Digital management method and system based on multivariate data

Abstract

The invention discloses a digital management method and system based on multivariate data, relates to the technical field of data processing, and solves the problem of lack of corresponding feature extraction, serial number construction and calculation values for user behaviors, portraits, social contact and position information data of classification categories. And personalized recommendation is difficult to generate for the user based on a collaborative filtering model framework. Comprising the following steps: collecting user multivariate data through various data sources; an interest category number is constructed through a clustering analysis method, a consumption preference clustering category is obtained according to a clustering result, a consumption preference number is obtained, the most active time period of the user on the platform is calculated, the corresponding most active time period number is found, and a user commodity purchase optimal value is constructed; using a collaborative filtering model to obtain and update the latest data of the user in real time, and inputting the model to generate a personalized recommendation list.

Description

Technical Field

[0001] The present invention belongs to the technical field of data processing, and in particular relates to a data management method and system based on multivariate data. Background Art

[0002] With the rapid development of information technology, all types of data are exploding. Traditional digital management methods often focus on processing a single type of data, making it difficult to efficiently integrate, analyze, and manage diverse data. Therefore, there is an urgent need for a method and system that can effectively manage diverse data to achieve comprehensive integration and in-depth analysis. This approach extracts multiple features from user behavior data, user profile data, social data, and location information data, generates various numbers or calculated values, and combines these into a comprehensive feature vector. This allows for personalized recommendations to be generated for users based on a collaborative filtering model framework.

[0003] In addition, existing data management methods lack effective coordination mechanisms in data collection, storage, processing and analysis, resulting in serious data silos. The value of data cannot be fully explored, and data from different sources and formats cannot be well linked for analysis, which affects the accuracy and efficiency of decision-making. It is difficult to extract features and analyze user multivariate data through clustering algorithms, TF-IDF algorithms and encoding, and it is difficult to fully integrate four types of user multivariate data. There is a lack of building a model based on the collaborative filtering framework to generate personalized recommendations. Summary of the Invention

[0004] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention provides a data management method and system based on multivariate data, which are used to solve the following technical problems: In addition, existing data management methods lack effective coordination mechanisms in data collection, storage, processing and analysis, resulting in serious data silos. The value of data cannot be fully explored, and data from different sources and formats cannot be well linked for analysis, which affects the accuracy and efficiency of decision-making. It is difficult to extract features and analyze user multivariate data through clustering algorithms, TF-IDF algorithms and encoding, and it is difficult to fully integrate four types of user multivariate data. There is a lack of building a model based on the collaborative filtering framework to generate personalized recommendations.

[0005] To solve the above problems, the first aspect of the present invention provides a digital management method based on multivariate data, comprising the following steps: Step 1: Collect user multivariate data through various data sources; Step 2: Preprocess the collected multivariate user data. Use cluster analysis to cluster the user's high-frequency keywords and interest categories, assigning them an interest category number. Combine the processed age, gender, occupation, and education characteristics into a feature matrix. Based on the clustering results, derive the consumption preference cluster categories and consumption preference numbers. Calculate the user's most active time period on the platform, find the corresponding most active time period number, and construct the user's optimal purchase value. Step 3: Combine the user's interest category ID, consumption preference ID, most active time period ID, and the user's optimal purchase value to obtain a feature vector to build a collaborative filtering model. The latest user data is obtained and updated in real time, and input into the trained collaborative filtering model to generate a personalized recommendation list.

[0006] Furthermore, the step 1 includes the following steps: The server records the user behavior data on websites with search functions and social media platforms. The user behavior data includes the user's search keywords, product browsing, adding to shopping carts and purchases, as well as the corresponding time. Basic information, including name, date of birth, and gender, is obtained through the personal information users enter when registering accounts on websites and applications. A questionnaire is embedded in the user registration process to obtain the user's occupation and education level information. User profile data is then generated based on the user's basic information, occupation, and education level information. Through the authorization mechanism or API interface provided by the social platform, find the application that the user has authorized to access and record the user's social data within the specified permissions. Social data includes the content posted in the social platform's circle of friends or dynamics, as well as likes, comments, and forwarding; The user's location information data is obtained through the GPS hardware of the user's mobile phone. The location information data includes the user's basic location coordinate data, altitude data and timestamp data.

[0007] Furthermore, the pre-processing operation on the collected user multivariate data includes the following steps: After the collected multivariate data is transmitted to the server, the collected user behavior data is cleaned and various user behaviors are classified into four categories, including searching keywords, browsing products, adding to shopping carts, and purchasing. All timestamps are converted into a unified format, and user behaviors are divided according to time series; Perform data integration and missing value processing on the collected user portrait data, and convert the various data contained in the user portrait data using coding; Clean the collected social data; The collected location information data is cleaned, and the geographic coordinates of the location information are converted into a coordinate system suitable for analysis. The address information contained in the location information is standardized and converted into a unified address format.

[0008] Furthermore, clustering the user's high-frequency keywords and interest categories by a cluster analysis method and assigning an interest category number includes the following steps: Collect user search keyword records within 30 days from the platform, including search time and keyword content, obtain user product browsing records, including product browsing time, product category and specific product information, and collect user purchase records, including purchase time and various attributes of purchased products, and add a timestamp to each user behavior data; perform text preprocessing on user search keywords, use word frequency statistics to count the probability of different keywords appearing in user search behavior to obtain the word frequency of each keyword, apply TF-IDF technology, divide the number of times a keyword appears in the user's search record by the total number of words in the search record to obtain the TF value of the keyword in the search record, calculate the corresponding number of each keyword appearing in all search records within 30 days, divide the total number of all search records by the total number of times the keyword appears, and then take the logarithm to obtain the IDF value of the keyword, and multiply the TF value of each keyword by the corresponding IDF value to obtain the TF-IDF value of the keyword in all search records within 30 days; Based on the calculated word frequency and TF-IDF value, high-frequency keywords are determined. Based on the high-frequency keywords and the category labels of the product classification system, the product categories that the user may be interested in are determined. A unique integer code starting from 0 is assigned to each product category. The user's high-frequency keywords and interest categories are converted into feature vectors suitable for clustering algorithm processing. The user's high-frequency keywords and interest categories are clustered through clustering analysis methods. Based on the clustering results, the users are divided into different groups, and each group is assigned an interest category number.

