Electronic commerce customer behavior analysis system

Abstract

The invention relates to the technical field of data analysis, in particular to an electronic commerce customer behavior analysis system which comprises a user behavior monitoring module, a commodity attribute feature analysis module, a consumption intention change recognition module, an abnormal behavior detection module and a short-time interest peak value capture module. According to the method, through dynamic classification of electronic commerce customer behavior data, fine granularity of data analysis is improved, deep matching of commodity attribute features strengthens association between user behaviors and commodity features, recommendation accuracy is improved, consumption intention path analysis refines a time window of consumption tendencies, prediction is made to be more timely, and user experience is improved. Abnormal behavior detection adopts short-term fluctuation analysis, an abnormal mode is rapidly recognized, abnormal data interference is reduced, short-term interest commodity recognition is based on browsing, clicking and dwell time changes, short-term interests are accurately captured, matched commodities are pushed in time, analysis accuracy and intelligence are improved, and efficient consumption behavior tracking and prediction are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis, and in particular, to an e-commerce customer behavior analysis system. Background Art

[0002] The technical field of data analysis includes multiple links such as data collection, storage, processing, analysis, and visualization, mainly involving multiple disciplinary fields such as computer science, statistics, artificial intelligence, and database management. In the process of data analysis, data collection is completed through sensors, logs, or application programming interfaces, and then the data is stored in relational databases, non-relational databases, or distributed storage. In the data processing stage, methods such as data cleaning, formatting, deduplication, and standardization are used to ensure data quality, and then methods such as mathematical modeling, regression analysis, and clustering analysis are used to extract patterns and features in the data. The data analysis part includes descriptive analysis, diagnostic analysis, predictive analysis, and prescriptive analysis, and various methods rely on technical means such as statistical inference, machine learning, and natural language processing to achieve, and the analysis results are presented through visualization tools or reporting systems for users to understand and make decisions.

[0003] Among them, an e-commerce customer behavior analysis system refers to analyzing user shopping behaviors and preferences through data mining, behavior modeling, and association analysis based on user interaction data on an e-commerce platform, and studying their purchase decision-making process. This system mainly covers links such as data collection, behavior recognition, user portrait construction, behavior pattern analysis, and recommendation strategy formulation. The data collection part extracts information from data sources such as user browsing records, click behaviors, shopping cart operations, order transactions, and evaluation feedback. The behavior recognition link uses classification methods to identify user browsing habits, purchase intentions, and loyalty. User portrait construction combines factors such as demographic characteristics, consumption preferences, and interest tags, and represents features in a vectorized manner. Behavior pattern analysis uses methods such as association rule mining and time series analysis to mine user shopping paths, consumption cycles, and purchase tendencies. Recommendation strategy formulation is based on strategies such as collaborative filtering, content recommendation, or hybrid recommendation.

[0004] The existing technologies have the deficiencies that the data classification dimensions in user behavior analysis are relatively fixed, making it difficult to accurately depict the short-term and long-term behavior changes of users. This leads to a lag in the analysis of consumption tendencies and an inability to promptly reflect the dynamic changes in user intentions. In terms of commodity matching, the existing technologies mostly adopt a fixed recommendation mode based on historical data and lack sensitivity to the real-time behavior preferences of users, resulting in a large deviation between the recommended commodities and the short-term interests of users. In the analysis of consumption intentions, less attention is paid to the time-series changes in user behavior, making the analysis of the consumption decision-making paths within different time windows less accurate and affecting the prediction accuracy. In anomaly behavior detection, the impact of short-term fluctuations is not fully considered, leading to misjudgment or omission of anomaly behaviors and affecting system security and user experience. In the aspect of short-term interest recognition, the existing technologies fail to accurately capture the peak of interest changes of users within a short period, resulting in a decrease in the effectiveness of short-term interest commodity recommendations and affecting the shopping conversion rate of users within a short period. Summary of the Invention

[0005] The purpose of the present invention is to solve the deficiencies existing in the existing technologies and propose an e-commerce customer behavior analysis system.

[0006] To achieve the above purpose, the present invention adopts the following technical solutions: An e-commerce customer behavior analysis system includes:

[0007] The user behavior monitoring module sorts out the commodity interaction records based on the commodity interaction data of e-commerce customers on the platform, analyzes the change trend of behavior categories, judges the change of user behavior in different commodity categories, analyzes the migration rate of behavior categories in the time series, and obtains the dynamic classification result of user behavior;

[0008] The commodity attribute feature analysis module conducts commodity attribute feature analysis based on the dynamic classification result of user behavior, identifies the correlation degree between user behavior categories and commodity features, screens the commodity attribute features preferred by user behavior, and obtains the commodity attribute matching degree distribution data set;

[0009] The consumption intention change recognition module conducts consumption behavior data analysis based on the commodity attribute matching degree distribution data set, analyzes the change of consumption behavior categories within the time window, judges the change trend of consumption intentions in different time periods, identifies the migration path of user consumption intentions, and obtains the change trend of consumption intentions;

[0010] The anomaly behavior detection module conducts short-term user behavior data analysis based on the change trend of consumption intentions, extracts the behavior features of abnormal fluctuations, judges whether the behavior changes within a short period exceed the normal range, classifies and marks the anomaly behaviors, and obtains the anomaly behavior volatility.

[0011] As a further solution of the present invention, the dynamic classification result of user behavior includes the change trend of behavior categories, the change of user behavior, and the migration rate of behavior categories. The dataset of the matching degree distribution of commodity attributes includes the matching relationship of commodity categories, the correlation degree of user behavior categories, and the preference characteristics of user behavior. The change trend of consumption intention includes the change of consumption behavior categories, the change trend of consumption intention, and the migration path of user consumption intention. The volatility of abnormal behavior includes abnormal behavior characteristics, the change range of behavior in a short time, and the classification mark of abnormal behavior.

[0012] As a further solution of the present invention, the acquisition steps of the user behavior monitoring module are specifically as follows:

[0013] The commodity interaction record extraction sub-module classifies the interaction data according to commodity categories based on the commodity interaction data of e-commerce customers on the platform, including click, browse, add to cart, and purchase behavior records, and eliminates abnormal data to form a commodity category behavior dataset.

