An intelligent marketing scene construction method and system based on user behavior trajectory big data

Abstract

The application discloses a kind of based on user behavior trajectory big data's intelligent marketing scene construction method and system, it is related to big data technical field;The method includes: collecting the data of user behavior trajectory data is preprocessed;Data information is identified, and behavior trajectory data information is classified and summarized;According to the data information of data information base, the distribution and law of data are obtained;According to the distribution and law of data, user portrait is constructed;According to user portrait, intelligent marketing scene model is constructed, and prediction marketing scene is recommended to user recommendation prediction marketing scene;According to the real-time data information received, real-time monitoring and analysis are carried out and data is returned to intelligent marketing scene model, and intelligent marketing scene model is optimized and adjusted;The application can obtain the behavior trajectory data information of user by big data, so that there is enough data support before user shopping, and the goods meeting the demand of each user are recommended according to the different demand of each user.

Description

Technical Field

[0001] The present invention relates to the field of big data technology, and in particular to a method and system for constructing intelligent marketing scenarios based on user behavior trajectory big data. Background Art

[0002] In the past marketing environment, on the one hand, production enterprises manufactured and produced, and product inventory piled up and could not be sold, resulting in overcapacity. On the other hand, due to information asymmetry, consumer demand was not fully met, resulting in a situation where sellers could not sell and buyers could not buy. With the rapid development of big data, the use of big data is becoming increasingly indispensable in daily life, so user behavior trajectory data is needed to accurately recommend what users need to meet the needs of different users. Therefore, how to recommend what users need through user data is a problem that urgently needs to be solved in this technical field. Summary of the Invention

[0003] The present invention provides a method for constructing intelligent marketing scenarios based on user behavior trajectory big data, comprising:

[0004] Step S1: The cloud platform collects user behavior trajectory data through web crawlers and pre-processes the data;

[0005] Step S2: The cloud platform identifies the data information, classifies and aggregates different behavioral trajectory data information, and generates a data information database;

[0006] Step S3: The cloud platform analyzes and summarizes the data information in the database to obtain the distribution and patterns of the data;

[0007] Step S4: The cloud platform constructs a user profile based on the distribution and patterns of the data;

[0008] Step S5: The cloud platform constructs an intelligent marketing scenario model based on the user portrait using a machine learning algorithm, predicts marketing scenarios based on the intelligent marketing scenario model, and recommends the predicted marketing scenarios to the user;

[0009] Step S6: The cloud platform performs real-time monitoring and analysis based on the received real-time user behavior trajectory data information and transmits the real-time data back to the intelligent marketing scenario model, continuously optimizing and adjusting the intelligent marketing scenario model.

[0010] In the above-mentioned method for constructing intelligent marketing scenarios based on user behavior trajectory big data, the cloud platform collects user behavior trajectory data through a web crawler and pre-processes the data in the following sub-steps:

[0011] Step S11: The cloud platform sends an HTTP request to the target website through a web crawler to obtain web page content;

[0012] Step S12: The cloud platform extracts the required user behavior trajectory data from the acquired webpage content;

[0013] Step S13: The cloud platform stores the extracted user behavior trajectory data in the MySQL database;

[0014] Step S14: The cloud platform calls the database data, processes the missing values ​​by deleting the missing values, identifies and replaces the outliers through the box plot, deletes the duplicate data in the data by querying the data, and obtains the cleaned data;

[0015] Step S15: The cloud platform converts the cleaned data into a form suitable for model training using normalization, standardization, and logarithmic transformation methods in numerical data conversion to obtain converted data;

[0016] Step S16: The cloud platform selects the main information data of the converted data through principal component analysis (PCA) and linear discriminant analysis (LDA) methods to obtain preprocessed data.

[0017] In the above-mentioned method for constructing intelligent marketing scenarios based on user behavior trajectory big data, the cloud platform identifies data information and classifies and aggregates different behavior trajectory data information to generate a data information database. The sub-steps are as follows:

[0018] Step S21: The cloud platform uses direct recognition to directly determine whether the pre-processed data meets the conditions and uses regular expressions to verify whether it meets the conditions;

[0019] Step S22: The cloud platform uses the keyword extraction method in indirect identification to extract all data that meets the specific conditions from the pre-processed data to obtain identification data;

[0020] Step S23: The cloud platform classifies and aggregates the identified data using a cluster analysis method to obtain classified aggregated data;

[0021] Step S24: The cloud platform transfers the classified and summarized data to the MySQL database to form a data information library.

[0022] In the above-mentioned method for constructing intelligent marketing scenarios based on user behavior trajectory big data, the cloud platform performs statistics and summaries based on the data information in the data information database through data analysis to obtain the distribution and regularity of the data in the following sub-steps:

[0023] Step S31: The cloud platform obtains the distribution of data through a histogram based on the data information in the data information database, obtains the relationship between the data through a scatter plot and a box plot, and obtains the relationship distribution value;

[0024] Step S32: The cloud platform calculates statistics such as mean, median, and standard deviation based on the data in the database to describe the central tendency and dispersion of the obtained data, and obtains the central dispersion value;

[0025] Step S33: The cloud platform groups the data by variables through grouping and aggregation based on the data information in the data information database, and performs aggregation analysis on each group of data to obtain the correlation between the variables and obtain the correlation value;

[0026] Step S34: The cloud platform calculates the probability regularity value based on the relationship distribution value, the concentrated discrete value and the correlation value;

[0027] Step S35: The cloud platform obtains the distribution and regularity of the data through the probability regularity value.

[0028] In the above-mentioned method for constructing intelligent marketing scenarios based on big data of user behavior trajectories, the cloud platform constructs user profiles based on the distribution and patterns of the data in the following sub-steps:

[0029] Step S41: The cloud platform establishes the user's information dimension based on the distribution and regularity of the target user's behavior trajectory data;

[0030] Step S42: The cloud platform extracts the user's purchasing interest and consumption capacity indicators based on the distribution and regularity of the user's behavior trajectory data, and establishes the user's characteristic dimensions;

[0031] Step S43: The cloud platform constructs a user profile based on the user behavior trajectory data information, the user's information dimension and the feature dimension.

[0032] In the above-described method for constructing intelligent marketing scenarios based on user behavior trajectory big data, the cloud platform constructs an intelligent marketing scenario model based on user portraits through a machine learning algorithm, predicts marketing scenario recommendations based on the intelligent marketing scenario model, and recommends the predicted marketing scenario to the user in the following sub-steps:

[0033] Step S51: The cloud platform uses machine learning to train user portraits and build an intelligent marketing scenario model;

[0034] Step S52: The cloud platform tests the performance of the intelligent marketing scenario model through a cross-validation method;

[0035] Step S53: The cloud platform determines whether the marketing scenario recommendation meets the recommendation conditions. If so, the predicted marketing scenario is recommended to the user. Otherwise, step S6 is executed.

[0036] Step S54: The cloud platform uses intelligent delivery technology to accurately push advertisements and promotional information to users based on the intelligent marketing scenario model.

[0037] In the above-mentioned method for constructing intelligent marketing scenarios based on user behavior trajectory big data, the cloud platform performs real-time monitoring and analysis based on the received real-time user behavior trajectory data and transmits the real-time data back to the intelligent marketing scenario model. The sub-steps of continuously optimizing and adjusting the intelligent marketing scenario model are as follows:

[0038] Step S61: The cloud platform obtains real-time user behavior trajectory data information through steps S1, S2, and S3, and uses data stream processing to perform real-time monitoring, analysis, and processing to obtain optimized data;

[0039] Step S62: The cloud platform transmits the optimized data to step S4 to obtain the optimized user portrait, and uses the optimized user portrait to train the intelligent marketing scenario model again to achieve continuous optimization of the model.

[0040] The present invention also provides an intelligent marketing scenario construction system based on user behavior trajectory big data, including:

[0041] The acquisition module is used to collect user behavior trajectory data through web crawlers and pre-process the data;

[0042] The classification and aggregation module identifies data information, classifies and aggregates different behavioral trajectory data information, and generates a data information database;

[0043] The distribution pattern analysis module obtains the distribution and patterns of data through data analysis and summary based on the data information in the data information database.