[0009] Furthermore, the processed age, gender, occupation and education level features are combined into a feature matrix, and the consumption preference cluster categories are obtained according to the clustering results to obtain the consumption preference numbers, including the following steps: Save the name as a separate field in the dataset, convert the user's date of birth into age, calculate the difference between the current date and the date of birth to get the user's age, divide the age into different age groups, and assign a value to each age group; convert the gender into a numerical code; determine the user's occupational category based on the collected user's occupational information and refer to the national occupational classification standards, specify the corresponding classification rules for each occupation, use the text similarity algorithm to calculate the similarity between the occupation name and the keywords of each occupational category, classify the occupation into the occupational category with the highest similarity, and create a two-dimensional code The matrix has the number of rows and columns of occupational categories. Initially, all elements are set to 0. For each occupational category, the corresponding row and column positions in the coding matrix are set to 1, so that each occupational category has a unique one-hot coding vector. The education level is converted from low to high into an ordered numerical code. The occupational codes are standardized, and the processed age, gender, occupation and education level features are combined into a feature matrix. The standardized feature matrix is ​​clustered using the selected clustering algorithm. The consumption preference cluster category is obtained based on the clustering results, and a unique consumption preference number is assigned to each label.

[0010] Furthermore, the calculation of the user's most active time period on the platform and finding the corresponding most active time period number includes the following steps: The content, publishing time, and publisher ID information of each Moments or dynamic post are recorded; like data includes the record of users liking or giving likes including the corresponding timestamp; comment data includes the collection of users' comments on Moments or dynamics and the frequency of comments, commenter ID, and comment time; forwarding data includes the record of the number of users forwarding content and the timestamp of forwarding. Each hour is simply numbered, and the number of content posted by users, the number of likes received and given, the number of comments, the number of replies, and the number of forwardings are counted every hour of every day within 30 days to obtain the comprehensive value of users' social interaction activity every hour; The formula for calculating the comprehensive value of user social activity is: in, Indicates the comprehensive value of user social activity. Indicates the number of user-posted content. Indicates the likes given by the user. Indicates the number of likes received by the user. Indicates the number of user comments, Indicates the number of comments on the user. Indicates the number of content forwarded by the user; Add up the user's social activity comprehensive values ​​corresponding to each hour of each day for 30 days to obtain the total comprehensive value for each hour. Find the hour with the highest total comprehensive value as the user's most active time period on the platform, and find the corresponding most active time period number.

[0011] Furthermore, the construction of the optimal value for user product purchases includes the following steps: Determine whether the user has enabled the location permission for the APP, use the positioning API provided by the mobile phone operating system to interact with the communication and data of the mobile phone GPS hardware, obtain the user's basic location coordinate data, altitude data and timestamp data, build a data receiving interface on the back end of the APP that matches the front end location information acquisition function module, use the HTTPS network communication protocol to encrypt and transmit the location information data, and store the user's basic location coordinate data, altitude data, timestamp data and the unique identifier information associated with the user account in the database on the platform server side, associate and integrate the inventory information of all products on the platform with the geographic location information of warehouses and stores in various places, establish a product inventory geographic location database, add the geographic location coordinates, inventory type, inventory quantity and basic product information of the warehouse or store where each product record is located in the database, calculate the straight-line distance between the user's location and the warehouse where the product is located, divide the product type that the user is interested in by the total product type to obtain the category coverage, normalize the quantity and popularity of the products to the maximum and minimum values ​​respectively to obtain the quantity normalization value and the popularity normalization value, and combine them to obtain the optimal value for the user's product purchase; The formula for calculating the optimal value of user product purchases is: in, Indicates the optimal value of the user's product purchase, Indicates the straight-line distance between the user's location and the product's warehouse. Indicates the types of products that the user is interested in. Indicates the total product category. represents the quantity normalized value, represents the normalized value of heat, and Indicates the weight coefficient corresponding to distance and product.

[0012] Furthermore, the step three includes the following steps: The collected user interest category numbers, consumption preference numbers, most active time period numbers and user product purchase optimal values ​​are combined into a feature vector. The feature vector similarity between users is calculated through cosine similarity and Pearson correlation coefficient, and the average of the two is taken as the final similarity. A similarity threshold is set, and the user group exceeding the similarity threshold is determined as the neighbor of the target user. The purchase of products by this user group is counted. The recommended products are screened according to the interest category number and consumption preference number of the target user. Combined with the most active time period number and the user's product purchase optimal value, the time is determined to push the nearest store with sufficient products to the user; the collected user data is used as part of the training set to train the collaborative filtering model. When the user logs in to the system or browses products, the user's latest data is obtained in real time, the information in the feature vector is updated, and the updated feature vector is input into the trained model to generate a personalized recommendation list.