[0014] The behavior category change analysis sub-module analyzes the change of customer behavior of different category commodities based on the commodity category behavior dataset, using the formula:

[0015]

[0016] Calculate the customer behavior change rate.

[0017] Among them, B Δ represents the customer behavior change rate, C i is the behavior category interaction volume on the i-th day, C i-1 is the behavior category interaction volume of the previous day, n is the number of days within the time window, D is the weighting factor of the behavior change range, E is the time window adjustment parameter, and W is the stability adjustment coefficient of behavior change.

[0018] The time series migration calculation sub-module calls the customer behavior change rate, calculates the migration rate of behavior categories through time series, and compares the user behavior change trajectories of different categories to obtain the dynamic classification result of user behavior.

[0019] As a further solution of the present invention, the acquisition steps of the commodity attribute feature analysis module are specifically as follows:

[0020] The behavior category matching sub-module extracts user behavior categories based on the dynamic classification result of user behavior, analyzes the proportion of user behavior categories in commodity categories, summarizes the bias of user behavior categories, and identifies the preliminary corresponding relationship between user behavior categories and commodity categories to obtain user behavior category matching data.

[0021] The commodity category feature weight analysis sub-module calls the user behavior category matching data, counts the distribution of attributes in different commodity categories, and uses the formula:

[0022]

[0023] Induce the matching relationship of commodity categories to obtain the matching degree value of commodity attribute features;

[0024] Among them, F' represents the matching degree value of commodity attribute features, P m represents the matching degree of the m-th attribute in the commodity category, A m represents the occurrence frequency of the m-th attribute in the commodity category, U b represents the interaction intensity of user behavior category b under the commodity category, G b represents the overall average interaction intensity of user behavior category b, α represents the commodity attribute matching weight coefficient, β represents the denominator adjustment parameter, and γ represents the user behavior deviation adjustment coefficient;

[0025] The commodity attribute preference screening sub-module calls the matching degree value of the commodity attribute features, analyzes the correlation between the commodity attribute features and the user behavior categories, screens the commodity attribute features with high matching degrees, and obtains the commodity attribute matching degree distribution data set.

[0026] As a further solution of the present invention, the acquisition steps of the consumption intention change recognition module are specifically as follows:

[0027] The behavior data analysis sub-module counts the browsing and purchase data of users in each time period based on the commodity attribute matching degree distribution data set, extracts the consumption behavior categories, and obtains the consumption behavior category analysis results;

[0028] The time window comparison sub-module calls the consumption behavior category analysis results, identifies the frequency and expenditure changes of the consumption behavior in the time period by setting a time window, and uses the formula:

[0029]

[0030] Calculate the consumption intention change index, analyze the consumption behavior frequency and expenditure changes, and adjust the overall trend through the consumption amount fluctuation to obtain the consumption intention change trend;

[0031] Among them, ΔC' represents the consumption intention change index, f t represents the consumption behavior frequency of the t-th time window, f t-1 represents the consumption behavior frequency of the t-1-th time window, d t represents the total consumption expenditure of the t-th time window, d t-1 represents the total consumption expenditure of the t-1-th time window, v trepresents the consumption amount fluctuation value of the t-th time window, v t-1 represents the consumption amount fluctuation value of the (t - 1)-th time window, O represents the consumption behavior change weight, U represents the consumption expenditure change adjustment coefficient, and L represents the consumption amount fluctuation balance factor.

[0032] As a further solution of the present invention, the acquisition step of the abnormal behavior detection module is specifically as follows:

[0033] The behavior data fluctuation extraction sub-module identifies the consumption intention change trend based on the consumption intention change trend, calculates the change value of the behavior data in adjacent time periods, screens the data points with fluctuations exceeding the threshold, and obtains the abnormal behavior fluctuation characteristics;

[0034] The short-term behavior change judgment sub-module calls the abnormal behavior fluctuation characteristics, analyzes the cumulative change rate of the fluctuation characteristics within the time window, and compares it with the normal fluctuation range. The formula is:

[0035]

[0036] Calculate the short-term behavior change rate, judge whether it exceeds the normal range, and obtain the short-term behavior abnormal judgment result;

[0037] Among them, B represents the short-term behavior change rate, M z represents the behavior data value at the z-th moment, T represents the time window length, represents the average value of the behavior data within the time window, M z-1 represents the behavior data value at the (z - 1)-th moment;

[0038] The abnormal behavior classification and marking sub-module calls the short-term behavior abnormal judgment result, screens the abnormal categories according to the abnormal change frequency and fluctuation degree, and obtains the abnormal behavior volatility.

[0039] As a further solution of the present invention, the system further includes a short-term interest peak capture module:

[0040] The short-term interest peak capture module analyzes the short-term browsing behavior data based on the abnormal behavior volatility, screens the short-term interest commodities, analyzes the change range of the browsing, clicking, and staying time, judges the commodity category with a sudden increase in interest, and outputs the recommended short-term interest peak commodities for the user;

[0041] The recommended short-term interest peak commodities for the user include short-term interest commodities, the change range of browsing, clicking, and staying time, and the commodity category with a sudden increase in interest.

[0042] As a further solution of the present invention, the acquisition step of the short-term interest peak capture module is specifically as follows:

[0043] Based on the abnormal behavior volatility, the behavior data analysis sub-module collects the short-term browsing, clicking, and staying data of users, and uses time series analysis to analyze the activity intensity of commodity categories to obtain the activity intensity change data;

[0044] The sudden interest increase judgment sub-module calls the activity intensity change data, analyzes the activity intensity fluctuation of commodity categories, identifies the sudden interest increase categories, and uses the formula:

[0045]

[0046] Obtain the sudden interest increase category data;

[0047] where R represents the sudden interest increase category value, A j is the activity intensity of commodity category j, is the average activity intensity, J is the total number of commodity categories, S is the standard deviation of activity intensity, and max(A j ) is the maximum activity intensity;

[0048] Based on the sudden interest increase category data, the commodity recommendation sub-module combines the original purchase data and market trends, analyzes the recent purchase preferences of users and the market popularity of commodities, prioritizes the commodities in the sudden interest increase categories, screens the commodities that meet the short-term interest peak of users, and outputs the recommended commodities for the short-term interest peak of users.

