E-commerce platform user portrait generation method based on neural network technology

Abstract

The invention relates to the technical field of data processing, in particular to an e-commerce platform user portrait generation method based on a neural network technology. Client data is cleaned, governed and subjected to clustering analysis to create an accurate user portrait for a user, and an algorithm model is iteratively adjusted in the development of operation experience; subtracting the mean value and then dividing the mean value by the standard deviation, selecting an initial point by optimizing a clustering analysis result and utilizing a strategy that the farther the clustering distance is, the better the clustering distance is, and continuously adjusting and repeatedly calculating to realize the best effect; the advantages of the user portraits based on the neural network technology can help an e-commerce platform to realize precise marketing, improve user experience, predict market trend, provide customized services, optimize advertisement putting and the like. By continuously and deeply knowing and analyzing the user portraits, the e-commerce platform can better meet the user requirements and improve the brand value, thereby realizing continuous and stable development.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method for generating user portraits of an e-commerce platform based on neural network technology. Background Art

[0002] In recent years, with the continuous development of artificial intelligence technology, more and more e-commerce platforms have begun to use neural networks to build user portraits in order to better understand user needs and behaviors. These user portraits can help companies develop more accurate marketing strategies, improve user satisfaction and loyalty, and promote sales growth.

[0003] When building a mall user portrait based on a neural network, it is usually necessary to collect a large amount of data, including basic user information (such as gender, age, region, etc.), purchase behavior, browsing history, search history, etc. Through these data, we can dig out information such as user interests, purchasing habits, and demand pain points.

[0004] In order to improve the accuracy of user portraits, many e-commerce platforms also use user behavior data as input and train neural network models. These models can learn user characteristics and preferences and automatically classify and label users. For example, based on data such as user purchase behavior and browsing history, users’ purchase intentions and interest preferences can be inferred, thereby recommending more accurate products and services to users.

[0005] In addition to improving the accuracy of marketing strategies, the mall user portrait based on neural networks can also help companies with supply chain management and inventory optimization. For example, by analyzing user purchasing behavior, sales trends in the future can be predicted, so that inventory planning and adjustments can be made in advance. In addition, product design and quality can be optimized based on user feedback and evaluation data to improve user satisfaction and loyalty. Summary of the invention

[0006] As one of the ways to promote e-commerce malls, user portraits have become an important tool for e-commerce companies to improve marketing effectiveness and optimize supply chain management. In the future, with the continuous development of artificial intelligence technology and the continuous expansion of application scenarios, it is believed that the user portraits of malls based on neural networks will become more and more refined and intelligent. Through the analysis of user portraits, e-commerce platforms can also provide customized services. By analyzing a large amount of user portrait data, e-commerce platforms can predict market trends and do inventory management and supply chain optimization in advance. For example, when the data shows that a large number of users begin to pay attention to a certain product of a certain brand, the platform can adjust the inventory in time to ensure that the supply of the product is sufficient. This can improve user satisfaction and reduce inventory costs.

[0007] The advantages of user portraits based on neural network technology can help e-commerce platforms achieve precision marketing, improve user experience, predict market trends, provide customized services, and optimize advertising. By continuously and deeply understanding and analyzing user portraits, e-commerce platforms can better meet user needs, enhance brand value, and thus achieve sustainable and stable development.

[0008] The present invention solves the existing technical problems and provides the following technical solutions: a method for generating user portraits of an e-commerce platform based on neural network technology, comprising the following steps:

[0009] 1. Data collection and processing: First, we need to collect various user data, including basic user information such as region, age, shopping mall browsing time, browsing history, purchase history, and search history; and process this data to clean out irrelevant and abnormal data.

[0010] Data cleaning is as follows:

[0011] Step 1: Eliminate erroneous data by identifying and correcting erroneous data, such as missing values, outliers, and duplicate values;

[0012] Step 2: Remove redundant data, delete duplicate data records, and reduce data redundancy;

[0013] Step 3: Standardize the data and convert it into a unified format and unit;

[0014] Step 4: Process missing data by filling missing values ​​or deleting missing values;

[0015] Step 5: Process abnormal data, identify and process outliers;

[0016] 2. Extract and model user information features, extract relevant features for different types of user groups, and establish user portrait models:

[0017] Extraction of basic information: For young user groups, focus on age, education level, personality type and social media preferences; for corporate user groups, focus on corporate size, industry and operating conditions; By extracting basic information of different types of user groups, understand their characteristics and needs;

[0018] Behavioral characteristics analysis: The behavioral characteristics of different types of user groups are also different. Young user groups tend to shop online and browse social media, while corporate user groups tend to engage in business activities such as purchasing and sales. By analyzing the behavioral characteristics of different types of user groups, we can understand their interests, hobbies, purchasing habits, and decision-making methods.