[0044] The user portrait construction module builds user portraits based on the distribution and patterns of data.

[0045] The intelligent marketing scenario recommendation module builds an intelligent marketing scenario model based on user portraits through machine learning algorithms, predicts marketing scenario recommendations based on the intelligent marketing scenario model, and recommends the predicted marketing scenario to users.

[0046] The optimization module conducts real-time monitoring and analysis based on the received real-time user behavior trajectory data information and transmits real-time data back to the intelligent marketing scenario model, continuously optimizing and adjusting the intelligent marketing scenario model.

[0047] In the above-mentioned intelligent marketing scenario construction system based on user behavior trajectory big data, the collection module specifically includes:

[0048] The web page content acquisition submodule sends HTTP requests to the target website through a web crawler to obtain web page content;

[0049] The user behavior trajectory data extraction submodule extracts the required user behavior trajectory data through the acquired web page content;

[0050] The data storage submodule stores the extracted user behavior trajectory data in the MySQL database;

[0051] The data cleaning submodule calls the database data, processes the missing values ​​by deleting the missing values, identifies and replaces the outliers through the box plot, and removes duplicate data by querying the data to obtain the cleaned data.

[0052] The data conversion submodule converts the cleaned data into a form suitable for model training using the normalization, standardization, and logarithmic transformation methods in numerical data conversion to obtain the converted data;

[0053] The data preprocessing submodule selects the main information data of the converted data through principal component analysis (PCA) and linear discriminant analysis (LDA) methods to obtain preprocessed data.

[0054] In the above-mentioned intelligent marketing scenario construction system based on user behavior trajectory big data, the classification and aggregation module specifically includes:

[0055] Direct recognition submodule uses direct recognition to directly determine whether the preprocessed data meets the conditions and uses regular expressions to verify whether the conditions are met;

[0056] The indirect recognition submodule uses the keyword extraction method in indirect recognition to extract all data that meet specific conditions from the preprocessed data and obtain recognition data;

[0057] The analysis submodule classifies and summarizes the identified data through cluster analysis to obtain classified summary data;

[0058] The data transmission submodule transmits the classified and summarized data to the MySQL database to form a data information database.

[0059] In the above-mentioned intelligent marketing scenario construction system based on user behavior trajectory big data, the distribution law analysis module specifically includes: a relationship distribution analysis submodule, which obtains the distribution of data through a histogram based on the data information in the data information database, obtains the relationship between the data through a scatter plot and a box plot, and obtains the relationship distribution value;

[0060] The centralization and dispersion analysis submodule calculates the mean, median, standard deviation and other statistical quantities based on the data in the database to describe the central tendency and dispersion of the obtained data and obtain the centralization and dispersion values;

[0061] The correlation analysis submodule groups the data by variables through grouping and aggregation based on the data information in the data information database, and performs aggregation analysis on each group of data to obtain the correlation between the variables and obtain the correlation value;

[0062] The probability law calculation submodule calculates the probability law value based on the relationship distribution value, the concentrated discrete value and the correlation value;

[0063] The distribution law acquisition submodule obtains the distribution and law of data through the probability law value.

[0064] In the above-mentioned intelligent marketing scenario construction system based on user behavior trajectory big data, the user portrait construction module specifically includes:

[0065] The information dimension establishment submodule establishes the user's information dimension based on the distribution and regularity of the target user's behavior trajectory data;

[0066] The feature dimension establishment submodule extracts the user's purchasing interest and consumption capacity indicators based on the distribution and regularity of user behavior trajectory data, and establishes the user's feature dimension;

[0067] Build a user portrait sub-module to build a user portrait based on user behavior trajectory data information, user information dimensions and feature dimensions.

[0068] In the above-mentioned intelligent marketing scenario construction system based on user behavior trajectory big data, the intelligent marketing scenario recommendation module specifically includes:

[0069] The training submodule uses machine learning to train user portraits and build intelligent marketing scenario models;

[0070] The testing submodule tests the performance of the intelligent marketing scenario model through cross-validation method;

[0071] The recommendation judgment submodule determines whether the marketing scenario recommendation meets the recommendation conditions. If so, the predicted marketing scenario is recommended to the user. Otherwise, the optimization module is executed.

[0072] The push sub-module uses intelligent delivery technology to accurately push advertising and promotional information to users based on the intelligent marketing scenario model.

[0073] In the above-mentioned intelligent marketing scenario construction system based on user behavior trajectory big data, the optimization module specifically includes:

[0074] Optimize the data transmission submodule, obtain real-time user behavior trajectory data information through the collection module, classification and summary module and distribution pattern analysis module, and use data stream processing to perform real-time monitoring, analysis and processing to obtain optimized data;

[0075] Optimize the user data sub-module, transfer the optimized data to the user portrait construction module, obtain the optimized user portrait, and use the optimized user portrait to train the intelligent marketing scenario model again to achieve continuous optimization of the model.

[0076] The beneficial effects achieved by the present invention are as follows:

[0077] Compared with the above background technology, the present invention can obtain user behavior trajectory data information through big data so that there is sufficient data support before the user shops, and recommend products that meet their needs according to the different needs of each user. The data of products purchased by the user is uploaded to the cloud platform, so that the model is continuously updated and optimized according to real-time data, and can recommend products that better meet the needs of users, thereby realizing customized and accurate recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0079] Figure 1 This is a flow chart of a method for constructing an intelligent marketing scenario based on user behavior trajectory big data provided in Example 1 of the present application;

[0080] Figure 2 This is a schematic diagram of an intelligent marketing scenario construction system based on user behavior trajectory big data provided in Example 2 of this application. DETAILED DESCRIPTION

[0081] The following is a clear and complete description of the technical solutions in the embodiments of the present invention, in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0082] Example 1

[0083] like Figure 1 As shown, the first embodiment of the present application provides a method for constructing an intelligent marketing scenario based on user behavior trajectory big data, which includes the following steps:

[0084] Step S1: The cloud platform collects user behavior trajectory data through web crawlers and preprocesses the data.

[0085] Specifically, the data of user behavior trajectory includes time, location, age, gender, search keywords, browsing history, types of purchased goods and the amount of purchased goods.

[0086] Further, the cloud platform collects user behavior trajectory data through a web crawler, and the data preprocessing sub-steps are as follows:

[0087] Step S11, the cloud platform sends an HTTP request to the target website through the web crawler to obtain the web page content.

[0088] Specifically, use the Requests library of the programming tool Python to send a request and obtain the HTML code of the web page.

[0089] Step S12, the cloud platform extracts the required user behavior trajectory data from the obtained web page content.

[0090] Specifically, use the HTML parsing library BeautifulSoup to complete the parsing of the HTML document and extract the data information.

[0091] Step S13, the cloud platform stores the extracted user behavior trajectory data into the database MySQL.

[0092] Step S14, the cloud platform calls the database data, processes missing values by deleting missing values, identifies and replaces abnormal values by box plot, deletes duplicate data in the data by querying data, and obtains the cleaned data.

[0093] Step S15, the cloud platform converts the cleaned data into a form suitable for model training using normalization, standardization and logarithmic transformation methods in numerical data conversion, and obtains the converted data.

[0094] Step S16, the cloud platform selects the main information data of the data through principal component analysis (PCA) and linear discriminant analysis (LDA) methods, and obtains the preprocessed data.

[0095] Step S2, the cloud platform identifies the data information and classifies and summarizes different behavior trajectory data information to generate a data information library.

[0096] Further, the cloud platform identifies the data information and classifies and summarizes different behavior trajectory data information to generate a data information library, and the sub-steps are as follows:

[0097] Step S21, the cloud platform uses direct recognition to directly determine whether the preprocessed data meets the conditions, and uses regular expression to verify whether the conditions are met.

[0098] Step S22, the cloud platform uses the keyword extraction method in indirect recognition to extract all data that meets the specific conditions from the preprocessed data, and obtains the identified data.

