Customer behavior analysis method and system based on retail data

Abstract

The invention relates to the technical field of data mining, in particular to a customer behavior analysis method and system based on retail data, and the method comprises the following steps: extracting a numerical set of customer purchase frequency, single consumption amount and shopping duration based on a customer group behavior feature matrix, and calling matrix operation to analyze a multi-feature distribution condition item by item. According to the invention, through extracting the purchase frequency, the consumption amount and the shopping duration in the customer behaviors, combining the analysis of the offset value and the item-by-item comparison of the distribution interval, accurately identifying the abnormal characteristics, and analyzing the offset value in the multi-period frequency change trend, the period characteristics of the behavior change are clearer, and the frequency intersection analysis of the commodity combination purchase behaviors is more accurate. Reconstruction accuracy of feature distribution and hierarchical marking of regional customer coverage data are improved, the assessment ability of regional feature distribution and an offset ratio is refined, and the precision and practicability of the operations in feature distribution capture, multi-period behavior analysis and regional data comparison are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data mining, and in particular to a customer behavior analysis method and system based on retail data. Background Art

[0002] The field of data mining technology includes technical methods that use computer technology to extract hidden useful information and knowledge from large amounts of data. Its core content includes data collection, storage, preprocessing, analysis and modeling. Data mining technology explores potential patterns and relationships in data through statistical analysis, pattern recognition and machine learning, providing a basis for decision support. This technology is widely used in finance, retail, medical care, communications and other fields. It mainly solves the needs of information mining in massive data. It has the characteristics of high efficiency, automated processing and intelligent analysis, and is an important part of the big data technology system.

[0003] Among them, the customer behavior analysis method refers to studying the customer's purchase behavior pattern by analyzing the purchase records, transaction information and other related data in the retail data. The patent subject mainly covers specific matters such as classification and clustering analysis of customer data, feature extraction, and behavior prediction. Usually, data preprocessing, behavior feature modeling, sequence analysis, and behavior rule mining are used to extract customer behavior features through time series analysis and association pattern mining of customer data to realize the construction of a method system for customer behavior analysis.

[0004] When analyzing customer data, existing technologies are not able to capture feature shifts and distinguish abnormal behaviors, making it difficult to accurately identify important feature change areas. Single or limited-dimensional data analysis models fail to fully tap the complex correlations between features, resulting in weak behavioral pattern classification. The ability to dynamically capture changes in time period behavior is not strong, and ignoring important feature changes in specific time periods affects prediction accuracy. Regionalized analysis lacks hierarchical labeling of shifted features, resulting in insufficient granularity in regional comparison results. These limitations reduce the ability to fully analyze complex customer behaviors and apply them to specific scenarios. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a customer behavior analysis method and system based on retail data.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: a customer behavior analysis method based on retail data, comprising the following steps: S1: Based on the customer group behavior feature matrix, extract the numerical set of customer purchase frequency, single consumption amount and shopping time, call matrix operation to analyze the multi-feature distribution item by item, separate the numerical set with the offset value greater than the overall feature change in the distribution, and generate the offset behavior feature value; S2: Based on the offset behavior characteristic value, extract the interval distribution data of sales and customer purchase amount, compare the offset value with the distribution interval data item by item, calculate the offset value proportion and cumulative characteristic value within the interval, compare and classify the change of the ratio with other interval data, and generate the offset interval proportion value; S3: Based on the offset interval ratio, call the fluctuation range of the time period purchase data, analyze the frequency change trend of the offset value in multiple time periods, calculate the cumulative difference between the offset frequency and the overall trend, extract the time period characteristics according to the frequency change amount and integrate the frequency weight results to generate the time period deviation frequency value; S4: extracting frequency distribution data in the commodity combination purchase behavior based on the time period deviation frequency value, calculating the intersection feature of the frequency change value and the deviation data, classifying and analyzing the commodity combination according to the frequency change range, and reconstructing and calculating the feature distribution to generate a reconstructed offset feature value; S5: Based on the reconstructed offset feature value, call the regional customer purchase coverage data, extract the feature distribution and offset feature distribution results within the regional coverage area, mark and layer the offset proportion data within the coverage area item by item, and establish the regional offset proportion mark value.

[0007] The offset behavior characteristic values ​​include purchase frequency offset values, single consumption amount offset values, and shopping duration offset values; the offset interval proportion values ​​include sales offset proportion values, customer purchase amount offset proportion values, and interval cumulative characteristic value offset proportions; the time period deviation frequency values ​​include frequency change characteristic values, time period frequency weight results, and cumulative difference frequency values; the reconstructed offset characteristic values ​​include commodity combination frequency change values, frequency deviation intersection characteristic values, and reconstructed characteristic distribution results; the regional offset proportion mark values ​​include characteristic distribution values ​​in the coverage area, offset proportion mark values, and layered processing results.

[0008] As a further solution of the present invention, the step of obtaining the offset behavior characteristic value is specifically as follows: S101: Call the numerical set of the customer's purchase frequency, single consumption amount and shopping time, establish a behavior feature matrix, calculate the mean vector and covariance matrix of multiple feature items through matrix transformation, and generate a feature distribution parameter set; S102: Based on the feature distribution parameter set, a standardized Euclidean distance between each feature point and the mean vector is calculated, a feature point set greater than a set offset threshold is selected, and an offset feature point set is generated; S103: Call the offset feature point set using the formula: ; Calculate the offset behavior strength of all feature points and generate offset behavior feature values; in, Representative The offset behavior strength of feature points, Representative The value of the feature point, represents the mean of the feature items, represents the standard deviation of the feature item, represents the weight of the feature item, Representative The characteristic value of the dimension, Representative The mean of the dimension features, It represents the dimension of the feature matrix, that is, the number of customer behavior features.