[0019] Mining interest preferences: Different types of user groups have different interest preferences. Young user groups are interested in fashion, music and movies, while corporate user groups are interested in industry dynamics, management knowledge and market trends. By mining the interest preferences of different types of user groups, we can provide accurate content recommendations and service customization.

[0020] Social network analysis: Different types of user groups have different social networks. Young user groups tend to share and communicate on social media, while corporate user groups focus on brand promotion and customer relationship management on social media. By analyzing the social networks of different types of user groups, we can understand their social habits, preferences, attitudes and other information and provide personalized services.

[0021] User geographic location analysis: Different types of user groups have different geographic locations. Urban users tend to shop online and order takeout, while rural users focus on agricultural product sales and agricultural production. By analyzing the geographic locations of different types of user groups, we can understand their activity ranges and travel habits, and provide accurate service positioning and promotion.

[0022] These features are analyzed by neural network clustering algorithm, the RFM model is expanded to GARFM model, and the neural network clustering algorithm K-means is used to cluster the data through unsupervised learning. By clustering similar data points together, patterns and structures in the data are discovered. In terms of feature selection, these algorithms are used to identify features associated with specific clusters.

[0023] Beneficial effects of the present invention:

[0024] 1. Customer portrait creation: Clean and manage customer data, and use cluster analysis to create accurate user portraits for users. At the same time, iteratively adjust the algorithm model based on the development of operational experience;

[0025] 2. Data standardization: subtract the mean and divide by the standard deviation, optimize the cluster analysis results, select the initial point using the strategy of the farther the cluster distance, the better, and continuously adjust and repeatedly calculate to achieve the best effect;

[0026] 3. Based on the neural network technology, we successfully divided customers into different groups by clustering customer shopping data, which are differentiated according to their purchasing power, purchasing habits, shopping mall browsing time, etc. This provides important market segmentation information for the mall, helping mall operators to formulate better marketing strategies and targeted product recommendations, thereby achieving customer satisfaction and loyalty, and increasing turnover and profits. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1A flowchart of a method for generating user portraits for an e-commerce platform based on neural network technology;

[0028] Figure 2 A schematic diagram of segmenting user information according to experience value;

[0029] Figure 3 This is the effect diagram of the first iteration of clustering;

[0030] Figure 4 This is the effect diagram of the second iteration of clustering;

[0031] Figure 5 It is a cluster scatter plot;

[0032] Figure 6 It is the classification rule diagram of customer value space of GARFM model; DETAILED DESCRIPTION

[0033] The present invention provides a method for generating user portraits of an e-commerce platform based on neural network technology to improve user satisfaction and loyalty and enhance market competitiveness in the current e-commerce development process. The specific scheme is as follows:

[0034] (I) Data collection and processing: First, we need to collect various user data, such as basic user information, such as the region, age, shopping mall browsing time, browsing history, purchase history, search history, etc., and process these data to clean out irrelevant and abnormal data.

[0035] The goal of data cleaning is to improve data quality, ensure the effectiveness and continuity of data analysis, and make the data suitable for subsequent analysis and modeling. The basic steps of data cleaning are as follows:

[0036] Step 1: Eliminate erroneous data and improve data accuracy by identifying and correcting erroneous data, such as missing values, outliers, duplicate values, etc.

[0037] Step 2: Remove redundant data, delete duplicate data records, reduce data redundancy, and improve data efficiency.

[0038] Step 3: Standardize the data and convert it into a unified format and unit for comparison and analysis.

[0039] Step 4: Process missing data by filling in missing values ​​or deleting missing values ​​to ensure data integrity.

[0040] Step 5: Process abnormal data, identify and process outliers to avoid adverse effects on analysis results.

[0041] (II) Extract and model user information features, extract relevant features for different types of user groups, and establish user portrait models. This can be done from the following perspectives:

[0042] Basic information extraction: The focus of basic information extraction varies for different types of user groups. For young user groups, basic information may focus more on age, education, personality type, and social media preferences; for corporate user groups, it may focus more on information such as corporate size, industry, and operating conditions. By extracting basic information of different types of user groups, we can better understand their characteristics and needs, and provide basic data for subsequent feature extraction and modeling.

[0043] Behavioral characteristics analysis: Different types of user groups have different behavioral characteristics. For example, young user groups may be more inclined to online shopping and social media browsing, while corporate user groups are more inclined to business activities such as purchasing and sales. By analyzing the behavioral characteristics of different types of user groups, we can better understand their interests, purchasing habits, decision-making methods, and other information, and then provide them with more personalized services and product recommendations.