[0099] Specifically, the pre-processed data is identified through direct identification and indirect identification. Direct identification is first used to determine whether the data meets specific conditions. Then, indirect identification is used to extract the data that meets the specific conditions. The data obtained by direct identification and the data obtained by indirect identification are then compared and verified. If the verified data are the same, step S23 is executed. If the verified data are different, the process returns to step S1. For example, first, direct identification is used to extract the clothing product data that the user has purchased from the pre-processed data using direct discriminants such as "coat equals coat" and "pants equals pants", and record it as data 1. Then, indirect identification is used to extract the clothing product data that the user has purchased from the pre-processed data using the overall query "clothing category" equals "coat, pants,..." and record it as data 2. The two sets of data are then compared. If data 1 is equal to data 2, step S23 is executed. If data 1 is not equal to data 2, the process returns to step S1.

[0100] Step S23: The cloud platform classifies and summarizes the identified data through cluster analysis to obtain classified summary data.

[0101] Specifically, the data is classified and summarized according to different types of time, place, person's age, person's gender, search keywords, browsing history, purchased product type, purchased product amount and payment method.

[0102] Step S24: The cloud platform transfers the classified and summarized data to the MySQL database to form a data information library.

[0103] Step S3: The cloud platform analyzes and summarizes the data information in the database to obtain the distribution and patterns of the data.

[0104] Furthermore, the cloud platform uses data analysis to perform statistics and summarize the data distribution and patterns based on the data information in the database. The sub-steps are as follows:

[0105] Step S31: The cloud platform obtains the distribution of data through a histogram based on the data information in the data information database, obtains the relationship between the data through a scatter plot and a box plot, and obtains the relationship distribution value.

[0106] Specifically, the data information in the database is directly queried based on the distribution and relationship of the data and a set is formed to obtain a relationship distribution value. The relationship distribution value is a data value set, which includes a search keyword set, a browsing history set, a purchase product category set, and a purchase product price set.

[0107] Step S32: The cloud platform calculates the mean, median, standard deviation and other statistical quantities based on the data in the database to describe the central tendency and dispersion of the obtained data and obtain the central dispersion value.

[0108] Specifically, the centralized discrete value is a data set, which includes a set of purchase times of the commodities obtained according to the data information base, and a set of statistics of each commodity, wherein the statistics set represents the purchase trend of each commodity by the user.

[0109] Step S33, the cloud platform groups the data according to the variables by grouping and aggregating the data information of the data information base, and performs an aggregated analysis on each group of data to obtain the correlation between the variables, and obtains the correlation value.

[0110] Specifically, the correlation is whether the commodities purchased by the user are related to the income demand and hobbies of the user and other factors, and the correlation value is calculated by using the Pearson correlation coefficient method of the data obtained by the aggregated analysis. The correlation value is used to describe the linear relationship between the variables, and the value range is from -1 to 1. Close to 1 or -1 indicates strong correlation, and close to 0 indicates no correlation.

[0111] Step S34, the cloud platform calculates the probability law value according to the relationship distribution value, the centralized discrete value and the correlation value.

[0112] Specifically, the probability law values of the commodities are integrated into a data set.

[0113] Further, the expression of the probability law value is as follows:

[0114]

[0115] Wherein, MTZ is the probability law value; FB is the set of relationship distribution values, y represents the entire set of prices of the commodities purchased by the user, and x represents the single price of a certain commodity purchased by the user, represents that the price x of a certain commodity is traversed in the set of prices y in the set of relationship distribution values, represents that all the prices in the set of prices y in the set of relationship distribution values are summed up; LS is the set of centralized discrete values, b represents the number of times of purchasing all the commodities by the user, and a represents the number of times of purchasing a certain commodity, represents that the number of times a of purchasing a certain commodity is traversed in the set of commodity purchase times b in the set of centralized discrete values; XG is the set of correlation values, c represents the correlation set purchased by the user, and d represents the correlation of a certain commodity purchased by the user, represents that the correlation value d of a certain commodity is traversed in the set of correlation values c.

[0116] Specifically, the probability law value is the probability law value of a certain commodity, and then the probability law values of all the commodities obtained are combined into the set of probability law values.

[0117] Step S35, the cloud platform obtains the distribution and law of the data through the probability law value.

[0118] Specifically, the cloud platform calculates the probability law value of all the products purchased by the user based on the expression of the probability law value, and associates the user's basic information, purchase information and the probability law value to obtain the distribution and law of the user behavior trajectory data information.

[0119] Step S4: The cloud platform constructs a user profile based on the distribution and patterns of the data.

[0120] Specifically, a user profile is a labeled user model abstracted from user behavior trajectory data, including its distribution and patterns. By mining massive amounts of data, it assigns highly refined user characteristics to form a comprehensive picture of the user. The core function of user profiles is to assign "labels" to users. These labels are refined descriptions of user characteristics, which together form a holistic understanding of the user. User profiles include a variety of data, including basic customer information, interests, purchase history, and behavioral data.

[0121] Furthermore, the cloud platform constructs user profiles based on data distribution and patterns in the following sub-steps:

[0122] Step S41: The cloud platform establishes the user's information dimension based on the distribution and regularity of the target user's behavior trajectory data.

[0123] Specifically, the information dimensions include demographic attributes (age, gender, location, occupation), behavioral attributes (search keywords, browsing history, purchase records, purchase frequency), and social attributes (consumption proportion, income); among them, demographic attribute data is obtained through user behavior trajectory information data; the search keyword browsing history and purchase records in the behavioral attributes are obtained through the relationship distribution value data set, and the purchase frequency in the behavioral attributes is obtained through the centralized discrete value; the total purchase amount of the user is obtained by summing the price set in the relationship distribution value and dividing it by the income to obtain the consumption proportion in the social attributes, among which the income is obtained from the user behavior trajectory information data.

[0124] Step S42: The cloud platform extracts the user's purchasing interest and consumption capacity indicators based on the distribution and regularity of the user's behavior trajectory data, and establishes the user's characteristic dimensions.

[0125] Specifically, purchasing interest refers to a user's shopping preferences in categories such as apparel, luggage, home furnishings, maternity and baby products, toiletries, and food. Spending capacity refers to a user's purchasing power. This distinguishes between a user's actual spending level and their perceived spending level in each category, establishing characteristic dimensions for each.

[0126] Step S43: The cloud platform constructs a user profile based on the user behavior trajectory data information, the user's information dimension and the feature dimension.

[0127] Further, the user portrait expression is as follows:

[0128]

[0129] wherein XHZ is the user portrait; RK is the population attribute, n represents age, x represents gender, d represents location, and z represents occupation, RK (n∪x∪d∪z) represents that the age n, gender x, location d, and occupation z are queried and collected into the population attribute RK from the user behavior trajectory information data; XW is the behavior attribute, s represents search keywords, l represents browsing history, m represents purchase records, and p represents purchase frequency, XW (s∪l∪m∪p) represents that the search keywords s, browsing history l, purchase records m, and purchase frequency p are taken out of the relationship distribution value set and collected into the behavior attribute XW; SH is the social attribute, f represents consumption proportion, and sr represents income, SH (f∪sr) represents that the calculated consumption proportion f and income sr are collected into the social attribute SH; [RK (n∪x∪d∪z) ∪XW (s∪l∪m∪p) ∪SH (f∪sr) ] represents that the population attribute RK, behavior attribute XW, and social attribute SH are collected into the user information dimension; XG is the relevance value set, c represents the user purchase overall relevance set, and q represents the user purchase relevance of a certain type of goods, represents that the user purchase relevance of a certain type of goods q is traversed out of the user purchase overall relevance set c; LS is the centralized discrete value set, b represents the number of times of purchase of all goods by the user, and w represents the number of times of purchase of a certain type of goods, represents that the number of times of purchase of a certain type of goods w is traversed out of the number of times of purchase of all goods b, represents that the user purchase relevance of a certain type of goods and the number of times of purchase of a certain type of goods are multiplied to obtain the purchase interest; FB is the relationship distribution value set, e represents the actual purchase price of a certain type of goods, and y represents the overall price set of goods purchased by the user, represents that the actual purchase price e of a certain type of goods is traversed out of the overall price set of goods purchased by the user y, ll represents the price of all goods browsed, and ml represents the price of a certain type of goods browsed, represents that the price ml of a certain type of goods is traversed out of the price of all goods browsed ll, mm represents the number of times of browsing a certain type of goods, and as represents the number of times of browsing all goods, represents that the number of times of browsing a certain type of goods mm is traversed out of the number of times of browsing all goods as, represents the user consumption capacity, using the actual purchase price of a certain type of goods minus the average value of the price of a certain type of goods browsed If the result is a negative number, it means that the user's consumption ability is weak. If the result is a positive number, it means that the user's consumption ability is strong. The smaller the result, the weaker the user's consumption ability. The larger the result, the stronger the user's consumption ability. It means that purchasing interest and consumption ability are combined into user characteristic dimensions.