[0013] The present invention also provides a digital management system based on multivariate data, comprising the following modules: Multivariate data collection module: collects user multivariate data through various data sources; User multivariate data preprocessing and feature construction module: This module preprocesses the collected user multivariate data, clusters the user's high-frequency keywords and interest categories using cluster analysis methods, assigns an interest category number, and combines the processed age, gender, occupation, and education level features into a feature matrix. Based on the clustering results, the consumption preference cluster categories are obtained and the consumption preference numbers are obtained. The user's most active time period on the platform is calculated, and the corresponding most active time period number is found to construct the optimal value for the user's product purchases. Collaborative filtering recommendation module: Combines the user's interest category number, consumption preference number, most active time period number, and the user's optimal purchase value to obtain a feature vector to build a collaborative filtering model. It obtains and updates the user's latest data in real time and inputs it into the trained collaborative filtering model to generate a personalized recommendation list.

[0014] Beneficial effects of the present invention: The present invention uses clustering algorithms to analyze user behavior data, such as purchase records, browsing history, and search keywords, to gain an in-depth understanding of the user's product interest categories; user portrait data includes age, gender, occupation, and education information. By analyzing, numbering, and calculating these data, a detailed user portrait can be constructed to understand the differences in needs and consumption preferences of users of different age groups; by analyzing users' behaviors such as likes, comments, and forwarding on social media platforms or in Moments, it is possible to understand the most active time of users using the corresponding application, classify users with the help of numbering and calculation, and push notifications to users with different active times, thereby increasing user stickiness to the platform; using positioning APIs and mobile phone GPS hardware to analyze user location information data, it helps companies provide localized services. Companies can place targeted advertisements based on the characteristics of the user's region, city, or even specific block, while providing users with a time-saving and labor-saving purchasing method; The present invention obtains the user's interest category number, consumption preference number, most active time number and the user's optimal purchase value through early data analysis and collection, and after establishing a unique identifier for each user, combines them into a feature vector, which can accurately characterize the unique preferences of each user, greatly improving the correlation between recommendations and users' actual needs. The final similarity matrix is ​​obtained by taking the average of cosine similarity and Pearson correlation coefficient, and the neighbor group of the target user is determined. A variety of factors are comprehensively considered to determine the recommendation list to be pushed to the target user. The collaborative filtering model is trained based on the user's feature vector and the corresponding user data as a training set. According to the latest acquired data, the updated feature vector is input into the trained collaborative filtering model. The model calculates and analyzes the input feature vector to generate a personalized recommendation list and recommends it to the user. For enterprises, the recommendation system can accurately push suitable products to users in need, thereby improving sales conversion rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 Schematic diagram of the method flow of the present invention; Figure 2 It is a schematic diagram of the module flow of the present invention. DETAILED DESCRIPTION

[0016] The following will clearly and completely describe 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 the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0017] See also Figure 1As shown, the present invention is a digital management method based on multivariate data, comprising the following steps: Step 1: Collect user multivariate data through various data sources; Step 2: Preprocess the collected multivariate user data. Use cluster analysis to cluster the user's high-frequency keywords and interest categories, assigning them an interest category number. Combine the processed age, gender, occupation, and education characteristics into a feature matrix. Based on the clustering results, derive the consumption preference cluster categories and consumption preference numbers. Calculate the user's most active time period on the platform, find the corresponding most active time period number, and construct the user's optimal purchase value. Step 3: Combine the user's interest category ID, consumption preference ID, most active time period ID, and the user's optimal purchase value to obtain a feature vector to build a collaborative filtering model. The latest user data is obtained and updated in real time, and input into the trained collaborative filtering model to generate a personalized recommendation list.

[0018] Specifically, user multivariate data is collected through various data sources, including user behavior data, user portrait data, social data, and location information data; for each keyword, its TF value is multiplied by the corresponding IDF value to obtain the TF-IDF value of the keyword in all search records within 30 days, and the TF-IDF value is selected. The top 100 keywords with the highest IDF value are identified as high-frequency keywords. Based on these high-frequency keywords and the category labels of the product classification system, a clustering algorithm is used to determine the product categories that users may be interested in. The processed features such as age, gender, occupation, and education level are merged into a feature matrix. Based on the clustering results, a unique consumer preference number is assigned to each cluster category, starting from 1. The user's social activity composite values ​​corresponding to each hour of each day for the past 30 days are added together to obtain a total composite value for each hour. The one-dimensional array storing the total composite values ​​is traversed to find the hour with the highest total composite value. The hour number is recorded, which is the number of the time period when the user was most active on the platform. The optimal value for the user's product purchase is obtained by jointly calculating the straight-line distance between the user's location and the product library, the category coverage rate, the quantity normalization value, and the popularity normalization value. The system's tracking and recording functions are used to obtain the user's latest data in real time, and the updated feature vector is input into the trained collaborative filtering model. The model calculates and analyzes the input feature vector to generate a personalized recommendation list for the user.

[0019] In one embodiment of the present invention, the step 1 includes the following steps: The server records the user behavior data on websites with search functions and social media platforms. The user behavior data includes the user's search keywords, product browsing, adding to shopping carts and purchases, as well as the corresponding time. Basic information, including name, date of birth, and gender, is obtained through the personal information users enter when registering accounts on websites and applications. A questionnaire is embedded in the user registration process to obtain the user's occupation and education level information. User profile data is then generated based on the user's basic information, occupation, and education level information. Through the authorization mechanism or API interface provided by the social platform, find the application that the user has authorized to access and record the user's social data within the specified permissions. Social data includes the content posted in the social platform's circle of friends or dynamics, as well as likes, comments, and forwarding; The user's location information data is obtained through the GPS hardware of the user's mobile phone. The location information data includes the user's basic location coordinate data, altitude data and timestamp data.