[0044] Mining interest preferences: Different types of user groups have different interest preferences. For example, young user groups may pay more attention to fashion, music, and movies, while corporate user groups pay more attention to industry dynamics, management knowledge, and market trends. By mining the interest preferences of different types of user groups, we can provide them with more accurate content recommendations and service customization.

[0045] Social network analysis: Different types of user groups have different social networks. For example, young user groups may be more inclined to share and communicate on social media, while corporate user groups pay more attention to brand promotion and customer relationship management on social media. By analyzing the social networks of different types of user groups, we can better understand their social habits, preferences, attitudes and other information, and provide them with more personalized services.

[0046] User geographic location analysis: Different types of user groups have different geographic locations. For example, urban users may prefer services such as online shopping and takeout ordering, while rural users pay more attention to services such as agricultural product sales and agricultural production. By analyzing the geographic locations of different types of user groups, we can better understand their activity range, travel habits and other information, and provide them with more accurate service positioning and promotion.

[0047] These features are analyzed by the neural network clustering algorithm. In this stage, we can expand the RFM model to the GARFM model, and further adopt the neural network clustering algorithm K-means to cluster the data through unsupervised learning. By clustering similar data points together, patterns and structures in the data can be discovered. In terms of feature selection, these algorithms can be used to identify features associated with specific clusters. For example, for a specific cluster, you can select those features that have the greatest impact on the clustering results. In addition, neural networks can be used to reduce the dimension of features to reduce the number of features and improve the performance of the model. Through the association rule machine learning FP-Growth algorithm, association rules in the data set are discovered. These rules reveal the potential relationship between data items and can be used for feature selection and association analysis between features. For example, association rules can be used to identify combinations of features that appear or are missing at the same time, which may be closely related to the target variable. By using association rules, interesting relationships between features can be discovered and combined to build better models.

[0048] (III) To be further specific: Among the numerous customer segmentation models of traditional enterprises and e-commerce, the RFM model is widely mentioned and used, namely Rencency (most recent consumption), Frequency (consumption frequency), and Monetary (consumption amount), a combination of the first letters of three indicators. With the current large e-commerce platforms, factors such as region and age account for an increasingly large proportion. We can expand the RFM model to the GARFM model, namely Region, Age, Rencency (most recent consumption), Frequency (consumption frequency), and Monetary (consumption amount).

[0049] (IV) Collect statistics on all users based on region, age, time of most recent consumption, number of consumptions, consumption amount, and browsing time.

[0050] In general, let the sample set D = {x1; x2; x3; ....; x m} contains m unlabeled samples, and sample x i =(x i1 ;x i2 ;x i3 ;....;x im ) is an n-dimensional feature vector. The clustering algorithm divides the sample D into p non-overlapping clusters {C p |p=1,2,3,...,k}, where

[0051] Then normalize And, j ∈{1,2,3,...,k} represents the sample xj The "cluster marker" of Clustering is performed with k points in space as the center, and the objects closest to them are classified. Through an iterative method, the value of each cluster center is updated successively until the best clustering result is obtained. Similarity calculation is performed, and the degree of difference between factors is measured by distance:

[0052]

[0053] Cluster centers, the centroids of the calculated clusters:

[0054]

[0055] According to the convergence criterion, calculate the square error function

[0056]

[0057] (V) Data standardization: subtract the mean and divide by the standard deviation

[0058]

[0059] Finally, the clustering result can be represented by a cluster label vector λ=(λ1; λ2; λ3; ...; λ m )express. Specific embodiments

[0061] Step 1: Use the order details of the platform in the past six months as the data source, sample and retain more than 50,000 items, with a span of 6 months and involving more than 10,000 different customers, and pre-process the data;

[0062] user_id,area,age,total_order_count,total_order_price,last_order_date

[0063] 21526959,110000,30,3010,8,2017-07-1319:22:43.0

[0064] 38972904,620105,40,230,35,2017-08-1110:51:06.0

[0065] 33666451,440000,25,7800,6,2017-02-2610:48:20.0

[0066] 39218462,440000,18,3000,22,2017-08-17 23:31:49.0

[0067] 33291929,120000,36,4500,20,2018-02-1218:17:54.0

[0068] 71742617,650000,50,132,9,2018-01-1000:57:43.0

[0069] 25114544,110000,32,5098,35,2017-02-1715:57:25.0 ......

[0071] Step 2: According to general experience, feature selection (GARFM model, from G-region, A-age, R-recent purchase, F-purchase frequency, M-average consumption amount), such as Figure 2 ;

[0072] Step 3: Randomly select a sample point as the first cluster center, such as Figure 3 ;

[0073] 1. Initialize each feature sample as a separate class;

[0074] 2. Calculate the distance between the remaining sample points and the existing cluster center (if there are multiple cluster centers, first calculate the center of these cluster centers, and then calculate the distance between the remaining sample points and the center). The greater the distance, the greater the probability of being selected as the next cluster center (assign a probability), and then use the roulette method to draw the next cluster center.

