the number of selected principal components, represents the sum of the absolute values of the first eigenvalues, is the smoothing adjustment parameter.

[0083] Set the covariance matrix of the commodity characteristics of commodity category A, eigenvalue , eigenvector , set , sum of the absolute values of the cumulative eigenvalues: ;

[0084] Assume the smoothing parameter , calculate the denominator: ;

[0085] Calculate each principal component:

[0086] ;

[0087] ;

[0088] ;

[0089] The finally calculated principal component matrix is: ;

[0090] The results show that by calculating the principal components of product characteristics, the dimensionality is successfully reduced to retain the main feature information, and a product principal component matrix is constructed, which can be used for subsequent data analysis and model construction.

[0091] Please refer to Figure 3 , and the steps for obtaining the user demand alignment matrix are specifically as follows:

[0092] S201: Extract user purchase records and preferences from e-commerce operation data, screen the product categories of user purchase records, call the product characteristic parameters of the product principal component matrix, match the products purchased by users according to the product classification labels, and extract the user demand vectors consistent with the characteristic dimensions of the product principal component matrix;

[0093] Extract the user's purchase history and preferences from e-commerce operation data, call the user's historical transaction records, screen the product IDs and product category labels included in the transaction details, eliminate the abnormal transaction data that do not completely record the purchase time, transaction quantity, and payment amount, calculate the user's average purchase cycle, divide the recent purchase behavior and long-term purchase behavior based on the time window, calculate the purchase frequency of the user under different product categories, obtain the purchase proportion of the user in each product category, call the product characteristic parameters of the product principal component matrix, screen the characteristic attributes of the products purchased by the user, perform normalization processing on the characteristic attributes of different product categories to ensure the comparability of product characteristics on the numerical scale, calculate the weighted user demand intensity according to the user's purchase frequency and purchase proportion, adjust the product characteristic weights, eliminate the abnormal data in the purchase history, such as the situation of multiple transactions of the same product in a short period of time but without payment, calculate the stability index of the user's purchase behavior, if the user stability index is lower than the set threshold, then reduce the weight of the user's purchase records to ensure that the overall analysis will not be affected by abnormal user behavior during the calculation process, and finally establish the user demand vector.

[0094] S202: Perform a linear projection on the user demand vector using the product principal component matrix, call the principal component weights of the product characteristic parameters, and according to the linear mapping rule of the projection matrix, transform the user demand vector on the product principal component matrix, calculate the projection coordinates, screen the eigenvectors that deviate from the central distribution in the projection results, adjust the user demand weights, and obtain the user demand principal component vector;

[0095] Perform a linear projection on the user demand vector using the commodity principal component matrix, call the principal component weights of the commodity characteristic parameters, construct a characteristic dimension conversion matrix according to the linear mapping rule of the projection matrix, calculate the projection coordinates of the user demand vector in the commodity principal component matrix, screen the eigenvectors deviating from the central distribution in the projection results, set a characteristic deviation threshold, calculate the standardized deviation of the user demand in each characteristic dimension, adjust the user demand weights if the deviation exceeds the threshold, calculate the weighted sum of the adjusted user demand vector to ensure the balance between characteristic dimensions, and generate the user demand principal component vector after normalization processing.

[0096] S203: Calculate the Euclidean distance between the user demand principal component vector and the commodity principal component matrix, using the formula:

[0097] ;

[0098] Obtain the distance sorting result;

[0099] Among them, the distance between the user demand principal component vector and the a-th principal component vector of the commodity principal component matrix is , the component of the user demand principal component vector in the j-th characteristic dimension is , the component of the commodity principal component matrix in the j-th characteristic dimension of the a-th principal component is , the weight parameter adjusts the influence degree of the differential characteristic dimension, and the correction parameter is used to avoid calculation errors caused by too small values. After calculating all distances, sort them in ascending order, and n represents the total number of characteristic dimensions.

[0100] Formula:

[0101] ;

[0102] The advantage of the formula is that by introducing the weight parameter the influence degree of different characteristic dimensions can be adjusted, and at the same time, the correction parameter is added to prevent the distance calculation from being too small and causing numerical instability, thus ensuring the robustness of the distance sorting.

[0103] Detailed explanation of the formula and the derivation process of the formula calculation:

[0104] Among them, represents the distance between the user demand principal component vector and the a-th principal component vector of the commodity principal component matrix, is the component of the user demand principal component vector in the j-th characteristic dimension, is the j-th dimensional component of the commodity principal component matrix in the i-th principal component, is the weight parameter, which adjusts the influence of different characteristic dimensions on the distance calculation, To correct the parameters and prevent data distortion caused by the calculation results tending to zero.