[0130] Specifically, a user portrait is a collection of user information dimensions and feature dimensions.

[0131] Step S5: The cloud platform constructs an intelligent marketing scenario model based on the user portrait through a machine learning algorithm, predicts marketing scenario recommendations based on the intelligent marketing scenario model, and recommends the predicted marketing scenario to the user.

[0132] Furthermore, the cloud platform builds an intelligent marketing scenario model based on user portraits through machine learning algorithms. Based on the intelligent marketing scenario model, it predicts marketing scenario recommendations. The sub-steps of recommending the predicted marketing scenario to users are as follows:

[0133] Step S51: The cloud platform uses machine learning to train user portraits and build an intelligent marketing scenario model.

[0134] Specifically, the intelligent marketing scenario model includes an existing data marketing scenario model and a predictive marketing scenario model. The existing data marketing scenario model is trained based on user profile data, while the predictive marketing scenario model is based on the results of the existing data marketing scenario model training. The intelligent marketing scenario model training process is the process of training the predictive marketing scenario model based on the existing data marketing scenario model. The cloud platform inputs the user's known data into the predictive marketing scenario model to obtain the model output. The loss function is then derived based on the model output and the results of manual annotation. Finally, the cloud platform corrects the data in the predictive model by backpropagating the loss function, thus completing the model training.

[0135] Among them, the existing data marketing scenario model is obtained by training the existing data marketing scenario model using a machine learning algorithm based on the user portrait, wherein age, gender, location, occupation, search keywords, browsing history, purchase records, purchase frequency, consumption proportion, income, purchase interest, and consumption capacity are used as known data for classification training, so that the training result is to derive other marketing scenario data from a single known data. For example, other marketing scenario data is obtained through age, {"age"→"gender", "location", "occupation", "search keywords", "browsing history", "purchase records", "purchase frequency", "consumption proportion", "income", "purchase interest", "consumption capacity"}, and then other marketing scenario data is obtained by training based on different marketing scenario keywords. For example, search keywords, browsing history, purchase records, purchase frequency, purchase interest, and consumption capacity are obtained based on age, gender, location, occupation, consumption proportion, and income, {"age", "gender", "location", "occupation", "consumption proportion", "income"→"search keywords", "browsing history", "purchase records", "purchase frequency", "purchase interest", "consumption capacity"}. The more marketing scenario keywords are combined, the fewer the required marketing scenario keywords are obtained, and the more refined the classification is.

[0136] In this case, predictions are made using the data from the existing data model through forward propagation based on the training results of the existing data model to train a predictive marketing scenario model. A predictive marketing scenario model is trained based on known user parameters using the parameters obtained by training the existing data marketing scenario model to generate predictive data. For example, given a user's age, gender, location, occupation, consumption percentage, and income, these parameters are input into the predictive marketing scenario model to obtain data on search keywords, browsing history, purchase history, purchase frequency, purchase interest, and spending power. Predictive training is then performed based on this known data to obtain predicted data for the user. For example, based on age, gender, location, occupation, consumption percentage, and income, predicted search keywords, predicted browsing, predicted purchases, predicted purchase frequency, predicted purchase interest, and predicted spending power are obtained: {"age","gender","location","occupation","consumption percentage","income"→"predicted search keywords","predicted browsing","predicted purchases","predicted purchase frequency","predicted purchase interest","predicted spending power"}.

[0137] Step S52: The cloud platform tests the performance of the intelligent marketing scenario model through a cross-validation method.

[0138] Specifically, we used the holdout method in cross-validation to divide the existing data marketing scenario model into two non-overlapping parts, one as the training dataset and the other as the test dataset. Through testing, we further optimized the model parameters to ensure the accuracy and effectiveness of the model.

[0139] Step S53: The cloud platform determines whether the marketing scenario recommendation meets the recommendation conditions. If so, the predicted marketing scenario is recommended to the user. Otherwise, step S6 is executed.

[0140] Furthermore, the cloud platform determines whether the marketing scenario recommendation meets the recommendation conditions. If so, the predicted marketing scenario is recommended to the user. Otherwise, the sub-steps of step S6 are executed as follows:

[0141] Step S531: The cloud platform uses a machine learning algorithm to obtain user purchase habit trajectory data based on the user portrait, and calculates and predicts the predicted purchase value of the products recommended to the user.

[0142] Specifically, the predicted purchase value is to predict the products recommended to the user based on the user's browsing history, search keywords and other user purchasing habit trajectory data, and then calculate a numerical value based on the expression of the predicted purchase value to determine whether the predicted purchase value of the recommended product is consistent with the user's purchasing habits. If it is consistent, the product predicted to be recommended to the user is transmitted to the intelligent marketing scenario model. If it is not consistent, the predicted purchase value of the next product recommended to the user is calculated.

[0143] Furthermore, the expression for predicting purchase value is as follows:

[0144]

[0145] Among them, GMT is the predicted purchase value; YC is the predicted recommended product price, mc represents the set of all recommended products, and ym represents a selected predicted recommended product. It means selecting a predicted recommended product price ym from the total recommended product price set mc; FB is the relationship distribution value set, y represents the total price set of the user's purchased products, and e represents the actual purchase price of a certain type of product. It means traversing and predicting the actual purchase price e of the recommended product in the price set y of all products purchased by the user; LS is a centralized discrete value set, b represents the number of times the user purchases all products, and w represents the number of times a certain type of product is purchased. It means that the number of times a user purchases all products b is used to traverse and predict the number of times w of products in the same category as the recommended product. It represents the comparison between the predicted recommended product price and the average price of similar products purchased by the user. The closer the value is to 0, the more it matches the user's consumption level. XG is the correlation set, yc represents the correlation set of the predicted recommended products, and yy represents the correlation of a selected predicted recommended product. Indicates the correlation value yy of the selected predicted recommended product from the correlation value set yc of the predicted recommended products. The closer the value is to -1 or 1, the stronger the correlation is. as represents the number of browsed products, mm represents the number of browsed products of a certain category. Indicates the number of products in the same category of the predicted recommended products that are browsed in the number of all products browsed as, The following means getting the number of all browsed products as, Indicates the number of products in the same category that are browsed for predicted recommendations Than the number of all products viewed The closer the value is to 1, the greater the user's interest in the product; the predicted purchase value GMT is calculated based on the expression. The closer its value is to 0, the greater the possibility that the user will purchase the predicted recommended product. Conversely, the closer the value is to 0, the less likely the user will purchase the predicted recommended product.

[0146] Step S532: The cloud platform determines whether the predicted recommended product meets the recommendation conditions based on the predicted purchase value. If so, the predicted purchase product is transmitted to the intelligent marketing scenario model. If not, step S6 is executed.

[0147] Step S54: The cloud platform uses intelligent delivery technology to accurately push advertisements and promotional information to users based on the intelligent marketing scenario model.

[0148] Specifically, intelligent delivery technology monitors the effectiveness of advertising in real time and analyzes users' behavioral data such as clicks, browsing, and conversions. Intelligent delivery technology can automatically adjust the time, frequency, channel and other strategies of advertising delivery based on the intelligent marketing scenario model to ensure that advertising reaches target users in the best way.

[0149] Step S6: The cloud platform performs real-time monitoring and analysis based on the received real-time user behavior trajectory data information and transmits the real-time data back to the intelligent marketing scenario model, continuously optimizing and adjusting the intelligent marketing scenario model.