[0049] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0050] In the present invention, through the dynamic classification of e-commerce customer behavior data, the behavior changes of users in different time dimensions are accurately identified, making data analysis more granular. Through the deep matching of commodity attribute features, the association between user behavior and commodity characteristics is strengthened, greatly improving the accuracy of commodity recommendation. By refining the analysis of the change path of consumption intention, the consumption tendency of users in different time windows can be captured more clearly, making consumption behavior prediction more timely. The detection of abnormal behavior uses short-time behavior fluctuation analysis, which can quickly identify abnormal patterns and reduce the interference of abnormal data on user portraits and recommendations. The identification of short-term interest commodities is based on the analysis of changes in browsing, clicking, and staying time, making the capture of short-term interest more accurate, being able to timely push commodities that meet the current interest preferences, enhancing the accuracy and timeliness of data analysis, making commodity recommendation, user portrait construction, and abnormal behavior detection more intelligent, and realizing the efficient tracking and accurate prediction of user consumption behavior. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 is the system flow chart of the present invention;

[0052] Figure 2This is the acquisition flowchart of the user behavior monitoring module in the present invention;

[0053] Figure 3 This is the acquisition flowchart of the commodity attribute feature analysis module in the present invention;

[0054] Figure 4 This is the acquisition flowchart of the consumption intention change recognition module in the present invention;

[0055] Figure 5 This is the acquisition flowchart of the abnormal behavior detection module in the present invention;

[0056] Figure 6 This is the acquisition flowchart of the short-term interest peak capture module in the present invention. Detailed implementation manners

[0057] In order to make the objectives, technical solutions and advantages of the present invention clearer and more 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.

[0058] 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 thus cannot be understood 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.

[0059] Please refer to Figure 1 , an e-commerce customer behavior analysis system includes:

[0060] The user behavior monitoring module sorts out the commodity interaction records based on the commodity interaction data of e-commerce customers on the platform, analyzes the change trend of behavior categories, judges the change of user behavior in different commodity categories, analyzes the migration rate of behavior categories in the time series, and obtains the dynamic classification result of user behavior;

[0061] The commodity attribute feature analysis module performs commodity attribute feature analysis based on the dynamic classification result of user behavior, extracts the matching relationship of commodity categories, identifies the correlation between user behavior categories and commodity features, screens the commodity attribute features preferred by user behavior, and obtains the commodity attribute matching degree distribution data set;

[0062] Based on the dataset of the distribution of the matching degree of commodity attributes, the consumption intention change recognition module conducts data analysis of consumption behaviors, analyzes the changes in the categories of consumption behaviors within the time window, judges the change trend of consumption intention in different time periods, identifies the migration path of users' consumption intention, and obtains the change trend of consumption intention;

[0063] Based on the change trend of consumption intention, the abnormal behavior detection module conducts data analysis of users' short-term behaviors, extracts the behavioral characteristics of abnormal fluctuations, judges whether the behavior changes within a short time exceed the normal range, classifies and marks abnormal behaviors, and obtains the abnormal behavior volatility;

[0064] Based on the abnormal behavior volatility, the short-term interest peak capture module conducts data analysis of short-term browsing behaviors, screens out the commodities of short-term interest, analyzes the change ranges of browsing, clicking, and staying times, judges the categories of commodities with a sudden increase in interest, and outputs the recommendation of commodities with short-term interest peak for users.

[0065] The dynamic classification results of user behaviors include the change trend of behavior categories, the change of user behaviors, and the migration rate of behavior categories. The dataset of the distribution of the matching degree of commodity attributes includes the matching relationship of commodity categories, the correlation degree of user behavior categories, and the user behavior preference characteristics. The change trend of consumption intention includes the change of consumption behavior categories, the change trend of consumption intention, and the migration path of users' consumption intention. The abnormal behavior volatility includes the abnormal behavior characteristics, the change range of behaviors within a short time, and the classification and marking of abnormal behaviors. The recommendation of commodities with short-term interest peak for users includes the commodities of short-term interest, the change ranges of browsing, clicking, and staying times, and the categories of commodities with a sudden increase in interest.

[0066] Please refer to Figure 2 , the acquisition steps of the user behavior monitoring module are specifically as follows:

[0067] The commodity interaction record extraction sub-module, based on the commodity interaction data of e-commerce customers on the platform, including the behavior records of clicking, browsing, adding to cart, and purchasing, classifies the interaction data according to commodity categories, and eliminates abnormal data to form a commodity category behavior dataset;

[0068] The commodity interaction data on the platform not only involves extracting data from the database, but also includes the preliminary processing of data, such as denoising and standardization, to ensure the accuracy of subsequent analysis. For example, an instance of an e-commerce platform involves monitoring the click-through rate and purchase rate of specific commodity categories (such as electronic products) during a specific shopping festival, screening out abnormal data, such as high-frequency clicks by the same user within a short time but no purchase behavior, indicating machine automatic operation or incorrect data, and the data will be eliminated to ensure the quality of the dataset. The remaining data is classified and summarized, sorted into a behavior dataset differentiated by commodity categories, which is achieved through statistical software and database management, but the key lies in the accurate collection and effective classification of data to obtain the commodity category behavior dataset.