[0105] Data collection method:

[0106] Calculation method: Obtain the demand distribution of users on product characteristics through their historical purchase records, and perform weighted calculation based on the purchase frequency, and obtain it after normalization.

[0107] Calculation method: Construct a product principal component matrix based on product characteristic parameters, and calculate the principal component vectors of each characteristic dimension through eigenvalue decomposition method.

[0108] Setting basis: Set according to the preference degree of users on different characteristic dimensions. If the user's demand for a certain characteristic dimension is high, a larger weight is given. The calculation method is as follows:

[0109] ;

[0110] Among them, is the mean value of this characteristic dimension among all users, is the standard deviation of this characteristic dimension. If the user's demand is far from the mean value, a higher weight is given.

[0111] Setting basis: Used to smooth the calculation process, usually adjusted according to the minimum value of the distance distribution. The calculation method is as follows:

[0112] ;

[0113] Calculate the minimum distance value to prevent data distortion caused by too small distance calculation.

[0114] Formula calculation example:

[0115] Set the component of the user demand vector as , the component of the product principal component matrix as , set the weight parameter , the correction parameter , substitute into the formula for calculation:

[0116] ;

[0117] ;

[0118] ;

[0119] The result shows that the Euclidean distance between the user demand vector and the first vector of the commodity principal component matrix is 0.2828. The smaller this value is, the closer the user demand is to the principal component vector. By calculating and sorting all distances subsequently, the characteristic vector that best meets the user demand can be obtained, and then the distance sorting result can be obtained in the subsequent steps.

[0120] S204: Call the distance sorting result, perform characteristic alignment according to the sorting order, screen the principal component vector that is closest to the user demand preference, adjust the proportion between characteristic dimensions, and generate the user demand alignment matrix.

[0121] Call the distance sorting result, based on the distance sorting order, perform characteristic alignment between the user demand vector and the commodity principal component vector, set the characteristic matching standard, calculate the relative deviation of each characteristic dimension. If the relative deviation is lower than the set threshold, it is determined that the characteristic matches; otherwise, recalculate the characteristic weight for correction, adjust the proportion between characteristic dimensions, ensure the stability of the projection of the principal component vector in the user demand vector, and finally generate the user demand alignment matrix.

[0122] Please refer to Figure 4 , and the specific steps for obtaining the reconstructed characteristic matching matrix are as follows:

[0123] S301: Statistically analyze the transaction cross-characteristic table in the e-commerce operation data, extract the correlation parameters between different transaction characteristics, call the user demand alignment matrix, and construct a characteristic correlation matrix based on the co-occurrence relationship of commodities in the transaction records. Screen the commodity characteristic items according to the co-occurrence frequency, calculate the correlation weight between transaction characteristics, and establish a characteristic cross matrix;

[0124] Statistically analyze the transaction cross-characteristic table in the e-commerce operation data, extract the commodity association information in all transaction records, call the transaction record timestamp, analyze the purchase frequency of different commodity combinations, construct a commodity combination co-occurrence matrix, calculate the similarity index of the characteristic vectors for high-frequency co-occurring commodity combinations, screen the commodity characteristic pairs with a similarity index higher than the set threshold, establish a characteristic correlation matrix, normalize the characteristic correlation matrix, calculate the transaction contribution rate of different characteristic combinations, eliminate the characteristic pairs with low contribution rates, based on the characteristic transaction contribution rate matrix, construct a cross-characteristic table, call the user demand alignment matrix, match the characteristic vectors of the user demand, screen the main characteristic dimensions where the user demand intersects with the transaction characteristics in the cross-characteristic table, calculate the matching degree between the user demand dimension and the transaction characteristic dimension, eliminate the dimensions with low matching degrees, optimize the characteristic combination rule, screen the characteristic combinations that finally meet the transaction cross-characteristic conditions, and establish a characteristic cross matrix.