[0020] Specifically, embed tracking code snippets into the website's front-end code. For example, use JavaScript to write event listeners to monitor user searches, page views, shopping cart interactions, and buy button clicks. When these events occur, trigger the corresponding function and send the relevant data to the server. Use a back-end programming language to build a server-side interface to receive the data sent by the front-end and store it in a database. You can choose a relational or non-relational database to store the data, depending on the data structure and query requirements. If a social media platform provides a developer API, such as Weibo Open Platform or WeChat Open Platform, develop according to the platform's API documentation, register as a developer, create an application, and obtain an API key and access token. Use the functions provided by the API to obtain user behavior data on social media. In the user registration interface design, ensure that there are clear fields for users to enter their name, date of birth, and gender information. At the same time, embed a questionnaire in an appropriate location. The questionnaire content should include questions such as occupation and education level, guiding users to jump to the authorization page provided by the platform. After the user logs in and confirms the authorization, the application will obtain an authorization code and use the obtained access token to access the user's social data through the platform's API interface. For example, the app can obtain dynamic content, likes, comments, and forwarded information posted by users on WeChat Moments. After obtaining location permissions, the app can obtain the user's location information through the phone's GPS hardware. The location service API provided by the phone's operating system can be used to obtain location coordinates, altitude, and timestamp data.

[0021] In one embodiment of the present invention, the pre-processing operation on the collected user multivariate data includes the following steps: After the collected multivariate data is transmitted to the server, the collected user behavior data is cleaned and various user behaviors are classified into four categories, including searching keywords, browsing products, adding to shopping carts, and purchasing. All timestamps are converted into a unified format, and user behaviors are divided according to time series; Perform data integration and missing value processing on the collected user portrait data, and convert the various data contained in the user portrait data using coding; Clean the collected social data; The collected location information data is cleaned, and the geographic coordinates of the location information are converted into a coordinate system suitable for analysis. The address information contained in the location information is standardized and converted into a unified address format.

[0022] Specifically, check the data set and delete duplicate records. For example, if the same user has multiple identical operation records for the same product at the same time, keep one record and delete the others. Convert all search keywords to lowercase letters to facilitate unified matching in subsequent analysis. For product information involved in browsing products, adding to shopping carts, and purchasing, ensure that the format of key identifiers such as product IDs is unified. Identify and process abnormal behavior data. For example, if a user completes the entire process from browsing products to purchasing in a very short time, and the product is of high value, it may be necessary to further verify whether the data is true. If it is judged to be abnormal, consider deleting or marking it. According to the behavior type, divide the data into four categories: search keywords, browsing products, adding to shopping carts, and purchasing; integrate user portrait data, such as In a relational database, use the JOIN operation to connect the basic information table containing name, date of birth, and gender with the questionnaire data table containing occupation and education information into a table containing comprehensive user portrait data. Merge any possible redundant or duplicate information and determine the fields that require unique-hot encoding, such as categorical variables such as occupation and education level. Social data cleaning includes deleting advertisements, promotional content, irrelevant information, and duplicate data. Identify and delete obviously erroneous location information, such as records whose geographic coordinates exceed a reasonable range. Determine a coordinate system suitable for analysis. Use professional geographic information system tools or geographic coordinate conversion libraries to convert the original geographic coordinates to the selected coordinate system. Establish a unified address format standard and use address parsing and standardization tools to parse and format address information.

[0023] In one embodiment of the present invention, clustering the user's high-frequency keywords and interest categories using a cluster analysis method and assigning an interest category number includes the following steps: Collect user search keyword records within 30 days from the platform, including search time and keyword content, obtain user product browsing records, including product browsing time, product category and specific product information, and collect user purchase records, including purchase time and various attributes of purchased products, and add a timestamp to each user behavior data; perform text preprocessing on user search keywords, use word frequency statistics to count the probability of different keywords appearing in user search behavior to obtain the word frequency of each keyword, apply TF-IDF technology, divide the number of times a keyword appears in the user's search record by the total number of words in the search record to obtain the TF value of the keyword in the search record, calculate the corresponding number of each keyword appearing in all search records within 30 days, divide the total number of all search records by the total number of times the keyword appears, and then take the logarithm to obtain the IDF value of the keyword, and multiply the TF value of each keyword by the corresponding IDF value to obtain the TF-IDF value of the keyword in all search records within 30 days; Based on the calculated word frequency and TF-IDF value, high-frequency keywords are determined. Based on the high-frequency keywords and the category labels of the product classification system, the product categories that the user may be interested in are determined. A unique integer code starting from 0 is assigned to each product category. The user's high-frequency keywords and interest categories are converted into feature vectors suitable for clustering algorithm processing. The user's high-frequency keywords and interest categories are clustered through clustering analysis methods. Based on the clustering results, the users are divided into different groups, and each group is assigned an interest category number.

[0024] Specifically, irrelevant characters in the search keywords are removed, all keywords are converted to lowercase letters, and a data processing library in a programming language is used to perform word frequency statistics on the preprocessed search keywords. For each search record, the total number of all keywords in it is counted. For each keyword, its TF value is multiplied by the corresponding IDF value to obtain the TF-IDF value of the keyword in all search records within 30 days. The top 100 keywords with the highest TF-IDF value are selected as high-frequency keywords. Based on the high-frequency keywords and the category labels of the product classification system, the product categories that the user may be interested in are determined. For example, if the high-frequency keyword includes "sports shoes", then it can be considered that the user is interested in the product category "sports-shoes". A unique integer code starting from 0 is assigned to each determined product category of interest. For example, "sports-shoes" is coded as 0, "women's clothing-dress" is coded as 1, and so on. The user's feature vector is clustered using the selected clustering algorithm. Divide users into different groups. Based on the clustering results, assign an interest category number to each group and create a mapping rule for the digital coding of interest categories. For example, "1" represents the outdoor sports interest group, "2" represents the technology product interest group, and "3" represents the literary creation interest group. Some groups may be interested in different fields such as outdoor sports, technology products, and literary creation. The dominant interest category can be determined by counting the frequency of users in the group mentioning relevant keywords, or analyzing their browsing volume and purchasing behavior on related content.