[0069] Based on the product category behavior dataset, the behavior category change analysis sub-module analyzes the changes in customer behavior for differentiated category products, using the formula:

[0070]

[0071] Calculate the customer behavior change rate through operations;

[0072] Among them, B Δ represents the customer behavior change rate, C i is the behavior category interaction volume on the i-th day, C i-1 is the behavior category interaction volume of the previous day, n is the number of days within the time window, D is the weighted factor for the behavior change amplitude, which adjusts the impact of behavior fluctuations on the rate, E is the time window adjustment parameter, which controls the impact of the time window size on the calculation result, and W is the stability adjustment coefficient for behavior changes, which avoids excessive deviations in the rate value due to small n values;

[0073] First, calculate the interaction frequency for each type of product, including statistics on behaviors such as clicks, add-to-carts, and purchases. Then, analyze the purchase conversion ratio and add-to-cart rate. The indicators can be optimized and calculated through specific algorithms to reflect the changes in user behavior for different category products. For example, when conducting year-end sales analysis, select home appliance products and calculate the changes in user interaction increments and conversion rates during the promotion period to quantitatively measure the customer behavior change rate;

[0074] Suppose the customer behavior of the smartphone category is being analyzed. Set the time window n = 5 days, that is, calculate the behavior change rate for the past five days. Let C1 = 1200, C2 = 1350, C3 = 1600, C4 = 1450, C5 = 1700, representing the daily interaction volume of users. Set the weight parameter D = 0.8 to amplify the impact of behavior changes, the time adjustment coefficient E = 3 to ensure that short-term fluctuations do not overly affect the calculation, and the stability adjustment coefficient W = 2 to smooth the calculated values. The calculation process is as follows:

[0075] Calculate the daily interaction change volume:

[0076] |C2 - C1| = |1350 - 1200| = 150;

[0077] |C3 - C2| = |1600 - 1350| = 250;

[0078] |C4 - C3| = |1450 - 1600| = 150;

[0079] |C5 - C4| = |1700 - 1450| = 250;

[0080] Calculate the change rate:

[0081]

[0082] This calculation shows that within the set time window, the rate of change of customer behavior for the smartphone category is 137.63. This value represents the fluctuation of customer behavior for this product category in the most recent five days. A higher rate indicates a greater change in user interest in this category in a short period. Subsequently, corresponding marketing strategies can be formulated in combination with trends, such as increasing advertising investment or inventory allocation, and finally, the rate of change of customer behavior is obtained.

[0083] The time series migration calculation sub-module calls the rate of change of customer behavior, calculates the migration rate of behavior categories through time series, compares the user behavior change trajectories of different differentiated categories, and obtains the dynamic classification result of user behavior.

[0084] Time series analysis is performed to calculate the migration rate of behavior categories. This calculation not only focuses on the interaction transfer probability between categories but also considers the dynamic changes of the time series, compares and analyzes the user behavior between different categories. For example, the data of two categories, household appliances and clothing, are compared, and the behavior migration of users from the household appliance category to the clothing category is identified through the dynamic time warping method. The calculation results show the interaction transfer probability between categories, and the user behavior change trajectories at different time points are visually displayed. Through analysis, the migration law of customer behavior between different product categories can be accurately grasped, and finally, the dynamic classification result of user behavior is generated, which can be directly used for the adjustment and optimization of marketing strategies.

[0085] Please refer to Figure 3 , and the specific acquisition steps of the commodity attribute feature analysis module are as follows:

[0086] Based on the dynamic classification result of user behavior, the behavior category matching sub-module extracts the user behavior categories, analyzes the proportion of user behavior categories in the commodity categories, summarizes the bias of user behavior categories, identifies the preliminary corresponding relationship between user behavior categories and commodity categories, and obtains the user behavior category matching data.

[0087] First, it is necessary to analyze the interaction data between users and product categories, which can be achieved by monitoring user behaviors such as click-through rate and purchase rate on e-commerce platforms. For example, on an online retail website, by analyzing the click and purchase records of users for specific products, the preference intensity of users for products in this category can be calculated. The data can be used to generalize the bias of user behavior categories. For example, if a user frequently browses and purchases baby products, this user can be classified as a "family user". By comparing with product categories, it can be identified which product categories have a high degree of match with this user behavior category. For example, family users also show a high preference for household cleaning products. This correspondence between the user behavior category calculated and the product category will ultimately provide a precise basis for product recommendations on the e-commerce platform and obtain user behavior category matching data.

[0088] The product category feature weight analysis sub-module calls the user behavior category matching data, counts the distribution of attributes in different product categories, and uses the formula:

[0089]

[0090] Generalize the matching relationship of product categories to obtain the product attribute feature matching degree value;

[0091] Among them, F' represents the product attribute feature matching degree value, P m represents the matching degree of the m-th attribute in the product category, A m represents the occurrence frequency of the m-th attribute in the product category, U b represents the interaction intensity of user behavior category b under the product category, G b represents the overall average interaction intensity of user behavior category b, α represents the product attribute matching weight coefficient, β represents the denominator adjustment parameter, and γ represents the user behavior deviation adjustment coefficient;

[0092] The cumulative sum part of the "product attribute feature matching degree value". In the calculation formula of this part, m and b are specific variables not shown. According to the formula, the cumulative sum of the "product attribute feature matching degree value" is mainly for the matching degree between each product attribute and the user behavior category. It sums up the matching degrees of each attribute in a weighted sum manner to reflect the matching degree of this product to a specific user behavior category. m and b represent certain specific indices or parameters of product attributes and user behavior categories;

[0093] The "overall average interaction intensity" is an indicator to measure the overall intensity of the interaction between user behavior and product categories. Specifically, it refers to the average interaction intensity between users and products in this category under the product category, reflecting the interaction frequency between a product category and all users. As a comparison reference, it is used to calculate the intensity of the interaction under a specific user behavior category relative to all users;

[0094] "Interaction intensity" is mainly used to measure the frequency of interaction between users and product categories, specifically the occurrence frequencies of behaviors such as clicks, browsing, and purchases, reflecting the degree of attention and preference of users for specific product categories. High interaction intensity means that users have a relatively high interest in that product category;

[0095] "Denominator adjustment parameter" is mainly used to adjust certain factors in the calculation process to ensure that the obtained value is within a suitable range, avoid excessive deviation of the calculation result, and help balance the influence of various factors. Especially when calculating the matching degree or interaction intensity, by adjusting the denominator, the influence of factors such as time window and user behavior deviation can be controlled to ensure the accuracy of the matching degree and recommendation results;