[0125] S302: Decompose and operate on the feature cross matrix, extract the principal components of multiple feature dimensions, calculate the contribution rate of components, call the principal component weight vector, sort according to the contribution rate of transaction features, filter out the feature components with contribution rate lower than the threshold, recalculate the contribution weights of the remaining feature items, and obtain the weight adjustment matrix;

[0126] Decompose and operate on the feature cross matrix, call the matrix decomposition method, calculate the principal component contribution rate of the feature vectors in different dimensions, sort according to the contribution rate, filter out the principal component features with high contribution rate, calculate the proportion weights of each principal component vector in the feature cross matrix, filter out the feature items with contribution rate lower than the set threshold, perform feature screening according to the feature contribution rate, remove the feature components with proportion weight lower than the threshold, call the transaction history data, calculate the influence weight of the screened feature components on the transaction records, adjust the feature vector weight coefficients in the feature cross matrix, establish the weight correction matrix, call the user purchase preference data, analyze the matching degree between the feature components and user preferences, calculate the user preference influence coefficient of each feature component, adjust the user preference correction coefficient of the feature weights, and construct the weight adjustment matrix.

[0127] S303: Reconstruct the feature cross matrix based on the weight adjustment matrix, calculate the weight of the adjusted feature matrix, using the formula:

[0128] ;

[0129] Calculate the adjusted feature matrix to obtain the reconstructed feature matrix;

[0130] Among them, represents the feature weight value of the j-th feature dimension in the z-th row of the adjusted feature matrix, represents the feature value at the corresponding position in the original feature cross matrix, represents the weight coefficient of the j-th feature dimension, Calculate the normalized proportion of the current feature dimension in the adjusted matrix, is the smoothing adjustment parameter to avoid the denominator being zero

[0131] Formula:

[0132] ;

[0133] The benefit of the formula is that by normalizing the calculation of the feature weights in the feature matrix, it avoids the feature matching deviation caused by uneven distribution of feature weights, and uses the smoothing parameter to control the adjustment range of feature weights, improving the stability of feature matching.

[0134] Detailed explanation of the formula and the formula calculation derivation process:

[0135] represents the z-th row and the The characteristic weight value represents the characteristic weight value at the i-th row and j-th column in the original characteristic cross matrix. represents the weight coefficient of the j-th characteristic dimension. Calculate the normalized proportion of the current characteristic dimension in the adjusted matrix. represents the smoothing adjustment parameter to avoid calculation instability caused by the denominator approaching zero.

[0136] Suppose the partial commodity characteristic cross matrix data is statistically obtained from the transaction data of an e-commerce platform as follows:

[0137] ;

[0138] Characteristic dimension weight coefficient:

[0139] ;

[0140] Let the smoothing adjustment parameter , and calculate the element (1,1) in the reconstructed characteristic matrix:

[0141] ;

[0142] ;

[0143] ;

[0144] ;

[0145] Similarly, calculate all matrix elements to obtain the reconstructed characteristic matrix:

[0146] ;

[0147] This result shows that by adjusting the characteristic weights and normalization calculation, the characteristic distribution of the characteristic matching matrix is optimized, the balance and stability of characteristic matching are improved, and finally the reconstructed characteristic matrix is obtained.

[0148] S304: Re-integrate the matrix structure according to the characteristic components of the reconstructed characteristic matrix, call the correlation parameters between transaction characteristics, calculate the interaction intensity of characteristic dimensions, screen the characteristic components with high interaction intensity, optimize the matching rules of characteristic components, adjust the characteristic alignment method, and generate the reconstructed characteristic matching matrix.

[0149] Re-integrate the matrix structure according to the characteristic components of the reconstructed characteristic matrix, call the correlation parameters between trading characteristics, calculate the synergy effect between different characteristic dimensions, call the trading records, analyze the frequently co-occurring characteristic items, calculate the interaction intensity of each characteristic component, screen the characteristic components with high interaction intensity, calculate the matching similarity between different characteristic dimensions according to the trading matching data, call the calculation results of the characteristic matching degree, adjust the matching rules of the characteristic dimensions, optimize the characteristic combination method, eliminate the characteristic items with abnormal characteristic matching, ensure the stability of the characteristic matching, call the characteristic component data after the characteristic matching adjustment, adjust the characteristic alignment method, construct the final characteristic matching structure, and generate the reconstructed characteristic matching matrix.