[0009] As a further solution of the present invention, the step of obtaining the offset interval ratio value is specifically as follows: S201: extracting interval distribution data of sales and customer purchase amounts, calling the offset behavior characteristic value, extracting offset characteristic values ​​corresponding to multiple intervals by matching the interval distribution data with the offset behavior characteristic value, calculating the offset behavior ratio of each interval based on the total interval characteristic value, and generating the interval offset ratio; S202: calling the interval offset ratio, calculating the offset ratios of adjacent intervals one by one, analyzing the offset feature difference values ​​between intervals, taking the average of the offset feature values ​​of all intervals as a benchmark, and using the formula: ; Generate interval offset value ratio data; in, Representative The offset value ratio of the interval, Representative The offset eigenvalue of the interval, Representative The offset eigenvalue of the interval, represents the mean of all interval-shifted eigenvalues, Indicates the total number of intervals divided in the entire data distribution; S203: calling the interval offset value ratio data, comparing the cumulative characteristic value ratio with the interval data volume, calculating the difference between the cumulative characteristic value and the offset value ratio interval by interval, classifying and identifying the offset interval, and generating the offset interval ratio value.

[0010] As a further solution of the present invention, the step of obtaining the time period deviation frequency value is specifically as follows: S301: Based on the offset interval ratio value, call the purchase data fluctuation range of multiple time periods, extract the upper and lower boundary values ​​of the purchase data of multiple time periods, calculate the offset behavior feature values ​​in multiple time periods, filter the feature points that exceed the time period mean deviation threshold, analyze the frequency distribution of the offset behavior in the differentiated time periods, and generate the time period data fluctuation analysis results; S302: Calculate the frequency of the offset value of each period from the period data fluctuation analysis results, using the formula: ; Calculate the cumulative difference value between the offset frequency and the overall trend to generate the offset frequency cumulative difference value data; in, Representative The offset frequency of the time period, Representative The offset value of the time point, Representative The shifted mean of the period, Representative The standard deviation of the shift for the period, represents the total number of time points captured; S303: Call the offset frequency cumulative difference value data, extract the time period with prominent offset frequency changes based on the fluctuation range of multiple time periods, calculate the characteristic frequency change amplitude of the time period, call the frequency weight parameter of the corresponding time period, calculate the weighted ratio of the frequency weight and the characteristic frequency change amplitude, and generate the time period deviation frequency value.

[0011] As a further solution of the present invention, the step of obtaining the reconstruction offset feature value is specifically: S401: extracting frequency distribution data of commodity combination purchase behavior based on the time period deviation frequency value, calling the purchase record of the commodity combination, calculating the number of occurrences of each combination, screening the combination that deviates from the average purchase frequency, classifying according to the purchase trend and combination stability parameter, and generating a frequency distribution data set; S402: calling the frequency distribution data set, analyzing the purchase frequency of each commodity combination, calculating the intersection with the time period deviation frequency value, identifying the overlapping features of the purchase frequency and deviation data, screening commodity combinations with similar purchase patterns, calculating the frequency change rate of the commodity combinations and classifying them, and generating intersection feature analysis results; S403: Call the intersection feature analysis result, classify according to the frequency change range and deviation data, and calculate the feature distribution of the product combination using the formula: ; Reconstruct and calculate the characteristic distribution to generate the reconstructed offset characteristic value; in, represents the reconstructed offset eigenvalue, Representative The frequency change value of a product combination, Representative Deviation data of product combinations, Represents the maximum value of all product combination deviation data. Represents the total number of all product combinations.

[0012] As a further solution of the present invention, the step of obtaining the area offset ratio mark value is specifically as follows: S501: Based on the reconstructed offset feature value, call the regional customer purchase coverage data, extract the statistical information of multiple purchase points in the region, analyze the purchase frequency and purchase amount distribution within the coverage area, calculate the standard deviation and mean of the coverage area, evaluate the degree of dispersion in the feature distribution, and generate regional feature distribution data; S502: calling the regional feature distribution data, filtering the offset ratio data in the coverage area, marking the size of the offset value item by item by comparing the offset value of each data point with the statistical benchmark of the area, dividing the data into a high offset group and a low offset group according to the offset amplitude, and generating an offset ratio marking list; S503: Call the offset ratio mark list, using the formula: ; Compare the offset values ​​of the differentiated areas at the same scale to generate a regional offset ratio mark value; in, .

[0013] A customer behavior analysis system based on retail data, the customer behavior analysis system based on retail data is used to perform the customer behavior analysis method based on retail data, the system comprises: The customer feature extraction module extracts the data set of purchase frequency, single consumption amount and shopping time based on the customer group behavior feature matrix, calculates the value distribution range of multiple features, compares the offset of multiple sets of values ​​with the overall feature change range, filters the value set whose offset exceeds the feature change range, and generates the offset behavior feature value; The sales interval analysis module extracts interval distribution data of sales and customer purchase amount based on the offset behavior characteristic value, compares the offset value with the interval value distribution item by item, calculates the proportion of the offset value to the value in the interval and the cumulative characteristic value, analyzes the difference interval value change ratio and performs classification operation, and generates the offset interval proportion value; The time period frequency analysis module extracts the fluctuation range of purchase data in multiple time periods based on the offset interval ratio, calculates the frequency distribution and trend change of the offset values ​​in multiple time periods, accumulates the difference between the offset frequency and the overall trend in the time period, selects the time period characteristics according to the frequency change amplitude and integrates the frequency weight to generate the time period deviation frequency value; The commodity combination feature reconstruction module extracts the commodity combination purchase frequency distribution data based on the time period deviation frequency value, calculates the intersection feature of the frequency change value and the deviation data, analyzes the intersection feature distribution range and the frequency change amplitude, reconstructs the commodity combination feature distribution, and generates a reconstructed offset feature value; The regional coverage marking module extracts the feature distribution data within the regional customer purchase coverage based on the reconstructed offset feature value, compares the offset features within the region with the overall feature distribution, marks the offset proportion within the region item by item and performs hierarchical classification to establish the regional offset proportion marking value.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are: In the present invention, by extracting the purchase frequency, consumption amount and shopping duration in customer behavior, combined with the analysis of offset values ​​and item-by-item comparison of distribution intervals, abnormal features can be accurately identified. The analysis of offset values ​​in the frequency change trend of multiple time periods makes the time period characteristics of behavior changes clearer. The frequency intersection analysis of commodity combination purchase behavior improves the accuracy of feature distribution reconstruction. The hierarchical labeling of regional customer coverage data refines the evaluation capabilities of regional feature distribution and offset ratio. These operations improve accuracy and practicality in feature distribution capture, multi-time period behavior analysis and regional data comparison. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a schematic diagram of the workflow of the present invention; Figure 2 A flow chart of the steps for obtaining the characteristic value of the offset behavior of the present invention; Figure 3 This is a flow chart of the steps for obtaining the offset interval ratio value of the present invention; Figure 4 A flow chart of the steps for obtaining the time period deviation frequency value of the present invention; Figure 5 A flow chart of the steps for obtaining the reconstructed offset feature value of the present invention; Figure 6 The present invention is a flowchart of the steps for obtaining the region offset ratio mark value. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with 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 intended to limit the present invention.