[0096] The sum of the products of the matching degree and the frequency divided by the corrected value of the frequency: This formula indeed involves multiple variables, including the association between product attribute characteristics and user behaviors. The formula aims to measure the matching degree between product attributes and user behaviors. By calculating the interaction intensity of each product attribute characteristic under different user behavior categories and adjusting the intensity value to obtain a comprehensive matching degree, the calculation of the sum divided by the corrected value is not a simple addition, but a correction based on factors such as weighting factors and adjustment coefficients, so as to ensure that the deviation of user behaviors and other influencing factors are taken into account during the analysis. The final "product attribute characteristic matching degree value" is obtained through complex calculations considering multiple factors. Although the formula seems complex, it can accurately reflect the correlation between products and user behaviors and ensure the accuracy of the recommendation system;

[0097] Obtaining the consumption intention change index: In the formula for the consumption intention change index, although the units of consumption behavior frequency and total consumption expenditure are different, each term in the formula has been adjusted by standardization or balance factors. For example, the fluctuation value of the consumption amount is used to adjust the overall influence of the consumption intention change trend. The purpose of this is to perform reasonable weighted processing on data with inconsistent units to obtain an index result with practical significance. This kind of cross-unit data processing is common in data analysis, especially when dealing with different types of quantitative data. Through appropriate balance factors, the final calculation result can have better interpretability and applicability;

[0098] Premise of dimensionless quantification: One of the operation premises of the formula is the quantification and standardization of data, which means that when dealing with data of different units, they must be converted into relatively comparable values through a certain method. Through this dimensionless quantification process, it is ensured that data with different units can be combined for calculation, avoiding calculation errors caused by differences in data units;

[0099] Therefore, the core of the doubt lies in how to reasonably adjust and correct the influence of different data, and the correction factor and weighting mechanism in the formula ensure that effective operations can still be carried out even when the units are different, providing reasonable and accurate analysis results;

[0100] First, count the occurrence frequencies of the attribute characteristics of commodity categories under different user behavior categories. For example, assume that we are analyzing the preferences of household users for cleaning supplies on an e-commerce platform. Select the following data as the analysis basis: The number of interactions of household users in the cleaning supplies category is 300 times, while the number of interactions in the food category is 150 times, and the number of interactions in the electronic products category is 50 times. This indicates that household users have a greater interest in cleaning supplies. Calculate the weights of each commodity category attribute under different user behavior categories. For example, in the cleaning supplies category, the matching degree of the "phosphate-free" attribute is 0.8, the matching degree of the "antibacterial" attribute is 0.6, and the matching degree of the "lemon fragrance" is 0.4;

[0101] Set specific values:

[0102] P1 = 0.8, P2 = 0.6, P3 = 0.4 (representing the matching degrees of the "phosphate-free", "antibacterial", and "lemon fragrance" attributes respectively);

[0103] A1 = 120, A2 = 100, A3 = 80 (representing the occurrence frequencies of these three attributes in the commodity category respectively);

[0104] U b = 300 (the interaction intensity of household users in the cleaning supplies category);

[0105] G b = 200 (the average interaction intensity of all users);

[0106] α = 1.2 (the commodity attribute matching weight coefficient);

[0107] β = 10 (the denominator adjustment parameter);

[0108] γ = 0.5 (the user behavior deviation adjustment coefficient);

[0109] First, calculate the numerator part:

[0110] ∑(αP m ×A m ) = (1.2×0.8×120) + (1.2×0.6×100) + (1.2×0.4×80) = 225.6;

[0111] Then, calculate the denominator part:

[0112]

[0113] Calculating the user behavior deviation part: |∑(U b -G b )| = |300 - 200| = 100;

[0114] Multiplying by the adjustment coefficient: γ×100 = 0.5×100 = 50;

[0115] Finally calculating F':

[0116] This result indicates that under the household user category, the matching degree of the product attribute characteristics of the cleaning supplies category is 62.81. This value can be used in the product recommendation system of the e-commerce platform to preferentially recommend products with a higher matching degree, thereby increasing the user's click-through rate and purchase rate, and obtaining the product attribute characteristic matching degree value.

[0117] The product attribute preference screening sub-module calls the product attribute characteristic matching degree value, analyzes the relevance between the product attribute characteristics and the user behavior categories, screens the product attribute characteristics with a high matching degree, and obtains the product attribute matching degree distribution data set;

[0118] First, it is necessary to call the product attribute characteristic matching degree value calculated previously to analyze which product attributes show a high relevance in different user behavior categories. For example, if household users show a high purchase frequency for products in the cleaning supplies category, then the attribute characteristics should be regarded as attributes with a high matching degree for household users. By setting a certain matching degree threshold, those product attributes whose matching degree exceeds this threshold are screened out. The threshold can be set according to the platform's marketing strategy and the results of user behavior analysis. For example, it can be set to 75% of the average matching degree of all user behavior data. This not only ensures that the screened attributes are indeed preferred by users but also avoids wasting marketing resources due to attributes with weak preferences. The screened attribute characteristics are summarized to form a product attribute matching degree distribution data set for different user behavior categories. The data set can be directly applied to the product recommendation system to improve user satisfaction and purchase rate, and generate the product attribute matching degree distribution data set.

[0119] Please refer to Figure 4 , the acquisition steps of the consumption intention change recognition module are specifically as follows:

[0120] The behavior data analysis sub-module, based on the product attribute matching degree distribution data set, statistically analyzes the browsing and purchase data of users in each time period, extracts the consumption behavior categories, and obtains the consumption behavior category analysis results;

[0121] First, by collecting users' browsing and purchasing data at different time periods, such as recording the number of clicks and purchases of home appliances and clothing products by users within a month. For example, user A clicks on home appliance products 100 times and makes 20 purchases in January, clicks on clothing products 150 times and makes 30 purchases. The data is automatically collected and organized through a data integration tool, and the analysis tool conducts preliminary processing on the collected data, such as cleaning error data and filling in missing values. Then, statistical analysis methods, such as frequency distribution and coefficient of variation analysis, are used to identify the main categories of users' consumption behaviors and their changing trends. The process covers data preprocessing, statistical analysis, and result interpretation to ensure the reliability and accuracy of the analysis results. Based on the analysis results, detailed basic data is provided for the next time window comparison, enabling the platform to design more personalized marketing strategies for different user groups, thereby improving user satisfaction and purchase conversion rate, and obtaining the analysis results of consumption behavior categories.