[0025] In one embodiment of the present invention, combining the processed age, gender, occupation, and education level features into a feature matrix, and obtaining the consumption preference cluster categories and consumption preference numbers based on the clustering results, includes the following steps: Save the name as a separate field in the dataset, convert the user's date of birth into age, calculate the difference between the current date and the date of birth to get the user's age, divide the age into different age groups, and assign a value to each age group; convert the gender into a numerical code; determine the user's occupational category based on the collected user's occupational information and refer to the national occupational classification standards, specify the corresponding classification rules for each occupation, use the text similarity algorithm to calculate the similarity between the occupation name and the keywords of each occupational category, classify the occupation into the occupational category with the highest similarity, and create a two-dimensional code The matrix has the number of rows and columns of occupational categories. Initially, all elements are set to 0. For each occupational category, the corresponding row and column positions in the coding matrix are set to 1, so that each occupational category has a unique one-hot coding vector. The education level is converted from low to high into an ordered numerical code. The occupational codes are standardized, and the processed age, gender, occupation and education level features are combined into a feature matrix. The standardized feature matrix is ​​clustered using the selected clustering algorithm. The consumption preference cluster category is obtained based on the clustering results, and a unique consumption preference number is assigned to each label.

[0026] Specifically, extract the name field separately from the dataset and store it in a new list or data frame for subsequent use. Add the calculated age to a new column in the dataset. Divide the age by every 10 years starting from 0 to determine the age group of each user. For example, 0-9 years old is the first age group, 10-19 years old is the second age group, and so on. Assign a unique numerical code to each age group, starting from 0 and adding 1 sequentially. For example, the first age group is coded as 0 and the second age group is coded as 1. Add the age group code to a new column in the dataset. For each user, if the gender is male, it is coded as 1; if the gender is female, it is coded as 0. According to the reference national occupational classification standards, sort out the keywords of each major occupation and its corresponding classification rules, use the text similarity algorithm to calculate the similarity between the occupation name and the keywords of each major occupation, select the major occupation with the highest similarity as the occupation classification result of the user, determine the number of major occupations, so that each major occupation has a unique one-hot encoding direction For example, if there are five major occupational categories, "medical and health" can be encoded as [1,0,0,0,0], "education" can be encoded as [0,1,0,0,0], and so on. According to the occupational classification results of each user, find the corresponding one-hot encoding vector, and determine an order from low to high according to the common education levels, such as elementary school, junior high school, high school, college, undergraduate, master's, doctoral, etc. The processed age, gender, occupation, and education level features are merged into a feature matrix. According to the clustering results, a unique consumption preference number is assigned to each cluster category, starting from 1. For example, "1" represents high consumption preference, "2" represents medium consumption preference, and "3" represents low consumption preference.

[0027] In one embodiment of the present invention, calculating the most active time period of the user on the platform and finding the corresponding most active time period number includes the following steps: The content, publishing time, and publisher ID information of each Moments or dynamic post are recorded; like data includes the record of users liking or giving likes including the corresponding timestamp; comment data includes the collection of users' comments on Moments or dynamics and the frequency of comments, commenter ID, and comment time; forwarding data includes the record of the number of users forwarding content and the timestamp of forwarding. Each hour is simply numbered, and the number of content posted by users, the number of likes received and given, the number of comments, the number of replies, and the number of forwardings are counted every hour of every day within 30 days to obtain the comprehensive value of users' social interaction activity every hour; The formula for calculating the comprehensive value of user social activity is: in, Indicates the comprehensive value of user social activity. Indicates the number of user-posted content. Indicates the likes given by the user. Indicates the number of likes received by the user. Indicates the number of user comments, Indicates the number of comments on the user. Indicates the number of content forwarded by the user; Add up the user's social activity comprehensive values ​​corresponding to each hour of each day for 30 days to obtain the total comprehensive value for each hour. Find the hour with the highest total comprehensive value as the user's most active time period on the platform, and find the corresponding most active time period number.

[0028] Specifically, a data storage structure is designed to save relevant information about Moments or dynamics, and a data collection program or interface is written. When a user posts a Moments or dynamic, the following information is automatically recorded: content, posting time, publisher ID, and this information is stored in a pre-designed data storage structure. The data storage structure is expanded to add a field for storing likes data. The data storage structure is further expanded to add a field for storing comment data. The data storage structure is further expanded to add a field for storing forwarded data and store it in the corresponding data storage structure. The 24 hours of a day are divided into 24 1-hour time periods, and each hour period is simply numbered. For example, 0:00-1:00 is hour period 0, 1:00-2:00 is hour period 1, and so on, until 23:00- 24:00 is the 23rd hour period. A data analysis program is written to extract data from the past 30 days from the data storage structure. For each record, the hour period and number to which the record belongs are determined based on its release time. The number of distributed content, the number of likes, the number of comments, and the number of reposts are counted separately to determine the hour period to which the corresponding data belongs. For each hour period, the set formula is used to calculate the user's social interaction activity comprehensive value for each hour of each day for the past 30 days. The corresponding user social activity comprehensive values ​​for each hour of each day for the past 30 days are added together to obtain the total comprehensive value for each hour. The one-dimensional array storing the total comprehensive values ​​is traversed to find the hour with the highest total comprehensive value. The number of that hour is recorded, which is the number of the time period when the user is most active on the platform.