[0150] Furthermore, the cloud platform monitors and analyzes the received real-time user behavior trajectory data in real time and transmits the real-time data back to the intelligent marketing scenario model. The sub-steps of continuously optimizing and adjusting the intelligent marketing scenario model are as follows:

[0151] Step S61: The cloud platform obtains real-time user behavior trajectory data information through steps S1, S2 and S3, and uses data stream processing to perform real-time monitoring, analysis and processing to obtain optimized data.

[0152] Specifically, the purpose of real-time monitoring and analysis is to monitor abnormal data, issue early warnings for abnormal data, and eliminate abnormal data.

[0153] Step S62, the cloud platform transmits the optimization data to step S4, obtains the optimized user portrait, and uses the optimized user portrait to train the intelligent marketing scenario model again, so as to realize continuous optimization of the model.

[0154] Specifically, the optimized user portrait is continuously filled with data according to real-time data information, so that the user portrait is always practical, thereby continuously training the intelligent marketing scenario model, thereby continuously optimizing the intelligent marketing scenario model, thereby realizing precision marketing and personalized service.

[0155] Embodiment two

[0156] As shown in Figure 2 Embodiment one of the present application provides an intelligent marketing scenario construction system based on user behavior trajectory big data, comprising:

[0157] The acquisition module 21 is configured to collect data of user behavior trajectory through a web crawler, and pre-process the data.

[0158] Specifically, the data of user behavior trajectory includes time, place, age of the person, gender of the person, search keywords, browsing history, types of purchased goods, and amount of purchased goods.

[0159] Further, the acquisition module 21 comprises the following sub-modules:

[0160] The web page content acquisition sub-module sends an HTTP request to a target website through a web crawler to acquire web page content.

[0161] Specifically, the Requests library of the programming tool Python is used to send a request and acquire HTML code of the web page.

[0162] The user behavior trajectory data extraction sub-module extracts the required user behavior trajectory data from the acquired web page content.

[0163] Specifically, the HTML parsing library BeautifulSoup is used to parse the HTML document and extract data information.

[0164] The data storage sub-module stores the extracted user behavior trajectory data in the database MySQL.

[0165] The data cleaning sub-module calls database data, processes missing values by deleting missing values, identifies and replaces abnormal values by box plot, deletes duplicate data in the data by querying and de-duplicating the data, and obtains cleaned data.

[0166] The data conversion submodule converts the cleaning data into a form suitable for model training using normalization, standardization, and logarithmic transformation methods in numerical data conversion, and obtains converted data.

[0167] The data preprocessing submodule selects the main information data of the data through principal component analysis (PCA) and linear discriminant analysis (LDA) methods, and obtains preprocessed data.

[0168] The classification and summary module 22 identifies data information and classifies and summarizes different behavior trajectory data information to generate a data information library.

[0169] Further, the classification and summary module includes the following submodules:

[0170] The direct recognition submodule directly determines whether the preprocessed data meets the conditions using regular expression verification.

[0171] The indirect recognition submodule extracts all data that meets specific conditions from the preprocessed data using keyword extraction in indirect recognition, and obtains recognized data.

[0172] Specifically, the preprocessed data is identified through direct recognition and indirect recognition, wherein the data is first determined whether it meets specific conditions through direct recognition, and then the data that meets specific conditions is extracted through indirect recognition. Then, the data obtained through direct recognition and the data obtained through indirect recognition are compared and verified. If the verification data is the same, the analysis submodule is executed, and if the verification data is different, the collection module 21 is returned. For example, first, the direct recognition method is used to take out the user's already purchased clothing goods data from the preprocessed data through "coat equals coat" and "trousers equals trousers" and the like, which is recorded as data 1. Then, the indirect recognition method is used to take out the user's already purchased clothing goods data from the preprocessed data through "clothing" equals "coat, trousers, …" and the like, which is recorded as data 2. Then, the two sets of data are compared. If data 1 is equal to data 2, the analysis submodule is executed, and if data 1 is not equal to data 2, the collection module 21 is returned.

[0173] The analysis submodule classifies and summarizes the recognized data through cluster analysis, and obtains classified and summarized data.

[0174] Specifically, the classified and summarized data is classified and summarized according to different types of time, location, age of the person, gender of the person, search keywords, browsing history, types of purchased goods, amount of purchased goods, and payment methods.

[0175] The data transmission submodule transmits the classified and summarized data to the MySQL database to form a data information library.

[0176] The distribution law analysis module 23 obtains the distribution and law of the data through data analysis and summary based on the data information in the data information database.

[0177] Furthermore, the distribution law analysis module 23 includes the following submodules:

[0178] The relationship distribution analysis submodule obtains the distribution of data through a histogram based on the data information in the data information database, obtains the relationship between the data through a scatter plot and a box plot, and obtains the relationship distribution value.

[0179] Specifically, the data information in the database is directly queried based on the distribution and relationship of the data and a set is formed to obtain a relationship distribution value. The relationship distribution value is a data value set, which includes a search keyword set, a browsing history set, a purchase product category set, and a purchase product price set.

[0180] The centralization and dispersion analysis submodule calculates the mean, median, standard deviation and other statistical quantities based on the data in the database to describe the central tendency and dispersion degree of the obtained data and obtain the centralization and dispersion values.

[0181] Specifically, the concentrated discrete value is a data set, which includes a set of the number of purchases of a product obtained from a data information database and a set of statistics for each product, wherein the statistics set represents a purchase trend of each product by users.

[0182] The correlation analysis submodule groups the data according to variables through grouping and aggregation based on the data information in the data information library, and performs aggregation analysis on each group of data to obtain the correlation between the variables and obtain the correlation value.

[0183] Specifically, the correlation refers to whether the products purchased by users are related to factors such as the users' income, needs, and hobbies. The data obtained through aggregate analysis is used to calculate the correlation value using the Pearson correlation coefficient method. The correlation value is used to describe the linear relationship between variables. The value range is from -1 to 1. Values ​​close to 1 or -1 indicate strong correlation, and values ​​close to 0 indicate no correlation.

[0184] The probability law calculation submodule calculates the probability law value based on the relationship distribution value, the centralized discrete value and the correlation value.

[0185] Specifically, the probability regularity values ​​of the commodities are integrated into a data set.

[0186] Furthermore, the expression of the probability law value is as follows:

[0187]

[0188] Among them, MTZ is the probability law value; FB is the relationship distribution value set, y represents the set of all prices of goods purchased by the user, and x represents the single price of a certain product purchased by the user. Indicates that the price x of a certain product is traversed in the price set y in the relation distribution value set. represents the sum of all prices in the price set y in the relational distribution value set; LS is a centralized discrete value set, b represents the number of times a user purchases all products, and a represents the number of times a user purchases a certain product. Indicates the number of times a a product is purchased in the set b of product purchase times in the centralized discrete value set; XG is the correlation value set, c represents the user's purchase of all correlation sets, and d represents the correlation of a user's purchase of a product. Indicates that the correlation value d of a certain product is traversed in the correlation value set c.

[0189] Specifically, the probability law value is the probability law value of a certain commodity, and then the obtained probability law values ​​of all commodities are combined into a probability law value set.

[0190] The distribution law acquisition submodule obtains the distribution and law of data through the probability law value.

[0191] Specifically, all commodities purchased by the user are calculated according to the expression of the probability law value to obtain the probability law value, and the user's basic information, purchase information and the probability law value are data-correlated to obtain the distribution and law of the user behavior trajectory data information.

[0192] The user portrait construction module 24 constructs a user portrait based on the distribution and regularity of the data.

[0193] Specifically, a user profile is a labeled user model abstracted from user behavior trajectory data, including its distribution and patterns. By mining massive amounts of data, it assigns highly refined user characteristics to form a comprehensive picture of the user. The core function of user profiles is to assign "labels" to users. These labels are refined descriptions of user characteristics, which together form a holistic understanding of the user. User profiles include a variety of data, including basic customer information, interests, purchase history, and behavioral data.

[0194] Furthermore, the user portrait construction module 24 includes the following submodules:

[0195] The information dimension establishment sub-module establishes the user's information dimension based on the distribution and regularity of the target user's behavior trajectory data.