[0150] Please refer to Figure 5 , and the steps for obtaining the long-tail characteristic optimization matrix are specifically as follows:

[0151] S401: Analyze the commodity long-tail distribution in the e-commerce operation data, call the sales frequency of the commodities in the trading records, calculate the distribution of commodity sales volume, screen the commodities with long-term low-frequency sales volume, extract the characteristic components of the low-frequency commodities in the reconstructed characteristic matching matrix, calculate the weight values of the long-tail commodities in the characteristic dimension, and establish the long-tail commodity characteristic weight matrix;

[0152] Analyze the commodity long-tail distribution in the e-commerce operation data, call the sales frequency of the commodities in the trading records, count the sales quantity of each commodity within a period of time, set the time window as the past 180 days, sort the sales frequencies of each commodity, calculate the cumulative sales proportion, use the top 20% of the commodities with the highest sales quantity as mainstream commodities, and the remaining commodities as long-tail commodities. Eliminate the long-tail characteristics caused by data anomalies, such as abnormal long-tail commodities caused by promotions and sudden demand growth. Extract the set of commodities that meet the long-tail characteristics, call the commodity characteristic parameters in the reconstructed characteristic matching matrix, screen the characteristic weights of the long-tail commodities, calculate the proportion of the characteristic weights of the long-tail commodities, call the maximum and minimum values of the characteristic weights to calculate the dispersion degree of the characteristic weights, eliminate the commodity characteristics with too large weight change range, adjust the scale range of the characteristic weights, normalize the weight values of different characteristic dimensions, calculate the average weight of the characteristic dimensions, eliminate the characteristic weight items below the average value, and finally construct the long-tail commodity characteristic weight matrix according to the distribution of the characteristic dimensions.

[0153] S402: Calculate the sparse distribution coefficient of the long-tail commodity characteristic weight matrix, call the characteristic weight vector, calculate the proportion of the weights of multiple characteristic dimensions, screen the characteristic dimensions with low weight proportion according to the dispersion degree of the characteristic weights, extract the characteristic components with the lowest weight proportion, calculate the average weight between the characteristic dimensions, call the characteristic dimension weight data below the average value, and construct the sparse characteristic weight matrix;

[0154] Calculate the sparse distribution coefficient of the feature weight matrix of long-tail products, call the feature weight vector, set the distribution threshold of the feature weight ratio, filter the feature dimensions whose feature weight ratio is lower than the threshold, call the weight value of long-tail products in each feature dimension, calculate the coefficient of variation of the feature dimension, and use the formula: in, represents the coefficient of variation of the j-th characteristic dimension, is the standard deviation of the feature weight, The feature weight is the mean, and the coefficient of variation of each feature dimension is calculated. The feature dimensions with a coefficient of variation greater than 0.5 are screened and defined as sparse features. The feature components with a coefficient of variation higher than 1.5 are removed. The mean and variance of the feature weight are called to adjust the distribution of the feature weight. The scale range of the sparse feature weight is renormalized, and the adjusted feature weight components are screened. The weight ratio calculated in the previous steps is called to match the sparse feature component. The feature weight value is adjusted according to the matching degree, the distribution uniformity of the feature dimension is optimized, and finally a sparse feature weight matrix is ​​constructed.

[0155] S403: Adjust the gain weight of the feature dimension in the sparse feature weight matrix using the formula:

[0156] ;

[0157] The gain adjustment characteristic matrix is ​​calculated;

[0158] in, represents the gain-adjusted feature weight, represents the original sparse feature weight, is the gain adjustment factor, which adjusts the proportion of low-weight characteristics. Calculate the normalized total weight of the current feature dimension, To smooth the parameters and avoid extreme data interfering with the calculation;

[0159] formula:

[0160] ;

[0161] The benefit of the formula is that by introducing the adjustment item of feature weight normalization, the weight ratio imbalance problem caused by too low weight values ​​of some features is avoided. This allows low-weight features to be optimized and improves the matching capability of long-tail features.

[0162] Detailed explanation of the formula and the process of formula calculation and derivation:

[0163] represents the gain-adjusted feature weight, represents the original sparse feature weight, is the gain adjustment factor, which adjusts the proportion of low-weight characteristics. Calculate the normalized total weight of the current characteristic dimension. is the smoothing parameter to prevent extreme data from interfering with the calculation.

[0164] Set parameters:

[0165] , , , , ;

[0166] Calculate the sum of the normalized characteristic weights:

[0167] ;

[0168] Calculate the gain adjustment term:

[0169] ;

[0170] Calculate the characteristic weights after gain adjustment:

[0171] ;

[0172] This result shows that after gain adjustment, the characteristic value with the original low weight has increased from 0.12 to 0.2387, optimizing the proportion of sparse characteristics in the weight distribution of long-tail product characteristics, enhancing the characteristic influence of long-tail products, and providing optimized characteristic data for the construction of the subsequent long-tail characteristic optimization matrix.

[0173] S404: Optimize the long-tail characteristic distribution based on the gain-adjusted characteristic matrix, call the characteristic weight values of long-tail products, screen the characteristic weights after gain adjustment, extract the characteristic components with the adjustment amplitude exceeding the set threshold, adjust the weight proportion of low-weight characteristics in long-tail products, optimize the characteristic structure of long-tail products, and generate the long-tail characteristic optimization matrix.