[0029] In one embodiment of the present invention, the step of constructing the optimal value for user product purchases includes the following steps: Determine whether the user has enabled the location permission for the APP, use the positioning API provided by the mobile phone operating system to interact with the communication and data of the mobile phone GPS hardware, obtain the user's basic location coordinate data, altitude data and timestamp data, build a data receiving interface on the back end of the APP that matches the front end location information acquisition function module, use the HTTPS network communication protocol to encrypt and transmit the location information data, and store the user's basic location coordinate data, altitude data, timestamp data and the unique identifier information associated with the user account in the database on the platform server side, associate and integrate the inventory information of all products on the platform with the geographic location information of warehouses and stores in various places, establish a product inventory geographic location database, add the geographic location coordinates, inventory type, inventory quantity and basic product information of the warehouse or store where each product record is located in the database, calculate the straight-line distance between the user's location and the warehouse where the product is located, divide the product type that the user is interested in by the total product type to obtain the category coverage, normalize the quantity and popularity of the products to the maximum and minimum values ​​respectively to obtain the quantity normalization value and the popularity normalization value, and combine them to obtain the optimal value for the user's product purchase; The formula for calculating the optimal value of user product purchases is: in, Indicates the optimal value of the user's product purchase, Indicates the straight-line distance between the user's location and the product's warehouse. Indicates the types of products that the user is interested in. Indicates the total product category. represents the quantity normalized value, represents the normalized value of heat, and Indicates the weight coefficient corresponding to distance and product.

[0030] Specifically, when a user installs or updates the APP for the first time and starts the APP, the user is explicitly requested to enable location permission through a pop-up window or other means, and the positioning API provided by the mobile phone operating system is used to communicate with the mobile phone GPS hardware. For example, in Android, you can use the integrated positioning service, write code logic, and obtain the user's basic location coordinate data, altitude data, and timestamp data through the positioning API. A data receiving interface that matches the front-end location information acquisition function module is built on the back-end of the APP. This interface can use the HTTPS network communication protocol to ensure the security and encryption of data transmission. After the front-end APP obtains the user's location information, this information is packaged in a certain format and sent to the back-end server through the HTTPS protocol. The table for storing user location information can contain fields such as user ID, longitude, latitude, altitude, and timestamp. The table for storing product inventory geographic location information can contain product I D. Warehouse or store ID, geographic location coordinates, inventory types, inventory quantities, and basic product information. On the server side, write corresponding code logic to receive the user location information sent by the front end, and store the parsed user location information and the associated user account unique identifier information in the platform server database. According to the inventory information of all products on the platform and the geographic location information of warehouses and stores in various places, associate and integrate the two to establish a product inventory geographic location database. Traverse all product records in the product inventory geographic location database. For each product, calculate the straight-line distance between the warehouse or store where it is located and the user's current location coordinates. Calculate the category coverage rate based on the product categories and total product categories that the user is interested in. Based on the straight-line distance between the user's location and the warehouse where the product is located, the category coverage rate, the quantity normalization value, and the popularity normalization value obtained by the above calculations, perform joint calculations to obtain the optimal value for the user's product purchase; among them, The value of is 0.6, The value of is 0.4.

[0031] In one embodiment of the present invention, step three includes the following steps: The collected user interest category numbers, consumption preference numbers, most active time period numbers and user product purchase optimal values ​​are combined into a feature vector. The feature vector similarity between users is calculated through cosine similarity and Pearson correlation coefficient, and the average of the two is taken as the final similarity. A similarity threshold is set, and the user group exceeding the similarity threshold is determined as the neighbor of the target user. The purchase of products by this user group is counted. The recommended products are screened according to the interest category number and consumption preference number of the target user. Combined with the most active time period number and the user's product purchase optimal value, the time is determined to push the nearest store with sufficient products to the user; the collected user data is used as part of the training set to train the collaborative filtering model. When the user logs in to the system or browses products, the user's latest data is obtained in real time, the information in the feature vector is updated, and the updated feature vector is input into the trained model to generate a personalized recommendation list.

[0032] Specifically, a unique identifier is established for each user, and the collected user interest category number, consumption preference number, most active time period number and user product purchase optimal value are matched with it to form a feature vector. The feature vectors of all users are traversed, and the cosine similarity of each two is calculated to obtain a cosine similarity matrix. Similarly, the feature vectors of all users are traversed, and their Pearson correlation coefficient is calculated between each two to obtain a Pearson correlation coefficient matrix. The corresponding values ​​of the cosine similarity and the Pearson correlation coefficient are averaged to obtain the final similarity matrix. According to the set similarity threshold, the final similarity matrix is ​​traversed to find user pairs that exceed the threshold. For each user, group other users whose purchases exceed the threshold into a set, which is the user's neighbors. For the target user, find the neighbor group determined in the previous step, and count the purchases of each product in the neighbor group. You can classify and count the products according to the product number or name. According to the interest category number and consumption preference number, filter out qualified products from the counted neighbor purchases. For example, if the target user's interest category number contains a specific category, select the products belonging to this category. If the target user's consumption preference number is a specific preference, further filter out the products that meet this preference. Get the target user's most active time period number and the user's optimal product purchase value. According to the most active time period number, find out the stores that are open and The store closest to the target user, at the same time, considering the optimal value of the user's product purchase, give priority to those stores with sufficient products; based on the above factors, determine the personalized store recommendation list that will be pushed to the target user; select a part of the collected user data as a training set, including the user's feature vector and their corresponding purchase behavior records, and use the training set to train the collaborative filtering model. When the user logs in to the system or browses products, the system's tracking and recording functions are used to obtain the user's latest data in real time, and the updated feature vector is input into the trained collaborative filtering model. The model calculates and analyzes the input feature vector to generate a personalized recommendation list and recommend it to the user; if the registered user is a new user, analysis is performed based on the new user's initial various data.