[0122] The time window comparison sub-module calls the analysis results of consumption behavior categories. By setting time windows, it identifies the frequency and expenditure changes of consumption behaviors in the time period, and uses the formula:

[0123]

[0124] Calculate the consumption intention change index, analyze the frequency and expenditure changes of consumption behaviors, and adjust the overall trend through the consumption amount fluctuation to obtain the changing trend of consumption intention;

[0125] Among them, ΔC' represents the consumption intention change index, f t represents the consumption behavior frequency in the t-th time window, f t-1 represents the consumption behavior frequency in the (t - 1)-th time window, d t represents the total consumption expenditure in the t-th time window, d t-1 represents the total consumption expenditure in the (t - 1)-th time window, v t represents the consumption amount fluctuation value in the t-th time window, v t-1 represents the consumption amount fluctuation value in the (t - 1)-th time window, O represents the consumption behavior change weight, U represents the consumption expenditure change adjustment coefficient, and L represents the consumption amount fluctuation balance factor;

[0126] The "Consumer Intention Change Index" is derived from the changes in user behavior categories (such as purchase frequency, expenditure amount, etc.) over different time windows, while the "Consumer Intention Change Trend" is the trend obtained through time series analysis of these change indices. The relationship between the two is that the consumer intention change trend is obtained through multiple comparisons and calculations of the consumer intention change index. By setting multiple time windows and analyzing data such as consumer behavior frequency and expenditure changes, the change trend of consumer intention is gradually obtained. The change index provides the basic data and calculation basis for the trend, and the two are interrelated and gradually deduced.

[0127] First, call the analysis results of consumer behavior categories. The data includes the click and purchase statistics of the aforementioned User A on household appliances and clothing products in different months. By setting specific time windows, such as comparing the data of two consecutive months, to observe the changes in consumer behavior;

[0128] This formula comprehensively evaluates the change in consumer intention by calculating the squared difference of consumption frequency and the absolute difference of consumption expenditure, combined with the fluctuation of consumption amount. For example, if the number of clicks and purchases of User A on household appliance products in February are 120 times and 25 times respectively, then by inserting into the formula: f t = 120, f t-1 = 100, d t = 25, d t-1 = 20, v t = 5, v t-1 = 4;

[0129] Set the weight parameters O = 1.5, U = 2, L = 1;

[0130] The calculation results are as follows:

[0131]

[0132] If this index is higher than a certain predetermined threshold, such as 200, it indicates that the consumer intention of the user has changed significantly. Further analysis shows that the change of this user may be related to market promotion activities or the improvement of personal consumption ability, thus providing a basis for adjusting marketing strategies and finally obtaining the consumer intention change trend.

[0133] Please refer to Figure 5 , and the specific acquisition steps of the abnormal behavior detection module are as follows:

[0134] Based on the consumer intention change trend, the behavior data fluctuation extraction sub-module identifies the consumer intention change trend, calculates the change value of behavior data in adjacent time periods, filters out the data points with fluctuations exceeding the threshold, and obtains the abnormal behavior fluctuation characteristics;

[0135] First, it is necessary to monitor and record the changing trend of the user's recent consumption intention. For example, on an e-commerce platform, this process can be achieved by analyzing the user's browsing history, purchase history, and search behavior. By collecting data within a specific time window, the change value of the user's behavior can be calculated, which involves extracting the activity records of each user at different time points and conducting a comparative analysis of the data points to identify abnormal fluctuation points. For a specific example, assume that user A suddenly increases from browsing 10 products per day to browsing 100 products per day within a week. This change value of the behavior data will be marked as an abnormal fluctuation. Based on this abnormal fluctuation, the consumption transformation of this user can be further analyzed to obtain the characteristics of abnormal behavior fluctuations.

[0136] The short-term behavior change judgment sub-module calls the characteristics of abnormal behavior fluctuations, analyzes the cumulative change rate of the fluctuation characteristics within the time window, and compares it with the normal range of fluctuations. The formula is used:

[0137]

[0138] Calculate the short-term behavior change rate, determine whether it exceeds the normal range, and obtain the judgment result of short-term behavior abnormality;

[0139] Among them, B represents the short-term behavior change rate, and M z represents the behavior data value at the z-th moment, T represents the length of the time window, represents the average value of the behavior data within the time window, and M z-1 represents the behavior data value at the (z - 1)-th moment;

[0140] First, extract the previously calculated characteristics of abnormal behavior fluctuations and set the time window T = 5 for calculation. Consider the purchase behavior data (unit: number of purchases) of the user within five days: 20 times on the first day, 50 times on the second day, 20 times on the third day, 45 times on the fourth day, and 50 times on the fifth day;

[0141] Among them, M t represents the number of purchases on the t-th day, and T = 5 represents the total number of days under investigation, represents the average number of purchases within five days. Calculate the average number of purchases:

[0142]

[0143] Calculate the change value of the daily data:

[0144] |M2 - M1| = |50 - 20| = 30;

[0145] |M3 - M2| = |20 - 50| = 30;

[0146] |M4 - M3| = |45 - 20| = 25;

[0147] |M5 - M4| = |50 - 45| = 5;

[0148] Calculate the rate of change:

[0149]

[0150] Calculate the standard deviation part:

[0151]

[0152] Finally calculate the short - term behavior change rate: B = 22.5×31.3 = 704.25

[0153] Assume the normal range is set to 500. Since 704.25 has exceeded the normal fluctuation range, it is determined that the user's behavior is abnormal in the short term, and finally the short - term behavior abnormality determination result is obtained.