[0017] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are 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 cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0018] Embodiment 1 See also Figure 1 The present invention provides a technical solution: a customer behavior analysis method based on retail data, comprising the following steps: S1: Based on the customer group behavior feature matrix, extract the numerical set of customer purchase frequency, single consumption amount and shopping time, call matrix operation to analyze the multi-feature distribution item by item, separate the numerical set with the offset value greater than the overall feature change in the distribution, and generate the offset behavior feature value; S2: Based on the offset behavior feature value, extract the interval distribution data of sales and customer purchase amount, compare the offset value with the distribution interval data item by item, calculate the offset value proportion and cumulative feature value within the interval, compare and classify the change of the ratio with other interval data, and generate the offset interval proportion value; S3: Based on the offset interval ratio, call the fluctuation range of the time period purchase data, analyze the frequency change trend of the offset value in multiple time periods, calculate the cumulative difference between the offset frequency and the overall trend, extract the time period characteristics according to the frequency change amount and integrate the frequency weight results to generate the time period deviation frequency value; S4: Based on the time period deviation frequency value, extract the frequency distribution data in the commodity combination purchase behavior, calculate the intersection feature of the frequency change value and the deviation data, classify and analyze the commodity combination according to the frequency change range, and reconstruct and calculate the feature distribution to generate the reconstructed offset feature value; S5: Based on the reconstructed offset feature value, call the regional customer purchase coverage data, extract the feature distribution and offset feature distribution results within the regional coverage area, mark and layer the offset proportion data within the coverage area item by item, and establish the regional offset proportion mark value.

[0019] The offset behavior characteristic values ​​include the purchase frequency offset value, the single consumption amount offset value, and the shopping time offset value. The offset interval proportion values ​​include the sales offset proportion value, the customer purchase amount offset proportion value, and the interval cumulative characteristic value offset proportion. The time period deviation frequency value includes the frequency change characteristic value, the time period frequency weight result, and the cumulative difference frequency value. The reconstructed offset characteristic value includes the commodity combination frequency change value, the frequency deviation intersection characteristic value, and the reconstructed characteristic distribution result. The regional offset proportion mark value includes the characteristic distribution value in the coverage area, the offset proportion mark value, and the layered processing result.

[0020] See also Figure 2 , the specific steps for obtaining the offset behavior characteristic value are: S101: Call the numerical set of the customer's purchase frequency, single consumption amount and shopping time, establish a behavior feature matrix, calculate the mean vector and covariance matrix of multiple feature items through matrix transformation, and generate a feature distribution parameter set; Construct a time series and calculate the mean and standard deviation of each feature. Set a time window, perform sliding statistics on customer consumption behavior in different time windows, calculate the number of purchases, total consumption amount and cumulative shopping time in each window, and normalize these data to construct a basic behavior feature matrix. Each row of the matrix represents a time window, and each column represents a feature type. Perform covariance calculation on this matrix to obtain the correlation matrix between features. Combined with PCA dimensionality reduction technology, extract principal components and calculate their contribution rates. Principal components with contribution rates higher than a certain threshold are used for subsequent analysis. Subsequently, the Mahalanobis distance is used to calculate the degree of outliers for each data point in order to preliminarily screen abnormal consumption behaviors and generate a set of feature distribution parameters.

[0021] S102: Based on the feature distribution parameter set, a standardized Euclidean distance between each feature point and the mean vector is calculated, a feature point set greater than a set offset threshold is selected, and an offset feature point set is generated; For multi-dimensional feature data, the weighted Euclidean distance is used to calculate the degree of deviation of a single data point from the mean vector. The calculation formula is: ,in, Indicates The weighted Euclidean distance between the data points and the mean vector, For the The data point in The values ​​in the dimensions, For the The mean of the dimensions, is the weight coefficient of this dimension, which is determined by the feature contribution rate and calculated for all data points And set the offset threshold for all Data points with values ​​greater than the threshold are marked to generate a set of offset feature points.

[0022] S103: Call the offset feature point set using the formula: ; Calculate the offset behavior strength of all feature points and generate offset behavior feature values; in, Representative The offset behavior strength of feature points, Representative The value of the feature point, represents the mean of the feature items, represents the standard deviation of the feature item, represents the weight of the feature item, Representative The characteristic value of the dimension, Representative The mean of the dimension features, It represents the dimension of the feature matrix, that is, the number of customer behavior features.

[0023] formula: ; The benefit of the formula is that by combining the standardized deviation calculation and the weighted Euclidean distance to measure the degree of abnormality of data points, the offset behavior feature value can not only reflect the deviation of a single feature, but also comprehensively consider the overall offset trend of all features, thereby improving the accuracy of anomaly identification.

[0024] Detailed explanation of the formula and the process of formula calculation and derivation: The numerical sets for customer purchase frequency, single consumption amount, and shopping duration are set as follows: , and its corresponding mean: , standard deviation: , weights: , calculate the standardized deviation: ; ; ; Compute the weighted Euclidean distance: ; ; ; Calculate the offset behavior characteristic value: ; The result shows that the customer's consumption behavior deviates significantly from the average of overall consumption behavior, especially the deviation of the single consumption amount is large. Therefore, it can be regarded as an abnormal consumption behavior. This value can be used for further behavior prediction, risk assessment or marketing strategy adjustment.