[0033] See also Figure 2 As shown, the present invention is a digital management system based on multivariate data, including the following modules: Multivariate data collection module: collects user multivariate data through various data sources; User multivariate data preprocessing and feature construction module: This module preprocesses the collected user multivariate data, clusters the user's high-frequency keywords and interest categories using cluster analysis methods, assigns an interest category number, and combines the processed age, gender, occupation, and education level features into a feature matrix. Based on the clustering results, the consumption preference cluster categories are obtained and the consumption preference numbers are obtained. The user's most active time period on the platform is calculated, and the corresponding most active time period number is found to construct the optimal value for the user's product purchases. Collaborative filtering recommendation module: Combines the user's interest category number, consumption preference number, most active time period number, and the user's optimal purchase value to obtain a feature vector to build a collaborative filtering model. It obtains and updates the user's latest data in real time and inputs it into the trained collaborative filtering model to generate a personalized recommendation list.

[0034] The above embodiments are only used to illustrate the technical method of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A digital management method based on multivariate data, characterized in that: The following steps are involved: Step 1: Collect user multivariate data through various data sources; Step 2: Preprocess the collected multivariate user data. Use cluster analysis to cluster the user's high-frequency keywords and interest categories, assigning them an interest category number. Combine the processed age, gender, occupation, and education characteristics into a feature matrix. Based on the clustering results, derive the consumption preference cluster categories and consumption preference numbers. Calculate the user's most active time period on the platform, find the corresponding most active time period number, and construct the user's optimal purchase value. Step 3: Combine the user's interest category ID, consumption preference ID, most active time period ID, and the user's optimal purchase value to obtain a feature vector to build a collaborative filtering model. The latest user data is obtained and updated in real time, and input into the trained collaborative filtering model to generate a personalized recommendation list.

2. The digital management method based on multivariate data according to claim 1, characterized in that: The step 1 comprises the following steps: The server records the user behavior data on websites with search functions and social media platforms. The user behavior data includes the user's search keywords, product browsing, adding to shopping carts and purchases, as well as the corresponding time. Basic information, including name, date of birth, and gender, is obtained through the personal information users enter when registering accounts on websites and applications. A questionnaire is embedded in the user registration process to obtain the user's occupation and education level information. User profile data is then generated based on the user's basic information, occupation, and education level information. Through the authorization mechanism or API interface provided by the social platform, find the application that the user has authorized to access and record the user's social data within the specified permissions. Social data includes the content posted in the social platform's circle of friends or dynamics, as well as likes, comments, and forwarding; The user's location information data is obtained through the GPS hardware of the user's mobile phone. The location information data includes the user's basic location coordinate data, altitude data and timestamp data.

3. The digital management method based on multivariate data according to claim 1, characterized in that: The pre-processing operation on the collected user multivariate data includes the following steps: After the collected multivariate data is transmitted to the server, the collected user behavior data is cleaned and various user behaviors are classified into four categories, including searching keywords, browsing products, adding to shopping carts, and purchasing. All timestamps are converted into a unified format, and user behaviors are divided according to time series; Perform data integration and missing value processing on the collected user portrait data, and convert the various data contained in the user portrait data using coding; Clean the collected social data; The collected location information data is cleaned, and the geographic coordinates of the location information are converted into a coordinate system suitable for analysis. The address information contained in the location information is standardized and converted into a unified address format.

4. The digital management method based on multivariate data according to claim 1, characterized in that: The method of clustering the user's high-frequency keywords and interest categories by a cluster analysis method and assigning an interest category number includes the following steps: Collect user search keyword records within 30 days from the platform, including search time and keyword content, obtain user product browsing records, including product browsing time, product category and specific product information, and collect user purchase records, including purchase time and various attributes of purchased products, and add a timestamp to each user behavior data; perform text preprocessing on user search keywords, use word frequency statistics to count the probability of different keywords appearing in user search behavior to obtain the word frequency of each keyword, apply TF-IDF technology, divide the number of times a keyword appears in the user's search record by the total number of words in the search record to obtain the TF value of the keyword in the search record, calculate the corresponding number of each keyword appearing in all search records within 30 days, divide the total number of all search records by the total number of times the keyword appears, and then take the logarithm to obtain the IDF value of the keyword, and multiply the TF value of each keyword by the corresponding IDF value to obtain the TF-IDF value of the keyword in all search records within 30 days; Based on the calculated word frequency and TF-IDF value, high-frequency keywords are determined. Based on the high-frequency keywords and the category labels of the product classification system, the product categories that the user may be interested in are determined. A unique integer code starting from 0 is assigned to each product category. The user's high-frequency keywords and interest categories are converted into feature vectors suitable for clustering algorithm processing. The user's high-frequency keywords and interest categories are clustered through clustering analysis methods. Based on the clustering results, the users are divided into different groups, and each group is assigned an interest category number.