[0196] Specifically, the information dimensions include demographic attributes (age, gender, location, occupation), behavioral attributes (search keywords, browsing history, purchase records, purchase frequency), and social attributes (consumption proportion, income); among them, demographic attribute data is obtained through user behavior trajectory information data; the search keyword browsing history and purchase records in the behavioral attributes are obtained through the relationship distribution value data set, and the purchase frequency in the behavioral attributes is obtained through the centralized discrete value; the total purchase amount of the user is obtained by summing the price set in the relationship distribution value and dividing it by the income to obtain the consumption proportion in the social attributes, among which the income is obtained from the user behavior trajectory information data.

[0197] The feature dimension establishes a sub-module, which extracts the user's purchasing interest and consumption ability indicators based on the distribution and patterns of user behavior trajectory data, and establishes the user's feature dimension.

[0198] Specifically, purchasing interest refers to a user's shopping preferences in categories such as apparel, luggage, home furnishings, maternity and baby products, toiletries, and food. Spending capacity refers to a user's purchasing power. This distinguishes between a user's actual spending level and their perceived spending level in each category, establishing characteristic dimensions for each.

[0199] Build a user portrait sub-module to build a user portrait based on user behavior trajectory data information, user information dimensions and feature dimensions.

[0200] Furthermore, the user portrait expression is as follows:

[0201]

[0202] Among them, XHZ is the user portrait; RK is the demographic attribute, n represents age, x represents gender, d represents location, z represents occupation, RK (n∪x∪d∪z) Indicates that age n, gender x, location d, and occupation z are queried from user behavior trajectory information data and aggregated into demographic attributes RK; XW is the behavior attribute, s represents the search keyword, l represents the browsing history, m represents the purchase record, p represents the purchase frequency, XW (s∪l∪m∪p) It means that the search keyword s, browsing history l, purchase record m and purchase frequency p are taken out from the relationship distribution value set to form the behavior attribute XW; SH is the social attribute, f represents the consumption proportion, sr represents income, SH (f∪sr) Indicates that the calculated consumption proportion f and income sr are combined into the social attribute SH; [RK (n∪x∪d∪z) ∪XW (s∪l∪m∪p) ∪SH (f∪sr) ] represents the combination of demographic attributes RK, behavioral attributes XW and social attributes SH into user information dimensions; XG is the correlation value set, c represents the user's purchase of all correlation sets, q represents the user's purchase of a certain type of goods correlation, Indicates that the correlation q of a user purchasing a certain type of goods is traversed in the user's purchase of all correlation sets c; LS is a concentrated discrete value set, b represents the number of times the user purchases all goods, and w represents the number of times a certain type of goods is purchased. Indicates the number of times a user purchases all products, w, from the number of times a user purchases all products. Indicates the relevance of users purchasing a certain type of goods and the number of purchases of a certain type of product Multiplying them together gives the purchase interest; FB is the relationship distribution value set, e represents the actual purchase price of a certain type of product, and y represents the total price set of the product purchased by the user. It means that the actual purchase price e of a certain type of goods is traversed from the price set y of all goods purchased by the user, ll means the price of all goods browsed, ml means the price of a certain type of goods browsed, Indicates that the price ml of a certain category of goods is traversed from the price ll of all the goods being browsed, mm indicates the number of goods of a certain category being browsed, and as indicates the number of goods being browsed in all categories. Indicates that the number of browsed products of a certain category mm is traversed from the number of browsed products of all categories as. Indicates the user's spending power, using the actual purchase price of a certain type of goods Subtract the average price of browsing a certain category of products If the result is a negative number, it means that the user's consumption ability is weak. If the result is a positive number, it means that the user's consumption ability is strong. The smaller the result, the weaker the user's consumption ability. The larger the result, the stronger the user's consumption ability. It means that purchasing interest and consumption ability are combined into user characteristic dimensions.

[0203] Specifically, a user portrait is a collection of user information dimensions and feature dimensions.

[0204] The intelligent marketing scenario recommendation module 25 constructs an intelligent marketing scenario model based on the user portrait through a machine learning algorithm, predicts marketing scenario recommendations based on the intelligent marketing scenario model, and recommends the predicted marketing scenario to the user.

[0205] Furthermore, the intelligent marketing scenario recommendation module 25 includes the following submodules:

[0206] The training sub-module uses machine learning to train user portraits and build intelligent marketing scenario models.

[0207] Specifically, the intelligent marketing scenario model includes an existing data marketing scenario model and a predicted marketing scenario model. The existing data marketing scenario model is trained according to user portrait data, and the predicted marketing scenario model is based on the training result of the existing data marketing scenario model. The process of training the intelligent marketing scenario model is the process of training the predicted marketing scenario model according to the existing data marketing scenario model. The known data of the user is input into the predicted marketing scenario model to obtain a model output result. Then, a loss function is obtained according to the model output result and the manually labeled result. Finally, the loss function is back propagated to correct the data in the predicted model, thereby completing the training of the model.

[0208] The existing data marketing scenario model is obtained by training a machine learning algorithm according to user portrait data. Age, gender, location, occupation, search keywords, browsing history, purchase records, purchase frequency, consumption proportion, income, purchase interest, and consumption capacity are used as known data for classification training. The training result is other marketing scenario data obtained by a single known data, for example, other marketing scenario data obtained by age, {“age”→“gender”, “location”, “occupation”, “search keywords”, “browsing history”, “purchase records”, “purchase frequency”, “consumption proportion”, “income”, “purchase interest”, “consumption capacity”}. Then, different marketing scenario keywords are combined to train other marketing scenario data, for example, search keywords, browsing history, purchase records, purchase frequency, purchase interest, and consumption capacity are obtained according to age, gender, location, occupation, consumption proportion, and income, {“age”, “gender”, “location”, “occupation”, “consumption proportion”, “income”→“search keywords”, “browsing history”, “purchase records”, “purchase frequency”, “purchase interest”, “consumption capacity”}. The more marketing scenario keywords are combined, the fewer marketing scenario keywords are obtained, and the classification is more detailed.

[0209] In this case, predictions are made using the data from the existing data model through forward propagation based on the training results of the existing data model to train a predictive marketing scenario model. A predictive marketing scenario model is trained based on known user parameters using the parameters obtained by training the existing data marketing scenario model to generate predictive data. For example, given a user's age, gender, location, occupation, consumption percentage, and income, these parameters are input into the predictive marketing scenario model to obtain data on search keywords, browsing history, purchase history, purchase frequency, purchase interest, and spending power. Predictive training is then performed based on this known data to obtain predicted data for the user. For example, based on age, gender, location, occupation, consumption percentage, and income, predicted search keywords, predicted browsing, predicted purchases, predicted purchase frequency, predicted purchase interest, and predicted spending power are obtained: {"age","gender","location","occupation","consumption percentage","income"→"predicted search keywords","predicted browsing","predicted purchases","predicted purchase frequency","predicted purchase interest","predicted spending power"}.

[0210] The testing submodule tests the performance of the intelligent marketing scenario model through the cross-validation method.

[0211] Specifically, we used the holdout method in cross-validation to divide the existing data marketing scenario model into two non-overlapping parts, one as the training dataset and the other as the test dataset. Through testing, we further optimized the model parameters to ensure the accuracy and effectiveness of the model.

[0212] The recommendation judgment submodule judges whether the marketing scenario recommendation meets the recommendation conditions. If so, the predicted marketing scenario is recommended to the user. Otherwise, the optimization module 26 is executed.

[0213] Furthermore, the recommendation judgment submodule includes the following submodules:

[0214] The purchase prediction submodule uses machine learning algorithms to obtain user purchase habit trajectory data based on user portraits, and calculates the predicted purchase value of the products recommended to the user.

[0215] Specifically, the predicted purchase value is to predict the products recommended to the user based on the user's browsing history, search keywords and other user purchasing habit trajectory data, and then calculate a numerical value based on the expression of the predicted purchase value to determine whether the predicted purchase value of the recommended product is consistent with the user's purchasing habits. If it is consistent, the product predicted to be recommended to the user is transmitted to the intelligent marketing scenario model. If it is not consistent, the predicted purchase value of the next product recommended to the user is calculated.