[0025] See also Figure 3 , the specific steps for obtaining the offset interval ratio value are: S201: extracting interval distribution data of sales and customer purchase amounts, calling offset behavior feature values, matching interval distribution data with offset behavior feature values, extracting offset feature values ​​corresponding to multiple intervals, calculating the offset behavior ratio of each interval based on the total interval feature value, and generating interval offset ratios; First, calculate the mean and standard deviation of sales data in all intervals. The mean calculation formula is as follows: ,in, Representative Sales data, Represents the total amount of data, and the standard deviation is calculated as follows: After calculating the mean and standard deviation, the sales data is transformed using the standardized formula to obtain a standardized sales data set. Then, the distribution of customer purchase amounts is analyzed, and the offset behavior feature value is called to calculate the mean and standard deviation of all data points of customer purchase amounts. The standardized transformation is performed according to the same standardized formula. Then, an interval matrix of sales and customer purchase amounts is established, and the cumulative sum of customer purchase amounts in each interval is calculated in turn, and normalized. The calculation formula is as follows: ,in, Representative The amount of customer purchases within a certain interval, Represents the number of customer data in the current interval, Represents the total amount of data in all intervals. After normalization, the offset behavior eigenvalue is compared with the normalized interval matrix. The deviation degree between the offset eigenvalue and the interval mean is calculated for each interval. The calculation formula is as follows: in, Representative The offset eigenvalue of the interval, Represents the mean purchase amount of all interval customers, Represents the standard deviation of the customer's purchase amount. By calculating the offset degree of all intervals, they are arranged into an offset comparison matrix. Then, the offset degree distribution of all interval data is counted, the offset degree values ​​of all intervals are obtained, and the total eigenvalue of the interval is called for calculation. The ratio of the offset eigenvalue of each interval to the total eigenvalue of the interval is compared to generate the interval offset ratio.

[0026] S202: Call the interval offset ratio, calculate the offset ratios of adjacent intervals one by one, analyze the offset feature difference values ​​between intervals, take the mean of the offset feature values ​​of all intervals as the benchmark, and use the formula: ; Generate interval offset value ratio data; in, Representative The offset value ratio of the interval, Representative The offset eigenvalue of the interval, Representative The offset eigenvalue of the interval, represents the mean of all interval-shifted eigenvalues, Indicates the total number of intervals divided in the entire data distribution; formula: ; The benefit of the formula is that it reduces the impact of extreme values ​​on the calculation results through square root conversion, and introduces the processing of adding one in the denominator to ensure that the data will not cause calculation errors due to zero values, making the weight of the offset interval smoother and more reasonable.

[0027] Detailed explanation of the formula and the process of formula calculation and derivation: In this formula, Representative The proportion of offset values ​​in the interval, Representative The offset eigenvalue of the interval, represents the mean of all interval-shifted eigenvalues, Representative The offset eigenvalue of the interval, represents the total number of intervals, the square root is used to calculate the absolute deviation, and the denominator ensures that the calculation will not fail due to the denominator being zero.

[0028] Parameter value acquisition: Obtained by statistically analyzing the distribution data of sales and customer purchase amounts, and calculating the deviation behavior characteristic value within the interval; The calculation formula is as follows: ; Calculation example: Assume that in a sales data, the offset behavior feature value dataset is , calculate the mean of all interval-shifted eigenvalues: ; Calculate the offset value ratio of the first interval: ; Calculate the denominator: ; Calculate the specific value: ; ; ; Calculate the final result: ; The result shows that the offset value proportion of the first interval is 0.129, which has a relatively small weight among all intervals. The comparison with the offset value proportions of other intervals will be used for further classification and analysis.

[0029] S203: Call the interval offset value ratio data, compare the cumulative characteristic value ratio with the interval data volume, calculate the difference between the cumulative characteristic value and the offset value ratio interval by interval, classify and identify the offset interval, and generate the offset interval ratio value.

[0030] Count the percentage of offset values ​​in all intervals, establish an interval offset statistics table, and calculate the cumulative sum of the percentage of offset values ​​in each interval. The calculation formula is as follows: ,in, Representative The cumulative offset value ratio of the interval, Representative The offset value ratio of each interval is calculated, and then the change trend of the offset value ratio of all intervals is calculated. The change rate calculation formula is as follows: ,in, Representative The rate of change of the offset in the interval, Represents the proportion of the cumulative offset value of the previous interval, analyzes the distribution of the change rate of each interval, compares the changes in the proportion of the cumulative eigenvalues, classifies and identifies the intervals with significant offsets, and generates the offset interval proportion value.

[0031] See also Figure 4 , the specific steps for obtaining the time period deviation frequency value are: S301: Based on the offset interval ratio value, call the purchase data fluctuation range of multiple time periods, extract the upper and lower boundary values ​​of the purchase data of multiple time periods, calculate the offset behavior feature values ​​in multiple time periods, filter the feature points that exceed the time period mean deviation threshold, analyze the frequency distribution of the offset behavior in the differentiated time periods, and generate the time period data fluctuation analysis results; The purchase data of each time period is statistically processed, and the data is divided into multiple intervals. Each interval represents a continuous time period. The mean, standard deviation and skewness of the data distribution of each time period are calculated. The upper and lower limits of the fluctuation range of each time period are determined based on the standard deviation. The characteristic value of the deviation behavior in each time period is calculated, and the data points are compared with the upper and lower boundaries. The abnormal data points that exceed the boundaries are screened out, and the abnormal data points are classified according to the time distribution. The cumulative frequency distribution of the deviation points is further calculated to evaluate the changes in the abnormal transaction frequency in different time periods. The degree of abnormality of each time period is determined by the cumulative deviation point ratio method. Based on the calculated abnormal ratio threshold, the data of each time period is divided into The degree of deviation is classified, and the proportion of the number of deviation points in the interval is used as the judgment basis. The mean of the proportion of deviation points is calculated, and the deviation value of the proportion of the number of deviation points in each time period is compared with the overall proportion of deviation points. The time periods with deviation values ​​greater than the set threshold are recorded, and the recorded proportion of time period deviation points is calculated as the abnormal fluctuation degree of the overall trading data. The deviation trend of the fluctuation range within the time period is summarized, and the weight proportion within the time period is calculated based on the data of the time period fluctuation trend and combined with the overall trading volume. The calculation results of the deviation characteristics of each time period are corrected using the time period weight proportion, and the calculation method of the deviation characteristics of different time intervals is adjusted to eliminate the influence of abnormal deviation points on the overall trend and generate the time period data fluctuation analysis results.