[0174] Optimize the long-tail feature distribution based on the gain-adjusted feature matrix, call the feature weight value of the long-tail product, calculate the mean of the feature weight after gain adjustment, screen the adjusted feature weight distribution, call the feature weight data before and after adjustment, calculate the feature dimensions with an adjustment range of more than 30%, eliminate the feature weight items with excessive gain, recalculate the mean and standard deviation of each feature dimension, normalize the adjusted feature weight, match the feature components of long-tail products, calculate the proportion of long-tail features in different product categories, optimize the feature matching rules based on the feature proportion of product categories, calculate the deviation value of feature weight matching, adjust the feature weight whose feature deviation exceeds the set range, eliminate the feature items with invalid adjustment, build the optimized feature matching matrix, call the feature weight data in the matching matrix, calculate the feature weight matching degree, screen the feature components with high matching degree, adjust the weight according to the matching degree, and finally generate the long-tail feature optimization matrix.

[0175] See also Figure 6 , the specific steps for obtaining the commodity demand characteristic matching ratio value are as follows:

[0176] S501: Count the user-product matching records in the e-commerce business data, call the user transaction data, select matching items according to the number of successful transactions and evaluations, extract the characteristic vector of the corresponding product, call the long-tail characteristic optimization matrix, screen the optimized characteristic matching parameters, calculate the matching degree between the user demand vector and the product characteristic vector, and establish the product characteristic matching matrix;

[0177] Statistics are collected on user-product matching records in e-commerce business data, historical transaction data of users are called, matching items with high transaction success times and positive evaluations are screened, feature vectors of corresponding products are extracted, contribution values ​​of product features in user demand matching are calculated, long-tail feature optimization matrix is ​​called, optimized feature matching parameters are screened, matching degree between user demand vector and product feature vector is calculated, weight distribution of matching features in different user groups is called, matching features with high contribution weights are screened, matching relationships of products under multiple user demand dimensions are calculated, matching data are normalized, abnormal matching values ​​are eliminated, and finally a product feature matching matrix is ​​established.

[0178] S502: calling the key characteristic dimensions of the product characteristic matching matrix, dynamically adjusting the matching components in the matrix, calculating the contribution weights of user needs in multiple characteristic dimensions, adjusting the matching characteristic weights according to user preferences, selecting characteristic dimensions with high contribution weights, calling the adjusted matching matrix, normalizing the matching weights, and obtaining a matching characteristic adjustment matrix;

[0179] Call the key feature dimensions of the product feature matching matrix to dynamically adjust the matching components in the matrix. Call the user interaction data to calculate the weight change trend of the matching features, adjust the matching feature weights according to the changes in user requirements, calculate the contribution values of each feature dimension, eliminate the feature components with too low contribution values, screen the feature dimensions with high contribution weights, calculate the dynamic distribution of user preferences, call the adjusted matching matrix, perform a normalization transformation on the matching weights, calculate the proportion of each feature dimension in the matching, eliminate the feature items with abnormal matching, and re-optimize the feature dimension matching parameters to finally obtain the matching feature adjustment matrix.

[0180] S503: Calculate the matching ratio of the product and the demand. Call the feature matching components in the matching feature adjustment matrix, and calculate the contribution proportion of the product in multiple feature dimensions based on the product demand matching weights. Screen the feature components with a proportion higher than the set threshold, and use the formula:

[0181] ;

[0182] Calculate the product matching degree parameter and obtain the product matching ratio matrix;

[0183] Among them, represents the product matching ratio, represents the feature matching weight, represents the matching degree of the product in the u-th feature dimension, Calculate the normalized proportion of all matching features, is the smoothing adjustment parameter, represents the demand expectation value of the product in the u-th feature dimension, represents the actual feature value after matching, is the adjustment factor, Calculate the normalized deviation of the demand expectation value;

[0184] Formula:

[0185] ;

[0186] The advantage of the formula is that by calculating the feature matching weight and the feature demand deviation simultaneously and normalizing the matching features, it can more accurately measure the matching degree between the product and the user's needs, and at the same time improve the robustness of the matching calculation.