[0216] Furthermore, the expression for predicting purchase value is as follows:

[0217]

[0218] Among them, GMT is the predicted purchase value; YC is the predicted recommended product price, mc represents the set of all recommended products, and ym represents a selected predicted recommended product. It means selecting a predicted recommended product price ym from the total recommended product price set mc; FB is the relationship distribution value set, y represents the total price set of the user's purchased products, and e represents the actual purchase price of a certain type of product. It means traversing and predicting the actual purchase price e of the recommended product in the price set y of all products purchased by the user; LS is a centralized discrete value set, b represents the number of times the user purchases all products, and w represents the number of times a certain type of product is purchased. It means that the number of times a user purchases all products b is used to traverse and predict the number of times w of products in the same category as the recommended product. It represents the comparison between the predicted recommended product price and the average price of similar products purchased by the user. The closer the value is to 0, the more it matches the user's consumption level. XG is the correlation set, yc represents the correlation set of the predicted recommended products, and yy represents the correlation of a selected predicted recommended product. Indicates the correlation value yy of the selected predicted recommended product from the correlation value set yc of the predicted recommended products. The closer the value is to -1 or 1, the stronger the correlation is. as represents the number of browsed products, mm represents the number of browsed products of a certain category. Indicates the number of products in the same category of the predicted recommended products that are browsed in the number of all products browsed as, The following means getting the number of all browsed products as, Indicates the number of products in the same category that are browsed for predicted recommendations Than the number of all products viewed The closer the value is to 1, the greater the user's interest in the product; the predicted purchase value GMT is calculated based on the expression. The closer its value is to 0, the greater the possibility that the user will purchase the predicted recommended product. Conversely, the closer the value is to 0, the less likely the user will purchase the predicted recommended product.

[0219] The judgment submodule judges whether the predicted recommended product meets the recommendation conditions based on the predicted purchase value. If so, the predicted purchase product is transmitted to the intelligent marketing scenario model. If not, the optimization module 26 is executed.

[0220] The push sub-module uses intelligent delivery technology to accurately push advertising and promotional information to users based on the intelligent marketing scenario model.

[0221] Specifically, intelligent delivery technology monitors the effectiveness of advertising in real time and analyzes users' behavioral data such as clicks, browsing, and conversions. Intelligent delivery technology can automatically adjust the time, frequency, channel and other strategies of advertising delivery based on the intelligent marketing scenario model to ensure that advertising reaches target users in the best way.

[0222] The optimization module 26 performs real-time monitoring and analysis based on the received real-time user behavior trajectory data information and returns the real-time data to the intelligent marketing scenario model, continuously optimizing and adjusting the intelligent marketing scenario model.

[0223] Furthermore, the cloud platform monitors and analyzes the received real-time user behavior trajectory data in real time and transmits the real-time data back to the intelligent marketing scenario model. The sub-steps of continuously optimizing and adjusting the intelligent marketing scenario model are as follows:

[0224] The optimization data submodule obtains real-time user behavior trajectory data information through the collection module 21, the classification and summary module 22 and the distribution pattern analysis module 23, and uses data stream processing to perform real-time monitoring, analysis and processing to obtain optimization data.

[0225] Specifically, the purpose of real-time monitoring and analysis is to monitor abnormal data, issue early warnings for abnormal data, and eliminate abnormal data.

[0226] The user data submodule is optimized, and the optimized data is transmitted to the user portrait construction module 24 to obtain the optimized user portrait. The optimized user portrait is used to train the intelligent marketing scenario model again to achieve continuous optimization of the model.

[0227] Specifically, optimizing user images is to continuously fill in user portraits with data based on real-time data information, so that user portraits are always practical, thereby continuing to train the intelligent marketing scenario model, and continuously optimizing the intelligent marketing scenario model, thereby achieving precise marketing and personalized services.

[0228] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent replacements, improvements, etc. made on the basis of the technical solution of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for constructing intelligent marketing scenarios based on user behavior trajectory big data, characterized in that: include: Step S1: The cloud platform collects user behavior trajectory data through web crawlers and pre-processes the data; Step S2: The cloud platform identifies the data information, classifies and aggregates different behavioral trajectory data information, and generates a data information database; Step S3: The cloud platform analyzes and summarizes the data information in the database to obtain the distribution and patterns of the data; Step S4: The cloud platform constructs a user profile based on the distribution and patterns of the data; Step S5: The cloud platform constructs an intelligent marketing scenario model based on the user portrait using a machine learning algorithm, predicts marketing scenario recommendations based on the intelligent marketing scenario model, and recommends the predicted marketing scenario to the user, including the following sub-steps: Step S51: The cloud platform uses machine learning to train user portraits and build an intelligent marketing scenario model; Step S52: The cloud platform tests the performance of the intelligent marketing scenario model through a cross-validation method; Step S53: The cloud platform determines whether the marketing scenario recommendation meets the recommendation conditions. If so, the predicted marketing scenario is recommended to the user. Otherwise, step S6 is executed, which includes the following sub-steps: Step S531: The cloud platform uses a machine learning algorithm to obtain user purchase habit trajectory data based on the user profile and calculates the predicted purchase value of the product recommended to the user; Specifically, the predicted purchase value is to predict the products recommended to the user based on the user's purchase habit trajectory data, and judge whether the predicted purchase value of the recommended product is consistent with the user's purchase habits. If it is consistent, the product predicted to be recommended to the user is transmitted to the intelligent marketing scenario model. If it is not consistent, the predicted purchase value of the next product recommended to the user is calculated. The predicted purchase value expression is used. Calculate the predicted purchase value, where GMT is the predicted purchase value; YC is the predicted recommended product price, mc represents the set of all recommended products, and ym represents a selected predicted recommended product. It means selecting a predicted recommended product price ym from the total recommended product price set mc; FB is the relationship distribution value set, y represents the total price set of the user's purchased products, and e represents the actual purchase price of a certain type of product. It means traversing and predicting the actual purchase price e of the recommended product in the price set y of all products purchased by the user; LS is a centralized discrete value set, b represents the number of times the user purchases all products, and w represents the number of times a certain type of product is purchased. It means that the number of times a user purchases all products b is used to traverse and predict the number of times w of products in the same category as the recommended product. It represents the comparison between the predicted recommended product price and the average price of similar products purchased by the user. The closer the value is to 0, the more it matches the user's consumption level. XG is the correlation set, yc represents the correlation set of the predicted recommended products, and yy represents the correlation of a selected predicted recommended product. Indicates the correlation value yy of the selected predicted recommended product from the correlation value set yc of the predicted recommended products. The closer the value is to -1 or 1, the stronger the correlation is. as represents the number of browsed products, mm represents the number of browsed products of a certain category. Indicates the number of products in the same category of the predicted recommended products that are browsed in the number of all products browsed as, The following means getting the number of all browsed products as, Indicates the number of products in the same category that are browsed for predicted recommendations Than the number of all products viewed The closer the value is to 1, the more interested the user is in the product. The predicted purchase value GMT is calculated based on the expression. The closer the value is to 0, the more likely the user is to purchase the predicted recommended product. Conversely, the less likely the user is to purchase the predicted recommended product. Step S532: The cloud platform determines whether the predicted recommended product meets the recommendation conditions based on the predicted purchase value. If so, the predicted purchase product is transmitted to the intelligent marketing scenario model. If not, step S6 is executed. Step S54: The cloud platform uses intelligent delivery technology to accurately push advertisements and promotional information to users based on the intelligent marketing scenario model; Step S6: The cloud platform performs real-time monitoring and analysis based on the received real-time user behavior trajectory data information and transmits the real-time data back to the intelligent marketing scenario model, continuously optimizing and adjusting the intelligent marketing scenario model.