[0032] S302: Calculate the frequency of the offset value of each period from the period data fluctuation analysis results using the formula: ; Calculate the cumulative difference value between the offset frequency and the overall trend to generate the offset frequency cumulative difference value data; in, Representative The offset frequency of the time period, Representative The offset value of the time point, Representative The shifted mean of the period, Representative The standard deviation of the shift for the period, represents the total number of time points captured; formula: ; The benefit of the formula is that by accumulating the absolute difference of the offset value at each time point within the time period and combining it with the standard deviation normalization calculation, the stability of the offset frequency calculation is improved, the abnormal offsets between different time periods can be compared more reasonably, the influence of extreme values ​​can be avoided, and the sensitivity to offset fluctuations is enhanced.

[0033] Detailed explanation of the formula and the process of formula calculation and derivation: Representative The offset value of a time point is calculated based on the fluctuation of actual transaction data. It is calculated period by period, with the change in transaction amount per hour as the calculation basis. Representative The mean deviation of a period is obtained by taking the arithmetic mean of the deviation values ​​in all periods. The calculation formula is as follows: ; in Represents the total number of time points in this period, Representative The standard deviation of the time period is calculated as follows: ; Substitute the data into the calculation and set the offset data within a certain period of time , then its mean is calculated as follows: ; Calculate the standard deviation: ; ; Enter the calculation formula: ; ; ; The result shows that the offset frequency value of this period is 1.16, which is slightly higher than the historical offset frequency average of 1.0 for this period, indicating that the purchase data of this period fluctuated greatly and was a time period with prominent offset characteristics.

[0034] S303: Call the offset frequency cumulative difference value data, extract the time period with prominent offset frequency changes based on the fluctuation range of multiple time periods, calculate the characteristic frequency change amplitude of the time period, call the frequency weight parameter of the corresponding time period, calculate the weighted ratio of the frequency weight and the characteristic frequency change amplitude, and generate the time period deviation frequency value.

[0035] The frequency difference between different time periods is calculated based on the change amplitude, and the time periods with frequency difference greater than the calculation threshold are screened. The characteristic frequency change amplitude of the time period is further calculated, and the weight factor of the time period is calculated in combination with the trading volume of each time period. The weight factor is normalized to ensure that the influencing factors of different time periods are comparable, and the weighted ratio of the frequency weight and the characteristic frequency change amplitude is calculated. The frequency offset value after weight correction is compared with the historical average value, and the degree of deviation of the time period is evaluated. Reasonable boundaries are set to classify the offset frequency of each time period, and the classified time period data is integrated with the overall trend data. The deviation results of different time periods are integrated and calculated, and the interval with the highest degree of time period deviation is taken as the abnormal time period to generate the time period deviation frequency value.

[0036] See also Figure 5 , the specific steps for obtaining the reconstructed offset eigenvalues ​​are: S401: extract frequency distribution data in the commodity combination purchase behavior based on the time period deviation frequency value, call the purchase record of the commodity combination, calculate the number of occurrences of each combination, filter out the combination that deviates from the average purchase frequency, classify according to the purchase trend and combination stability parameters, and generate a frequency distribution data set; First, the purchase records are divided into multiple intervals according to the time dimension. Each interval corresponds to the purchase frequency data of a commodity combination. The number of occurrences of each commodity combination in each interval is calculated. The basic purchase frequency of each commodity combination is determined based on the historical purchase data of the commodity combination and the time period deviation frequency value. According to the deviation data in the time period, the purchase frequency and the time period deviation frequency value are matched and calculated to screen out the commodity combinations that deviate from the mean. The purchase anomalies of the commodity combination are determined by calculating the standard deviation of the deviation from the mean. When classifying commodity combinations with abnormal purchase frequencies, they are divided into three categories: high-frequency purchases, low-frequency purchases, and fluctuating purchases according to the fluctuation of the purchase frequency in the time period. Among them, high-frequency purchase product combinations refer to combinations that appear more than a certain threshold in multiple time periods, low-frequency purchase product combinations refer to combinations that appear significantly less than the threshold, and fluctuating purchase product combinations refer to combinations whose appearances vary greatly in different time periods. According to the distribution of purchase frequency, the purchase stability of each product combination is calculated, and the coefficient of variation is used to calculate the stability parameter. The higher the coefficient of variation of the product combination, the stronger the volatility of its purchase frequency. The product combinations with strong volatility are selected, and their frequency change values ​​are calculated. The degree of change in the purchase behavior of the product combination is quantified by calculating the purchase frequency change ratio of adjacent time periods to obtain the frequency distribution data set.

[0037] S402: Call the frequency distribution data set, analyze the purchase frequency of each commodity combination, calculate the intersection with the time period deviation frequency value, identify the overlapping features of the purchase frequency and deviation data, screen commodity combinations with similar purchase patterns, calculate the frequency change rate of the commodity combinations and classify them, and generate the intersection feature analysis results; First, it is necessary to determine the purchase frequency distribution characteristics of each commodity combination within the time period, use the cumulative frequency to calculate the cumulative number of occurrences of the commodity combination in each time period, normalize the purchase frequency of each time period so that the frequency values ​​of different time periods can be compared horizontally, calculate the matching degree of the normalized purchase frequency and the time period deviation frequency value, use the Euclidean distance to calculate the difference between the purchase frequency and deviation data of each commodity combination, screen out commodity combinations with significant frequency changes through the difference threshold, calculate the frequency change trend of the commodity combination, use the sliding window to calculate the frequency change rate of the commodity combination, and classify the commodity combinations based on the calculation results. In the classification process, according to the different distribution ranges of the frequency change rate, the commodity combinations are divided into three categories: high change frequency, medium change frequency and low change frequency. The commodity combination with high change frequency refers to the commodity combination that fluctuates significantly during the peak period of the time period deviation frequency value, the commodity combination with medium change frequency refers to the commodity combination that changes slowly within the fluctuation range of the time period deviation frequency value, and the commodity combination with low change frequency refers to the combination that maintains a stable purchase frequency in all time periods. Filter commodity combinations with similar purchase patterns, calculate the frequency change rate of the commodity combination and classify it, and obtain the intersection feature analysis results.