[0187] Detailed explanation of the formula and the formula calculation derivation process:

[0188] Set parameters:

[0189] ;

[0190] ;

[0191] ;

[0192] ;

[0193] ;

[0194] 。

[0195] Calculate the normalized sum of matching characteristics:

[0196] ;

[0197] Calculate the first part of the matching degree:

[0198] ;

[0199] ;

[0200] Calculate the normalized sum of demand matching deviation:

[0201] ;

[0202] Calculate the deviation of demand matching degree:

[0203] ;

[0204] ;

[0205] Calculate the final matching ratio:

[0206] ;

[0207] The result shows that the calculated matching ratio is 0.9152, indicating a relatively high matching degree of product characteristics under the current user needs. This matching ratio can be used in the subsequent matching optimization process and as a reference for adjusting the matching weights.

[0208] S504: Call the product matching ratio matrix, integrate multiple matching characteristic components, calculate the weighted matching ratio of the product in the dimension of differentiated user needs, call the weighted matching ratio data, screen the products whose matching degree exceeds the set threshold, adjust the weights of the products with relatively low matching ratios, optimize the matching rules, and generate the product demand characteristic matching ratio value.

[0209] Call the commodity matching ratio matrix, calculate the contribution value of different matching characteristic components in the overall matching, call the user demand parameters to calculate the matching weights of each characteristic dimension, calculate the weighted matching ratio of the commodity in different user demand dimensions, eliminate the commodities with low matching ratios, call the weighted matching ratio data, screen the commodities whose matching degree exceeds the set threshold, adjust the commodity weights according to the dynamic change trend of the matching ratio, screen the characteristic dimensions with small changes in the matching ratio, optimize the matching rules, call the adjusted matching data to calculate the final characteristic matching ratio value, eliminate the characteristics with excessive matching fluctuations, and finally generate the commodity demand characteristic matching ratio value.

[0210] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any person skilled in the art may use the disclosed technical content to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, as long as the technical content of the present invention is not departed from, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention still belong to the protection scope of the technical solution of the present invention.

Claims

1. An e-commerce business data analysis and evaluation method, characterized in that: The following steps are involved: S1: Based on the commodity classification information and commodity characteristic parameters in the e-commerce business data, extract the commodity characteristic vector covariance matrix, calculate the eigenvalues ​​and eigenvectors in the matrix, sort the vectors according to the eigenvalues ​​and complete the dimensionality reduction, extract the main characteristic components, and generate the commodity principal component matrix; S2: extracting user purchase preferences from e-commerce business data, combining the commodity principal component matrix, linearly projecting the user demand vector according to the commodity characteristic dimension, calculating the Euclidean distance between the demand and the principal component vector, sorting by distance and completing characteristic alignment, and generating a user demand alignment matrix; S3: Count the transaction cross-feature table in the e-commerce business data, combine it with the user demand alignment matrix, extract the commodity association information in all transaction records, call the transaction record timestamp, analyze the purchase frequency of different commodity combinations, calculate the similarity index of the feature vector, screen the commodity feature pairs with a similarity index higher than the set threshold, establish a feature association matrix and normalize the feature association matrix, calculate the transaction contribution rate of different feature combinations, match the feature vector of user demand, calculate the matching degree between the user demand dimension and the transaction feature dimension, establish a feature cross matrix, perform decomposition operation on the feature cross matrix, extract the component contribution rate and adjust the matrix weight, reintegrate the component matrix according to the adjustment result, complete the feature reconstruction operation, and generate a reconstructed feature matching matrix; S4: Analyze the long-tail distribution of commodities in the e-commerce business data, extract the sparse distribution coefficient of the characteristic weight in the matrix in combination with the reconstructed characteristic matching matrix, analyze the characteristic weight of the long-tail commodities and add weight to the sparse characteristic dimension, optimize the long-tail characteristic distribution, and generate a long-tail characteristic optimization matrix; S5: Count the user-product matching records in the e-commerce business data, combine the long-tail feature optimization matrix, dynamically adjust the key feature dimension matching components in the matrix, calculate the matching ratio of products to demand and integrate the matching values, and generate a product demand feature matching ratio value.

2. The e-commerce business data analysis and evaluation method according to claim 1, characterized in that: The commodity principal component matrix includes main characteristic components, eigenvalue vectors, and eigenvalue sorting; the user demand alignment matrix includes user demand vectors, linear projection results, and Euclidean distances; the reconstructed characteristic matching matrix includes characteristic cross matrix components, component contribution rates, and weight adjustment results; the long-tail characteristic optimization matrix includes characteristic weight sparse distribution coefficients, long-tail commodity characteristic weights, and sparse characteristic dimension gain weights; the commodity demand characteristic matching ratio value includes key characteristic dimension matching components, dynamic adjustment results, and commodity demand matching ratios.