2. The method for constructing intelligent marketing scenarios based on user behavior trajectory big data according to claim 1, characterized in that: The cloud platform collects user behavior trajectory data through web crawlers and preprocesses the data in the following sub-steps: Step S11: The cloud platform sends an HTTP request to the target website through a web crawler to obtain web page content; Step S12: The cloud platform extracts the required user behavior trajectory data from the acquired web page content; Step S13: The cloud platform stores the extracted user behavior trajectory data in the database; Step S14: The cloud platform calls the database data, processes the missing values ​​by deleting the missing values, identifies and replaces the outliers through the box plot, deletes the duplicate data in the data by querying the data, and obtains the cleaned data; Step S15: The cloud platform converts the cleaned data into a form suitable for model training using normalization, standardization, and logarithmic transformation methods in numerical data conversion to obtain converted data; Step S16: The cloud platform selects the main information data of the converted data through principal component analysis and linear discriminant analysis methods to obtain preprocessed data.

3. The method for constructing intelligent marketing scenarios based on user behavior trajectory big data according to claim 1, characterized in that: The cloud platform uses the data information in the database to perform statistics and summarize the data distribution and patterns through data analysis. The sub-steps are as follows: Step S31: The cloud platform obtains the distribution of data through a histogram based on the data information in the data information database, obtains the relationship between the data through a scatter plot and a box plot, and obtains the relationship distribution value; Step S32: The cloud platform calculates the mean, median, and standard deviation statistics based on the data in the database to describe the central tendency and dispersion of the obtained data, and obtains the central dispersion value; Step S33: The cloud platform groups the data by variables through grouping and aggregation based on the data information in the data information database, and performs aggregation analysis on each group of data to obtain the correlation between the variables and obtain the correlation value; Step S34: The cloud platform calculates the probability regularity value based on the relationship distribution value, the concentrated discrete value and the correlation value; Step S35: The cloud platform obtains the distribution and regularity of the data through the probability regularity value.

4. The method for constructing intelligent marketing scenarios based on user behavior trajectory big data according to claim 1, characterized in that: The cloud platform constructs user profiles based on data distribution and patterns in the following sub-steps: Step S41: The cloud platform establishes the user's information dimension based on the distribution and regularity of the target user's behavior trajectory data; Step S42: The cloud platform extracts the user's purchasing interest and consumption capacity indicators based on the distribution and regularity of the user's behavior trajectory data, and establishes the user's characteristic dimensions; Step S43: The cloud platform constructs a user profile based on the user behavior trajectory data information, the user's information dimension and the feature dimension.

5. An intelligent marketing scenario construction system based on user behavior trajectory big data, characterized by: include: The acquisition module is used to collect user behavior trajectory data through web crawlers and pre-process the data; The classification and aggregation module identifies data information, classifies and aggregates different behavioral trajectory data information, and generates a data information database; The distribution law analysis module obtains the distribution and law of data through data analysis and summary based on the data information in the data information database; User portrait construction module, which builds user portraits based on the distribution and patterns of data; The intelligent marketing scenario recommendation module builds an intelligent marketing scenario model based on user portraits through machine learning algorithms, predicts marketing scenarios based on the intelligent marketing scenario model, and recommends the predicted marketing scenarios to users. It includes the following sub-modules: The training submodule uses machine learning to train user portraits and build intelligent marketing scenario models; The testing submodule tests the performance of the intelligent marketing scenario model through cross-validation method; The recommendation judgment submodule determines whether the marketing scenario recommendation meets the recommendation conditions. If so, the predicted marketing scenario is recommended to the user. Otherwise, the optimization module is executed. It includes the following submodules: The purchase prediction submodule uses machine learning algorithms to obtain user purchase habit trajectory data based on user profiles and calculates the predicted purchase value of products recommended to users; Specifically, the predicted purchase value is to predict the products recommended to the user based on the user's purchase habit trajectory data, and judge whether the predicted purchase value of the recommended product is consistent with the user's purchase habits. If it is consistent, the product predicted to be recommended to the user is transmitted to the intelligent marketing scenario model. If it is not consistent, the predicted purchase value of the next product recommended to the user is calculated. The predicted purchase value expression is used. Calculate the predicted purchase value, where GMT is the predicted purchase value; YC is the predicted recommended product price, mc represents the set of all recommended products, and ym represents a selected predicted recommended product. It means selecting a predicted recommended product price ym from the total recommended product price set mc; FB is the relationship distribution value set, y represents the total price set of the user's purchased products, and e represents the actual purchase price of a certain type of product. It means traversing and predicting the actual purchase price e of the recommended product in the price set y of all products purchased by the user; LS is a centralized discrete value set, b represents the number of times the user purchases all products, and w represents the number of times a certain type of product is purchased. It means that the number of times a user purchases all products b is used to traverse and predict the number of times w of products in the same category as the recommended product. It represents the comparison between the predicted recommended product price and the average price of similar products purchased by the user. The closer the value is to 0, the more it matches the user's consumption level. XG is the correlation set, yc represents the correlation set of the predicted recommended products, and yy represents the correlation of a selected predicted recommended product. Indicates the correlation value yy of the selected predicted recommended product from the correlation value set yc of the predicted recommended products. The closer the value is to -1 or 1, the stronger the correlation is. as represents the number of browsed products, mm represents the number of browsed products of a certain category. Indicates the number of products in the same category of the predicted recommended products that are browsed in the number of all products browsed as, The following means getting the number of all browsed products as, Indicates the number of products in the same category that are browsed for predicted recommendations Than the number of all products viewed The closer the value is to 1, the more interested the user is in the product. The predicted purchase value GMT is calculated based on the expression. The closer the value is to 0, the more likely the user is to purchase the predicted recommended product. Conversely, the less likely the user is to purchase the predicted recommended product. The judgment submodule determines whether the predicted recommended product meets the recommendation conditions based on the predicted purchase value. If so, the predicted purchase product is transmitted to the intelligent marketing scenario model. If not, the optimization module is executed; The push submodule uses intelligent delivery technology to accurately push advertising and promotional information to users based on intelligent marketing scenario models; The optimization module conducts real-time monitoring and analysis based on the received real-time user behavior trajectory data information and transmits real-time data back to the intelligent marketing scenario model, continuously optimizing and adjusting the intelligent marketing scenario model.

6. The intelligent marketing scenario construction system based on user behavior trajectory big data according to claim 5, characterized in that: The acquisition module specifically includes: The web page content acquisition submodule sends HTTP requests to the target website through a web crawler to obtain web page content; The user behavior trajectory data extraction submodule extracts the required user behavior trajectory data through the acquired web page content; The data storage submodule stores the extracted user behavior trajectory data in the database; The data cleaning submodule calls the database data, processes the missing values ​​by deleting the missing values, identifies and replaces the outliers through the box plot, and removes duplicate data by querying the data to obtain the cleaned data. The data conversion submodule converts the cleaned data into a form suitable for model training using the normalization, standardization, and logarithmic transformation methods in numerical data conversion to obtain the converted data; The data preprocessing submodule selects the main information data of the converted data through principal component analysis and linear discriminant analysis methods to obtain preprocessed data.

7. The intelligent marketing scenario construction system based on user behavior trajectory big data according to claim 5, characterized in that: Distribution law analysis module, specifically including: The relationship distribution analysis submodule obtains the data distribution through the histogram according to the data information of the data information database, obtains the relationship between the data through the scatter plot and box plot, and obtains the relationship distribution value; The centralization and dispersion analysis submodule calculates the mean, median, and standard deviation statistics based on the data in the database to describe the central tendency and dispersion of the obtained data and obtain the centralization and dispersion values; The correlation analysis submodule groups the data by variables through grouping and aggregation based on the data information in the data information database, and performs aggregation analysis on each group of data to obtain the correlation between the variables and obtain the correlation value; The probability law calculation submodule calculates the probability law value based on the relationship distribution value, the concentrated discrete value and the correlation value; The distribution law acquisition submodule obtains the distribution and law of data through the probability law value.

8. The intelligent marketing scenario construction system based on user behavior trajectory big data according to claim 5, characterized in that: User portrait construction module, specifically including: The information dimension establishment submodule establishes the user's information dimension based on the distribution and regularity of the target user's behavior trajectory data; The feature dimension establishment submodule extracts the user's purchasing interest and consumption capacity indicators based on the distribution and regularity of user behavior trajectory data, and establishes the user's feature dimension; Build a user portrait sub-module to build a user portrait based on user behavior trajectory data information, user information dimensions and feature dimensions.

Images