[0038] S403: Call the intersection feature analysis results, classify according to the frequency change range and deviation data, and calculate the characteristic distribution of the product combination using the formula: ; Reconstruct and calculate the characteristic distribution to generate the reconstructed offset characteristic value; in, represents the reconstructed offset eigenvalue, Representative The frequency change value of a product combination, Representative Deviation data of product combinations, Represents the maximum value of all product combination deviation data. Represents the total number of all product combinations.

[0039] formula: ; The formula is useful in that by incorporating the frequency change value and product mix deviation data , to normalize the maximum deviation value As an adjustment factor, it improves the comparability of the characteristic distribution of different commodity combinations and avoids the influence of commodity combinations with large data volume or small variation on the overall deviation characteristic construction.

[0040] Detailed explanation of the formula and the process of formula calculation and derivation: in, represents the reconstructed offset eigenvalue, Representative The frequency change value of a product combination, Representative Deviation data of product combinations, Represents the maximum value among all product combination deviation data.

[0041] Calculate the specific values ​​of each parameter: Frequency change value By calculating the change rate of purchase frequency in adjacent time periods, the purchase frequency of a certain product combination in multiple time periods is set to {12, 15, 18, 14, 20}, and the change rate of frequency in adjacent time periods is calculated: ; Product mix deviation data By calculating the mean deviation of the product combination in all time periods, it is found that the mean of a product combination is 15, then: ; Calculate the maximum value of all product combination deviation data, set the maximum deviation value of all combinations to 5, then ; Substitute the formula to calculate the reconstruction offset eigenvalue: ; ; ; This result shows that the calculated The value is used to measure the overall impact of the change in the purchase frequency of the product combination and the deviation characteristics. A lower value means that the frequency change of the product combination is less affected by the deviation, while a higher value means that the frequency change of the product combination is less affected by the deviation. The value indicates that the purchase behavior of the product combination has a large deviation in different time periods. This result can be further used to optimize the sales strategy of the product combination and obtain the reconstruction offset feature value.

[0042] See also Figure 6 , the specific steps for obtaining the regional offset ratio mark value are: S501: Based on the reconstructed offset feature value, call the regional customer purchase coverage data, extract the statistical information of multiple purchase points in the region, analyze the purchase frequency and purchase amount distribution within the coverage area, calculate the standard deviation and mean of the coverage area, evaluate the degree of dispersion in the feature distribution, and generate regional feature distribution data; Determine the customer distribution in each region, collect the source of purchase data, calculate the transaction frequency and transaction amount of each purchase point by region, use time series analysis to identify the cyclical consumption characteristics of each region, calculate the transaction fluctuation range of each region within a specific time range, screen outliers in transaction amount and transaction frequency, calculate the deviation of outliers compared to the regional mean, and obtain the standard deviation of transaction characteristics of each region after normalization. Use hierarchical clustering method to identify consumption deviation categories in different regions, mark high volatility areas and low volatility areas, calculate the transaction center of gravity of these areas, adjust weight parameters to optimize regional influencing factors, establish regional feature distribution matrix based on consumption offset data, calculate the offset center points of each area within the coverage area, match historical purchase trend data, conduct time series regression analysis, further confirm the fluctuation pattern of purchase data in each region, screen out areas with high transaction stability and large transaction volatility, use different weights to calculate offset characteristics for different categories of areas, normalize the offset calculation results of all areas, and generate regional feature distribution data.

[0043] S502: Calling regional feature distribution data, filtering offset ratio data in the coverage area, marking the offset values ​​one by one by comparing the offset value of each data point with the statistical benchmark of the area, dividing the data into a high offset group and a low offset group according to the offset amplitude, and generating an offset ratio marking list; Extract the offset ratio data in the coverage area, normalize the offset values ​​of all areas, build a regional offset matrix, map the regional customer purchase behavior to the feature space by calculating the offset distance of each data point, calculate the weighted mean of the purchase frequency and purchase amount in each area, calculate the offset distribution of each data point, set the offset ratio threshold based on the offset statistical data of each area, filter out data points with offset values ​​higher than the set threshold, and classify them into high-offset areas. At the same time, filter out areas with offset values ​​lower than the set threshold and classify them into low-offset areas. Use the hierarchical clustering method to stratify the offset characteristics of each area, further calculate the transaction center of gravity of the high-offset area, filter out the area with the highest offset, calculate its standardized deviation from the overall regional mean, compare the mark value of the high-offset area with the low-offset area, adjust the threshold range, improve the accuracy of offset identification, calculate the offset change rate of each area, combine time series analysis to evaluate the trend of the offset characteristics of each area, calculate the offset ratio of each area, and correct the marks of the high-offset area and low-offset area to generate an offset ratio mark list.

[0044] S503: Call the offset ratio mark list, using the formula: ; Compare the offset values ​​of the differentiated areas at the same scale to generate a regional offset ratio mark value; in, .

[0045] formula: ; The formula is beneficial in that it normalizes the offset data and introduces a regional weight factor To measure the impact of different regions, the calculation results can more accurately reflect the differences in offset distribution between regions and improve the adaptability and accuracy of offset calculation.

[0046] Detailed explanation of the formula and the process of formula calculation and derivation: Parameter acquisition and calculation process: How to obtain: weight factor Reflects the importance of the region, adjusted by the weight of historical sales data, calculated by: ,in, Representative Sales volume in the region, Represents the total sales of all regions. The weight factor is dynamically calculated based on the sales proportion of each region.

[0047] How to obtain: offset data By calculating the deviation between the actual purchase frequency and the expected purchase frequency in each region, the calculation formula is: ,in, Representative The actual purchase frequency of the region, Represents the expected purchase frequency in this area, which is obtained by regression analysis of historical purchase data.