3. The e-commerce business data analysis and evaluation method according to claim 2, characterized in that: The steps for obtaining the commodity principal component matrix are specifically as follows: S101: Based on the commodity classification information and commodity characteristic parameters in the e-commerce business data, construct a commodity characteristic vector space, call the hierarchical structure parameters in the commodity classification information, screen the classification nodes with independent characteristic descriptions in the commodity classification hierarchy, match the commodity characteristic parameters under the classification nodes, eliminate the redundant characteristic items caused by the classification intersection, establish a characteristic parameter set, adjust the data scale according to the numerical range, data type and comparability of the commodity characteristic parameters, use normalization transformation to unify the data scale, and construct a commodity characteristic covariance matrix; S102: Perform eigendecomposition on the covariance matrix of the product characteristics, calculate the eigenvalues ​​and eigenvectors of the covariance matrix, call the matrix decomposition results, sort them in descending order according to the size of the eigenvalues, calculate the eigenvalue contribution rate based on the eigenvalue distribution, set the contribution rate threshold, filter the eigenvalues ​​whose contribution rate exceeds the threshold, extract the corresponding eigenvectors, and form an eigenvalue sorting vector matrix; S103: Perform dimensionality reduction processing on the eigenvalue sorting vector matrix, set a cumulative contribution rate threshold, retain the first k eigenvectors whose contribution rates reach the set value, and use the formula: ； Calculate the principal components of product characteristics and establish the principal component matrix of products; in, represents the commodity principal component matrix, represents the i-th eigenvalue of the covariance matrix, represents the cth eigenvector of the covariance matrix, represents the number of selected principal components, To smoothly adjust the parameters and ensure the numerical stability of the dimensionality reduction results.

4. The e-commerce business data analysis and evaluation method according to claim 3 is characterized in that: The steps for obtaining the user demand alignment matrix are specifically as follows: S201: extracting user purchase records and preferences from e-commerce business data, screening the commodity categories of user purchase records, calling commodity characteristic parameters of the commodity principal component matrix, matching the commodities purchased by the user according to commodity classification labels, and extracting user demand vectors consistent with the characteristic dimensions of the commodity principal component matrix; S202: linearly project the user demand vector using the commodity principal component matrix, call the principal component weights of the commodity characteristic parameters, transform the user demand vector on the commodity principal component matrix according to the linear mapping rule of the projection matrix, calculate the projection coordinates, filter the eigenvectors that deviate from the central distribution in the projection results, adjust the user demand weights, and obtain the user demand principal component vector; S203: Calculate the Euclidean distance between the user demand principal component vector and the product principal component matrix using the formula: ； Get the distance sorting results; Among them, the distance between the user demand principal component vector and the ath principal component vector of the product principal component matrix is , the component of the user demand principal component vector on the jth characteristic dimension is , the component of the commodity principal component matrix on the jth characteristic dimension of the ath principal component is , Represents the weight coefficient of the jth characteristic dimension, adjusts the influence of the differentiated characteristic dimension, and corrects the parameter It is used to avoid calculation errors caused by too small values. After calculating all distances, they are arranged in ascending order. n represents the total number of feature dimensions. S204: calling the distance sorting result, performing feature alignment according to the sorting order, selecting the principal component vector closest to the user's demand preference, adjusting the proportions between the feature dimensions, and generating a user demand alignment matrix.

5. The e-commerce business data analysis and evaluation method according to claim 4, characterized in that: The steps of obtaining the reconstructed characteristic matching matrix are specifically as follows: S301: Counting the transaction cross-feature table in the e-commerce business data, extracting the correlation parameters between the differentiated transaction features, calling the user demand alignment matrix, constructing a feature correlation matrix based on the co-occurrence relationship of the commodities in the transaction records, screening commodity feature items based on the co-occurrence frequency, calculating the correlation weights between the transaction features, and establishing a feature cross-matrix; S302: Decomposing the characteristic cross matrix, extracting the principal components of multiple characteristic dimensions, calculating the component contribution rates, calling the principal component weight vector, sorting according to the transaction characteristic contribution rates, screening the characteristic components whose contribution rates are lower than the threshold, recalculating the contribution weights of the retained characteristic items, and obtaining the weight adjustment matrix; S303: Reconstruct the characteristic cross matrix based on the weight adjustment matrix, and calculate the adjusted characteristic matrix weight using the formula: ； Calculate the adjusted characteristic matrix to obtain a reconstructed characteristic matrix; in, represents the feature weight value of the jth feature dimension in the zth row in the adjusted feature matrix, represents the characteristic value of the corresponding position in the original characteristic cross matrix, Calculate the normalized weight of the current feature dimension in the adjusted matrix, To smoothly adjust the parameters and avoid the denominator being zero, n represents the total number of feature dimensions; S304: Reintegrate the matrix structure according to the characteristic components of the reconstructed characteristic matrix, call the correlation parameters between the transaction characteristics, calculate the interaction intensity of the characteristic dimension, screen the characteristic components with high interaction intensity, optimize the matching rules of the characteristic components, adjust the characteristic alignment method, and generate a reconstructed characteristic matching matrix.