[0154] The abnormal behavior classification and marking sub - module calls the short - term behavior abnormality determination result, and filters the abnormal categories according to the abnormal change frequency and fluctuation degree to obtain the abnormal behavior volatility;

[0155] Define the types of abnormal behaviors. For example, they can be classified according to the frequency, fluctuation degree and duration of abnormal behaviors. For instance, a short - term high - frequency purchase behavior is classified as a promotional response type, while a long - term high - value purchase indicates consumption upgrading. Each classification needs to be marked according to the specific behavior volatility, and the specific classification process can be executed through threshold settings. For example, if the behavior change rate exceeds a preset high threshold, this behavior will be marked as a high - risk abnormal behavior. In this way, marks for different types of abnormal behaviors are established, and the abnormal behavior volatility is generated.

[0156] Please refer to Figure 6 , and the acquisition steps of the short - term interest peak capture module are specifically as follows:

[0157] Based on the abnormal behavior volatility, the behavior data analysis sub - module collects the user's short - term browsing, clicking and staying data, and uses time - series analysis to analyze the activity intensity of commodity categories to obtain the activity intensity change data;

[0158] Monitor the change in the activity intensity of the user for different commodity categories to judge whether the user's interest has changed. In practical applications, assume that a user browses 5 commodity categories daily, and during a specific promotion period, the browsing times suddenly increase to 20 times. At this time, the behavior data analysis sub - module will mark this abnormal fluctuation, calculate the growth rate, standardize the data, and output the activity intensity index, which provides basic data for further analysis, and finally obtains the activity intensity change data.

[0159] The sudden increase in interest judgment sub-module calls the activity intensity change data, analyzes the activity intensity fluctuations of commodity categories, identifies the categories with sudden increase in interest, and uses the formula:

[0160]

[0161] Obtain the data of the category with sudden increase in interest;

[0162] where R represents the value of the category with sudden increase in interest, and A j is the activity intensity of commodity category j, is the average activity intensity, J is the total number of commodity categories, S is the standard deviation of activity intensity, and max(A j ) is the maximum activity intensity;

[0163] First, calculate the activity intensity A j of each commodity category, and obtain the average activity intensity and the standard deviation S. Assume that the system monitors the activity intensities of five commodity categories as follows:

[0164] A1 = 50, A2 = 65, A3 = 40, A4 = 80, A5 = 90;

[0165] First, calculate the average activity intensity of the commodity category

[0166]

[0167] Then calculate the sum of the absolute values of the deviations of each category from the average value:

[0168]

[0169] Next, calculate the standard deviation S:

[0170]

[0171] Assume that the maximum activity intensity max(A j ) = 90, and substitute it into the improved formula:

[0172]

[0173] The data of the category with sudden increase in interest R = 43.1 is calculated, indicating that among the five commodity categories analyzed currently, there are categories with significant interest growth, which are commodity category 4 and category 5, that is, A4 = 80, A5 = 90, and finally the category with sudden increase in interest is identified.

[0174] Based on the data of the interest surge categories, combined with the original purchase data and market trends, the product recommendation sub-module analyzes the user's recent purchase preferences and the market popularity of products, prioritizes the products in the interest surge categories, filters out the products that meet the user's short-term interest peak, and outputs the recommendation of products with the user's short-term interest peak.

[0175] Through the optimized recommendation algorithm, it accurately recommends products that the user is interested in. In a specific implementation, assuming that the user has browsed a large number of electronic products, especially newly launched smartphones, in the past week, it will recommend special offers and new products of relevant smartphones based on the interest surge data. This not only increases the user's purchase probability but also improves user satisfaction and the platform's revenue. Finally, it outputs the recommendation of products with the user's short-term interest peak.

[0176] The above is only a preferred embodiment of the present invention, and it does not limit the present invention in other forms. Any person skilled in the art may use the technical content disclosed above to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, as long as it does not depart from the technical solution content of the present invention, any simple modification, equivalent change, and modification made to the above embodiments based on the technical essence of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. An e-commerce customer behavior analysis system, characterized in that, The system includes: The user behavior monitoring module sorts out the product interaction records based on the product interaction data of e-commerce customers on the platform, analyzes the changing trend of behavior categories, judges the changes in user behavior of different product categories, analyzes the migration rate of behavior categories in the time series, and obtains the dynamic classification result of user behavior; The product attribute feature analysis module conducts product attribute feature analysis based on the dynamic classification result of user behavior, identifies the correlation between user behavior categories and product features, filters out the product attribute features preferred by user behavior, and obtains the product attribute matching degree distribution data set; The consumption intention change recognition module conducts consumption behavior data analysis based on the product attribute matching degree distribution data set, analyzes the changes in consumption behavior categories within the time window, judges the changing trend of consumption intention in different time periods, identifies the migration path of user consumption intention, and obtains the changing trend of consumption intention; The abnormal behavior detection module conducts short-term user behavior data analysis based on the changing trend of consumption intention, extracts the behavior characteristics of abnormal fluctuations, judges whether the behavior changes within a short period exceed the normal range, classifies and marks abnormal behaviors, and obtains the abnormal behavior volatility; 2. The e-commerce customer behavior analysis system according to claim 1, characterized in that, The dynamic classification result of user behavior includes the changing trend of behavior categories, the changes in user behavior, and the migration rate of behavior categories. The product attribute matching degree distribution data set includes the matching relationship of product categories, the correlation between user behavior categories, and the preferred features of user behavior. The changing trend of consumption intention includes the changes in consumption behavior categories, the changing trend of consumption intention, and the migration path of user consumption intention. The abnormal behavior volatility includes the characteristics of abnormal behavior, the amplitude of behavior changes within a short period, and the classification mark of abnormal behavior; 3. The e-commerce customer behavior analysis system according to claim 1, wherein The specific acquisition steps of the user behavior monitoring module are as follows: The product interaction record extraction sub-module sorts the interaction data according to product categories based on the product interaction data of e-commerce customers on the platform, including click, browse, add to cart, and purchase behavior records, and eliminates abnormal data to form a product category behavior data set; The behavior category change analysis sub-module analyzes the changes in customer behavior of different product categories based on the product category behavior data set, using the formula: Calculate to obtain the customer behavior change rate; Among them, B Δ represents the customer behavior change rate, C i is the behavior category interaction volume on the i-th day, C i-1 is the behavior category interaction volume of the previous day, n is the number of days within the time window, D is the weighted factor of the behavior change amplitude, E is the time window adjustment parameter, and W is the stability adjustment coefficient of the behavior change; The time series migration calculation sub-module calls the customer behavior change rate, calculates the migration rate of behavior categories through time series, compares the changing trajectories of user behavior in different categories, and obtains the dynamic classification result of user behavior; 4. The e-commerce customer behavior analysis system according to claim 3, characterized in that The specific acquisition steps of the product attribute feature analysis module are as follows: The behavior category matching sub-module extracts user behavior categories based on the dynamic classification result of user behavior, analyzes the proportion of user behavior categories in product categories, summarizes the bias of user behavior categories, identifies the preliminary corresponding relationship between user behavior categories and product categories, and obtains the user behavior category matching data; The product category feature weight analysis sub-module calls the user behavior category matching data, counts the distribution of attributes in different product categories, using the formula: Summarize the matching relationship of product categories to obtain the product attribute feature matching degree value; Among them, F' represents the matching degree value of commodity attribute features, and P m represents the matching degree of the m-th attribute in the commodity category, and A m represents the occurrence frequency of the m-th attribute in the commodity category, and U b represents the interaction intensity of user behavior category b under the commodity category, and G b represents the overall average interaction intensity of user behavior category b. α represents the commodity attribute matching weight coefficient, β represents the denominator adjustment parameter, and γ represents the user behavior deviation adjustment coefficient; The product attribute preference screening sub-module calls the product attribute feature matching degree value, analyzes the relevance between the product attribute features and the user behavior categories, screens the product attribute features with high matching degree, and obtains the product attribute matching degree distribution data set.