5. The digital management method based on multivariate data according to claim 1, characterized in that: The method combines the processed age, gender, occupation and education level features into a feature matrix, and obtains the consumption preference cluster category and the consumption preference number according to the clustering result, including the following steps: Save the name as a separate field in the dataset, convert the user's date of birth into age, calculate the difference between the current date and the date of birth to get the user's age, divide the age into different age groups, and assign a value to each age group; convert the gender into a numerical code; determine the user's occupational category based on the collected user's occupational information and refer to the national occupational classification standards, specify the corresponding classification rules for each occupation, use the text similarity algorithm to calculate the similarity between the occupation name and the keywords of each occupational category, classify the occupation into the occupational category with the highest similarity, and create a two-dimensional code The matrix has the number of rows and columns of occupational categories. Initially, all elements are set to 0. For each occupational category, the corresponding row and column positions in the coding matrix are set to 1, so that each occupational category has a unique one-hot coding vector. The education level is converted from low to high into an ordered numerical code. The occupational codes are standardized, and the processed age, gender, occupation and education level features are combined into a feature matrix. The standardized feature matrix is ​​clustered using the selected clustering algorithm. The consumption preference cluster category is obtained based on the clustering results, and a unique consumption preference number is assigned to each label.

6. The digital management method based on multivariate data according to claim 1, characterized in that: Calculating the user's most active time period on the platform and finding the corresponding most active time period number includes the following steps: The content, publishing time, and publisher ID information of each Moments or dynamic post are recorded; like data includes the record of users liking or giving likes including the corresponding timestamp; comment data includes the collection of users' comments on Moments or dynamics and the frequency of comments, commenter ID, and comment time; forwarding data includes the record of the number of users forwarding content and the timestamp of forwarding. Each hour is simply numbered, and the number of content posted by users, the number of likes received and given, the number of comments, the number of replies, and the number of forwardings are counted every hour of every day within 30 days to obtain the comprehensive value of users' social interaction activity every hour; The formula for calculating the comprehensive value of user social activity is: in, Indicates the comprehensive value of user social activity. Indicates the number of user-posted content. Indicates the likes given by the user. Indicates the number of likes received by the user. Indicates the number of user comments, Indicates the number of comments on the user. Indicates the number of content forwarded by the user; Add up the user's social activity comprehensive values ​​corresponding to each hour of each day for 30 days to obtain the total comprehensive value for each hour. Find the hour with the highest total comprehensive value as the user's most active time period on the platform, and find the corresponding most active time period number.

7. The digital management method based on multivariate data according to claim 1, characterized in that: The method of constructing the optimal value for user commodity purchases includes the following steps: Determine whether the user has enabled the location permission for the APP, use the positioning API provided by the mobile phone operating system to interact with the communication and data of the mobile phone GPS hardware, obtain the user's basic location coordinate data, altitude data and timestamp data, build a data receiving interface on the back end of the APP that matches the front end location information acquisition function module, use the HTTPS network communication protocol to encrypt and transmit the location information data, and store the user's basic location coordinate data, altitude data, timestamp data and the unique identifier information associated with the user account in the database on the platform server side, associate and integrate the inventory information of all products on the platform with the geographic location information of warehouses and stores in various places, establish a product inventory geographic location database, add the geographic location coordinates, inventory type, inventory quantity and basic product information of the warehouse or store where each product record is located in the database, calculate the straight-line distance between the user's location and the warehouse where the product is located, divide the product type that the user is interested in by the total product type to obtain the category coverage, normalize the quantity and popularity of the products to the maximum and minimum values ​​respectively to obtain the quantity normalization value and the popularity normalization value, and combine them to obtain the optimal value for the user's product purchase; The formula for calculating the optimal value of user product purchases is: in, Indicates the optimal value of the user's product purchase, Indicates the straight-line distance between the user's location and the product's warehouse. Indicates the types of products that the user is interested in. Indicates the total product category. represents the quantity normalized value, represents the normalized value of heat, and Indicates the weight coefficient corresponding to distance and product.

8. The digital management method based on multivariate data according to claim 1, characterized in that: The step three comprises the following steps: The collected user interest category numbers, consumption preference numbers, most active time period numbers and user product purchase optimal values ​​are combined into a feature vector. The feature vector similarity between users is calculated through cosine similarity and Pearson correlation coefficient, and the average of the two is taken as the final similarity. A similarity threshold is set, and the user group exceeding the similarity threshold is determined as the neighbor of the target user. The purchase of products by this user group is counted. The recommended products are screened according to the interest category number and consumption preference number of the target user. Combined with the most active time period number and the user's product purchase optimal value, the time is determined to push the nearest store with sufficient products to the user; the collected user data is used as part of the training set to train the collaborative filtering model. When the user logs in to the system or browses products, the user's latest data is obtained in real time, the information in the feature vector is updated, and the updated feature vector is input into the trained model to generate a personalized recommendation list.

9. A digital management system based on multivariate data, using the digital management method based on multivariate data according to any one of claims 1 to 8, characterized in that: Includes the following modules: Multivariate data collection module: collects user multivariate data through various data sources; User multivariate data preprocessing and feature construction module: This module preprocesses the collected user multivariate data, clusters the user's high-frequency keywords and interest categories using cluster analysis methods, assigns an interest category number, and combines the processed age, gender, occupation, and education level features into a feature matrix. Based on the clustering results, the consumption preference cluster categories are obtained and the consumption preference numbers are obtained. The user's most active time period on the platform is calculated, and the corresponding most active time period number is found to construct the optimal value for the user's product purchases. Collaborative filtering recommendation module: Combines the user's interest category number, consumption preference number, most active time period number, and the user's optimal purchase value to obtain a feature vector to build a collaborative filtering model. It obtains and updates the user's latest data in real time and inputs it into the trained collaborative filtering model to generate a personalized recommendation list.

Images