6. The e-commerce business data analysis and evaluation method according to claim 5, characterized in that: The steps for obtaining the long-tail feature optimization matrix are specifically as follows: S401: Analyze the long-tail distribution of commodities in the e-commerce business data, call the sales frequency of commodities in the transaction records, calculate the distribution of commodity sales, screen commodities whose sales are in a low-frequency range for a long time, extract the characteristic components of the low-frequency commodities in the reconstructed characteristic matching matrix, calculate the weight values ​​of the long-tail commodities in the characteristic dimension, and establish the long-tail commodity characteristic weight matrix; S402: Calculate the sparse distribution coefficient of the long-tail product characteristic weight matrix, call the characteristic weight vector, calculate the weight ratio of multiple characteristic dimensions, select the characteristic dimensions with low weight ratio according to the discrete degree of the characteristic weight, extract the characteristic component with the lowest weight ratio, calculate the weight mean between the characteristic dimensions, call the characteristic dimension weight data below the mean, and construct a sparse characteristic weight matrix; S403: Performing gain weight adjustment on the feature dimension in the sparse feature weight matrix, using the formula: ； The gain adjustment characteristic matrix is ​​calculated; in, represents the gain-adjusted feature weight, represents the original sparse feature weight, is the gain adjustment factor, which adjusts the proportion of low-weight characteristics. Calculate the normalized total weight of the current feature dimension, To smooth the parameters and avoid extreme data interfering with the calculation; S404: Optimize the long-tail feature distribution according to the gain-adjusted feature matrix, call the feature weight values ​​of the long-tail products, screen the feature weights after gain adjustment, extract the feature components whose adjustment range exceeds the set threshold, adjust the weight proportion of low-weight features in the long-tail products, optimize the feature structure of the long-tail products, and generate a long-tail feature optimization matrix.

7. The e-commerce business data analysis and evaluation method according to claim 6, characterized in that: The steps for obtaining the commodity demand characteristic matching ratio value are specifically as follows: S501: Count the user-product matching records in the e-commerce business data, call the user transaction data, select matching items according to the number of successful transactions and evaluations, extract the characteristic vector of the corresponding product, call the long-tail characteristic optimization matrix, screen the optimized characteristic matching parameters, calculate the matching degree between the user demand vector and the product characteristic vector, and establish the product characteristic matching matrix; S502: calling the key characteristic dimensions of the product characteristic matching matrix, dynamically adjusting the matching components in the matrix, calculating the contribution weights of user needs in multiple characteristic dimensions, adjusting the matching characteristic weights according to user preferences, screening characteristic dimensions with high contribution weights, calling the adjusted matching matrix, and performing normalization transformation on the matching weights to obtain a matching characteristic adjustment matrix; S503: Calculate the matching ratio between the product and the demand, call the characteristic matching component in the matching characteristic adjustment matrix, calculate the contribution ratio of the product in multiple characteristic dimensions based on the product demand matching weight, and select the characteristic component whose ratio is higher than the set threshold, using the formula: ； Calculate product matching parameters and obtain product matching ratio matrix; in, Represents the product matching ratio, represents the feature matching weight, Represents the matching degree of the product in the u-th characteristic dimension, Calculate the normalized weights of all matching features, To smoothly adjust the parameters, Represents the expected demand value of the product in the uth characteristic dimension, Represents the actual matched characteristic value, is the adjustment factor, Calculate the normalized deviation of the expected value of the requirement, where n represents the total number of feature dimensions; S504: calling the product matching ratio matrix, integrating multiple matching characteristic components, calculating the weighted matching ratio of products in the differentiated user demand dimension, calling the weighted matching ratio data, screening products whose matching degree exceeds a set threshold, adjusting the weights of products with lower matching ratios, optimizing matching rules, and generating product demand characteristic matching ratio values.

Images