[0048] How to obtain: offset mean It is calculated by averaging all the region offset values: ; How to obtain: Standard deviation Calculated from the offset data of all regions: ; Numerical assignment calculation process: Set the sales volume of the three regions to be , , , calculate the weight factor: ; ; ; Set the actual purchase frequency and expected purchase frequency in the region: ; Calculate the offset value: ; Compute the shifted mean: ; Calculate the standard deviation: ; ; ; Calculate the offset ratio mark value: ; ; ; The result shows that the regional offset ratio marker value is , this value is closer to 1, indicating that there is a certain degree of deviation in the purchasing behavior of each region, but it has not reached the abnormal range. Subsequently, it can be further determined whether some regions need to make marketing adjustments or re-evaluate the purchasing model according to the set deviation threshold to generate a regional deviation ratio mark value.

[0049] A customer behavior analysis system based on retail data, the customer behavior analysis system based on retail data is used to execute the customer behavior analysis method based on retail data, the system comprises: The customer feature extraction module extracts the data set of purchase frequency, single consumption amount and shopping time based on the customer group behavior feature matrix, calculates the value distribution range of multiple features, compares the offset of multiple sets of values ​​with the overall feature change range, filters the value set whose offset exceeds the feature change range, and generates the offset behavior feature value; The sales interval analysis module extracts the interval distribution data of sales and customer purchase amount based on the offset behavior characteristic value, compares the offset value and interval value distribution item by item, calculates the proportion of the offset value to the value in the interval and the cumulative characteristic value, analyzes the change ratio of the differentiated interval value and performs classification operation to generate the offset interval proportion value; The time period frequency analysis module extracts the fluctuation range of purchase data in multiple time periods based on the offset interval ratio, calculates the frequency distribution and trend changes of the offset values ​​in multiple time periods, accumulates the difference between the offset frequency and the overall trend in the time period, selects the time period characteristics according to the frequency change amplitude and integrates the frequency weight to generate the time period deviation frequency value; The commodity combination feature reconstruction module extracts the commodity combination purchase frequency distribution data based on the time period deviation frequency value, calculates the intersection feature of the frequency change value and the deviation data, analyzes the distribution range of the intersection feature and the frequency change amplitude, reconstructs the commodity combination feature distribution, and generates the reconstruction offset feature value; The regional coverage marking module is based on the reconstruction of the offset feature value, extracts the feature distribution data within the regional customer purchase coverage, compares the offset features within the region with the overall feature distribution, marks the offset proportion within the region item by item and performs hierarchical classification to establish the regional offset proportion marking value.

[0050] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A customer behavior analysis method based on retail data, characterized in that: The following steps are involved: S1: Based on the customer group behavior feature matrix, extract the numerical set of customer purchase frequency, single consumption amount and shopping time, call matrix operation to analyze the multi-feature distribution item by item, separate the numerical set with the offset value greater than the overall feature change in the distribution, and generate the offset behavior feature value; S2: Based on the offset behavior characteristic value, extract the interval distribution data of sales and customer purchase amount, compare the offset value with the distribution interval data item by item, calculate the offset value proportion and cumulative characteristic value within the interval, compare and classify the change of the ratio with other interval data, and generate the offset interval proportion value; S3: Based on the offset interval ratio, call the fluctuation range of the time period purchase data, analyze the frequency change trend of the offset value in multiple time periods, calculate the cumulative difference between the offset frequency and the overall trend, extract the time period characteristics according to the frequency change amount and integrate the frequency weight results to generate the time period deviation frequency value; S4: extracting frequency distribution data in the commodity combination purchase behavior based on the time period deviation frequency value, calculating the intersection feature of the frequency change value and the deviation data, classifying and analyzing the commodity combination according to the frequency change range, and reconstructing and calculating the feature distribution to generate a reconstructed offset feature value; S5: Based on the reconstructed offset feature value, call the regional customer purchase coverage data, extract the feature distribution and offset feature distribution results within the regional coverage area, mark and layer the offset proportion data within the coverage area item by item, and establish the regional offset proportion mark value.

2. The customer behavior analysis method based on retail data according to claim 1, characterized in that: The offset behavior characteristic values ​​include purchase frequency offset values, single consumption amount offset values, and shopping duration offset values; the offset interval proportion values ​​include sales offset proportion values, customer purchase amount offset proportion values, and interval cumulative characteristic value offset proportions; the time period deviation frequency values ​​include frequency change characteristic values, time period frequency weight results, and cumulative difference frequency values; the reconstructed offset characteristic values ​​include commodity combination frequency change values, frequency deviation intersection characteristic values, and reconstructed characteristic distribution results; the regional offset proportion mark values ​​include characteristic distribution values ​​in the coverage area, offset proportion mark values, and layered processing results.

3. The customer behavior analysis method based on retail data according to claim 2, characterized in that: The steps for obtaining the offset behavior characteristic value are specifically as follows: S101: Call the numerical set of the customer's purchase frequency, single consumption amount and shopping time, establish a behavior feature matrix, calculate the mean vector and covariance matrix of multiple feature items through matrix transformation, and generate a feature distribution parameter set; S102: Based on the feature distribution parameter set, a standardized Euclidean distance between each feature point and the mean vector is calculated, a feature point set greater than a set offset threshold is selected, and an offset feature point set is generated; S103: Call the offset feature point set using the formula: ； Calculate the offset behavior strength of all feature points and generate offset behavior feature values; in, Representative The offset behavior strength of feature points, Representative The value of the feature point, represents the mean of the feature items, represents the standard deviation of the feature item, represents the weight of the feature item, Representative The characteristic value of the dimension, Representative The mean of the dimension features, It represents the dimension of the feature matrix, that is, the number of customer behavior features.