5. The e-commerce customer behavior analysis system according to claim 4, characterized in that The obtaining steps of the consumption intention change recognition module are specifically as follows: The behavior data analysis sub-module, based on the product attribute matching degree distribution data set, counts the browsing and purchasing data of users in each time period, extracts the consumption behavior categories, and obtains the consumption behavior category analysis result; The time window comparison sub-module calls the consumption behavior category analysis result, and by setting a time window, identifies the frequency and expenditure changes of the consumption behavior in the time period, and uses the formula: Calculates the consumption intention change index, analyzes the consumption behavior frequency and expenditure changes, and adjusts the overall trend through the consumption amount fluctuation to obtain the consumption intention change trend; Among them, ΔC' represents the consumption intention change index, f t represents the consumption behavior frequency in the t-th time window, f t-1 represents the consumption behavior frequency in the (t - 1)-th time window, d t represents the total consumption expenditure in the t-th time window, d t-1 represents the total consumption expenditure in the (t - 1)-th time window, v t represents the consumption amount fluctuation value in the t-th time window, v t-1 represents the consumption amount fluctuation value in the (t - 1)-th time window, O represents the consumption behavior change weight, U represents the consumption expenditure change adjustment coefficient, and L represents the consumption amount fluctuation balance factor.

6. The e-commerce customer behavior analysis system according to claim 5, wherein The obtaining steps of the abnormal behavior detection module are specifically as follows: The behavior data fluctuation extraction sub-module, based on the consumption intention change trend, identifies the consumption intention change trend, calculates the change value of the behavior data in adjacent time periods, screens the data points with fluctuations exceeding the threshold, and obtains the abnormal behavior fluctuation characteristics; The short-term behavior change judgment sub-module calls the abnormal behavior fluctuation characteristics, analyzes the cumulative change rate of the fluctuation characteristics within the time window, and compares it with the normal fluctuation range, and uses the formula: Calculates the short-term behavior change rate, judges whether it exceeds the normal range, and obtains the short-term behavior abnormality determination result; Among them, B represents the short-term behavior change rate, M z represents the behavior data value at the z-th moment, T represents the time window length, represents the mean value of the behavior data within the time window, M z-1 represents the behavior data value at the (z - 1)-th moment; The abnormal behavior classification and marking sub-module calls the short-term behavior abnormality determination result, and according to the abnormal change frequency and the fluctuation degree, screens the abnormal categories to obtain the abnormal behavior volatility.

7. The e-commerce customer behavior analysis system according to claim 1, wherein The system further includes a short-term interest peak capture module: The short-term interest peak capture module, based on the abnormal behavior volatility, conducts short-term browsing behavior data analysis, screens the short-term interest products, analyzes the change ranges of browsing, clicking, and staying times, judges the product categories with sudden interest increase, and outputs the recommended products of the user's short-term interest peak; The recommended products of the user's short-term interest peak include short-term interest products, the change ranges of browsing, clicking, and staying times, and the product categories with sudden interest increase.

8. The e-commerce customer behavior analysis system according to claim 7, wherein The obtaining steps of the short-term interest peak capture module are specifically as follows: The behavior data analysis sub-module, based on the abnormal behavior volatility, collects the user's short-term browsing, clicking, and staying data, and uses time series to analyze the activity intensity of the product categories to obtain the activity intensity change data; The sudden interest increase judgment sub-module calls the activity intensity change data, analyzes the activity intensity fluctuation of the product categories, and identifies the sudden interest increase categories, and uses the formula: Obtains the sudden interest increase category data; Among them, R represents the interest surge category value, and A j is the activity intensity of product category j, is the average activity intensity, J is the total number of product categories, S is the standard deviation of activity intensity, and max(A j ) is the maximum activity intensity; The product recommendation sub-module, based on the sudden interest increase category data, combines the original purchase data and the market trend, analyzes the user's recent purchase preferences and the market popularity of the products, prioritizes the products in the sudden interest increase categories, screens the products that meet the user's short-term interest peak, and outputs the recommended products of the user's short-term interest peak.

Images