4. The method for analyzing customer behavior based on retail data according to claim 3, characterized in that: The specific steps for obtaining the offset interval ratio value are as follows: S201: extracting interval distribution data of sales and customer purchase amounts, calling the offset behavior characteristic value, extracting offset characteristic values ​​corresponding to multiple intervals by matching the interval distribution data with the offset behavior characteristic value, calculating the offset behavior ratio of each interval based on the total interval characteristic value, and generating the interval offset ratio; S202: calling the interval offset ratio, calculating the offset ratios of adjacent intervals one by one, analyzing the offset feature difference values ​​between intervals, taking the average of the offset feature values ​​of all intervals as a benchmark, and using the formula: ； Generate interval offset value ratio data; in, Representative The offset value ratio of the interval, Representative The offset eigenvalue of the interval, Representative The offset eigenvalue of the interval, represents the mean of all interval shift eigenvalues, Indicates the total number of intervals divided in the entire data distribution; S203: calling the interval offset value ratio data, comparing the cumulative characteristic value ratio with the interval data volume, calculating the difference between the cumulative characteristic value and the offset value ratio interval by interval, classifying and identifying the offset interval, and generating the offset interval ratio value.

5. The method for analyzing customer behavior based on retail data according to claim 4, characterized in that: The specific steps for obtaining the time period deviation frequency value are as follows: S301: Based on the offset interval ratio value, call the purchase data fluctuation range of multiple time periods, extract the upper and lower boundary values ​​of the purchase data of multiple time periods, calculate the offset behavior feature values ​​in multiple time periods, filter the feature points that exceed the time period mean deviation threshold, analyze the frequency distribution of the offset behavior in the differentiated time periods, and generate the time period data fluctuation analysis results; S302: Calculate the frequency of the offset value of each period from the period data fluctuation analysis results, using the formula: ； Calculate the cumulative difference value between the offset frequency and the overall trend to generate the offset frequency cumulative difference value data; in, Representative The offset frequency of the time period, Representative The offset value of the time point, Representative The shifted mean of the period, Representative The standard deviation of the shift for the period, represents the total number of time points captured; S303: Call the offset frequency cumulative difference value data, extract the time period with prominent offset frequency changes based on the fluctuation range of multiple time periods, calculate the characteristic frequency change amplitude of the time period, call the frequency weight parameter of the corresponding time period, calculate the weighted ratio of the frequency weight and the characteristic frequency change amplitude, and generate the time period deviation frequency value.

6. The method for analyzing customer behavior based on retail data according to claim 5, characterized in that: The steps for obtaining the reconstructed offset feature value are specifically as follows: S401: extracting frequency distribution data of commodity combination purchase behavior based on the time period deviation frequency value, calling the purchase record of the commodity combination, calculating the number of occurrences of each combination, screening the combination that deviates from the average purchase frequency, classifying according to the purchase trend and combination stability parameter, and generating a frequency distribution data set; S402: calling the frequency distribution data set, analyzing the purchase frequency of each commodity combination, calculating the intersection with the time period deviation frequency value, identifying the overlapping features of the purchase frequency and deviation data, screening commodity combinations with similar purchase patterns, calculating the frequency change rate of the commodity combinations and classifying them, and generating intersection feature analysis results; S403: Call the intersection feature analysis result, classify according to the frequency change range and deviation data, and calculate the feature distribution of the product combination using the formula: ； Reconstruct and calculate the characteristic distribution to generate the reconstructed offset characteristic value; in, represents the reconstructed offset eigenvalue, Representative The frequency change value of a product combination, Representative Deviation data of product combinations, Represents the maximum value of all product combination deviation data. Represents the total number of all product combinations.

7. The method for analyzing customer behavior based on retail data according to claim 6, characterized in that: The specific steps for obtaining the area offset ratio mark value are as follows: S501: Based on the reconstructed offset feature value, call the regional customer purchase coverage data, extract the statistical information of multiple purchase points in the region, analyze the purchase frequency and purchase amount distribution within the coverage area, calculate the standard deviation and mean of the coverage area, evaluate the degree of dispersion in the feature distribution, and generate regional feature distribution data; S502: calling the regional feature distribution data, filtering the offset ratio data in the coverage area, marking the size of the offset value item by item by comparing the offset value of each data point with the statistical benchmark of the area, dividing the data into a high offset group and a low offset group according to the offset amplitude, and generating an offset ratio marking list; S503: Call the offset ratio mark list, using the formula: ； Compare the offset values ​​of the differentiated areas at the same scale to generate a regional offset ratio mark value; in, Representative area The offset ratio mark value, represents the total number of all data points in the differentiation region, Representative data points The weight parameter is used to adjust the contribution of the data point to the overall offset mark value. Representative data points The offset value of Representative area The statistical benchmark value of Representative area The standard deviation of , used to normalize the magnitude of the offset values.

8. A customer behavior analysis system based on retail data, characterized in that: According to any one of claims 1 to 7, the customer behavior analysis method based on retail data comprises: The customer feature extraction module extracts the data set of purchase frequency, single consumption amount and shopping time based on the customer group behavior feature matrix, calculates the value distribution range of multiple features, compares the offset of multiple sets of values ​​with the overall feature change range, filters the value set whose offset exceeds the feature change range, and generates the offset behavior feature value; The sales interval analysis module extracts interval distribution data of sales and customer purchase amount based on the offset behavior characteristic value, compares the offset value with the interval value distribution item by item, calculates the proportion of the offset value to the value in the interval and the cumulative characteristic value, analyzes the difference interval value change ratio and performs classification operation, and generates the offset interval proportion value; The time period frequency analysis module extracts the fluctuation range of purchase data in multiple time periods based on the offset interval ratio, calculates the frequency distribution and trend change of the offset values ​​in multiple time periods, accumulates the difference between the offset frequency and the overall trend in the time period, selects the time period characteristics according to the frequency change amplitude and integrates the frequency weight to generate the time period deviation frequency value; The commodity combination feature reconstruction module extracts the commodity combination purchase frequency distribution data based on the time period deviation frequency value, calculates the intersection feature of the frequency change value and the deviation data, analyzes the intersection feature distribution range and the frequency change amplitude, reconstructs the commodity combination feature distribution, and generates a reconstructed offset feature value; The regional coverage marking module extracts the feature distribution data within the regional customer purchase coverage based on the reconstructed offset feature value, compares the offset features within the region with the overall feature distribution, marks the offset proportion within the region item by item and performs hierarchical classification to establish the regional offset proportion marking value.

Images