[0075] 3. The two classes with the smallest distance are merged into a new class;

[0076] 4. Recalculate the distance between the new class and other classes;

[0077] 5. Repeat the calculation until there is only one class left.

[0078] Number of samples G A R F M 0 320 0.488287 0.712173 -0.645968 1.153902 0.260523 1 234 0.496397 -0.546175 -0.206534 -0.546175 0.366205 2 189 -0.206534 0.488434 0.488433 -0.645968 -0.673498 3 12 0.363572 0.712173 0.612173 0.812173 7.363321 4 232 0.137703 0.496397 0.863572 -0.727579 0.273264 5 321 -0.243234 -0.727579 -0.643460 -0.137703 -0.853512 6 233 0.143234 1.096498 -0.592992 1.196498 0.152670

[0079] The ideal effect pursued by KMeans is to minimize the distance within the same cluster and maximize the distance between different cluster groups. In fact, after clustering to this step, some conclusions can be drawn. In fact, some categories contain a small number of samples, such as category 3 and category 4, so you can run the above steps again and set the k value to a smaller value.

[0080] Step 4: After drawing the cluster diagram using SPSS, continue to adjust the elements and observe the cluster density.

[0081] In the first iteration, six cluster centers are randomly generated. Based on the cluster centers, we generate a dividing line according to the most recent access time, with 7 points within the last 20 days, 8 points between 20 and 40 days, and 4 points greater than 40 days.

[0082] In the second iteration, clustering is regenerated based on the first iteration.

[0083] The nth iteration is adjusted repeatedly in a similar manner.

[0084] Step 5: Finally, we get the target customer group and formulate operation strategies for the customer group. The following conclusions are drawn through data analysis:

[0085] 1. User regional distribution: The top three regions where consumer users are located are Guangdong, Shanghai, and Beijing. That is, users in first-tier cities account for a large proportion, and operational strategies can be prioritized for such users.

[0086] 2. User age distribution: The male-female ratio is close to 1:1, and operations can be conducted regardless of gender;

[0087] 3. User age distribution: The top three age groups are [18-25], [26-35], and [36-45]. Consumer users are mainly concentrated in the 18-35 age group, accounting for 70%. Operational strategies can be based on user age and priorities can be set.

[0088] 4. User grouping: The ratio of general customers to important customers is about 1:1, among which general retention customers (R low\F low\M low) account for the highest proportion, about 30%.

[0089] 5. For input indicator selection, only G, A, R, F, and M are selected because the source data information is limited. If conditions permit, more indicators can be included in the modeling. The number of categories is determined based on actual needs and the feedback effect of the model results.

[0090] Glossary

[0091] User portrait: Modeling is done based on the user's basic information, consumption behavior, social background, interests and hobbies, etc., to provide enterprises with comprehensive user information. By analyzing this user information, enterprises can gain a deep understanding of the characteristics of target users, including their age, gender, region, occupation, income, etc. Through user portraits, enterprises can better understand users' consumption habits, as well as their needs and preferences for products. In this way, they can better formulate product design and marketing strategies for these target users, and increase the market share and sales of products.

[0092] Neural network technology: (Neural Network, NN) is a complex network system formed by a large number of simple processing units (called neurons) widely connected to each other. It reflects many basic characteristics of human brain functions and is a highly complex nonlinear dynamic learning system. Neural networks have large-scale parallel, distributed storage and processing, self-organization, self-adaptation and self-learning capabilities, and are particularly suitable for processing imprecise and fuzzy information processing problems that require considering many factors and conditions at the same time.

Claims

1. A method for generating user portraits of e-commerce platforms based on neural network technology, characterized in that The steps include:

1. Data collection and processing: First, we need to collect various user data, including user basic information such as region, age, shopping mall browsing time, browsing history, purchase history, and search history; and process these data to clean out irrelevant and abnormal data; Data cleaning is as follows: Step 1: Eliminate erroneous data by identifying and correcting erroneous data, such as missing values, outliers, and duplicate values; Step 2: Remove redundant data, delete duplicate data records, and reduce data redundancy; Step 3: Standardize the data and convert it into a unified format and unit; Step 4: Process missing data by filling missing values ​​or deleting missing values; Step 5: Process abnormal data, identify and process outliers; 2. Extract and model user information features, extract relevant features for different types of user groups, and establish user portrait models: Extraction of basic information: For young user groups, focus on age, education level, personality type and social media preferences; for corporate user groups, focus on corporate size, industry and operating conditions; By extracting basic information of different types of user groups, understand their characteristics and needs; Behavioral characteristics analysis: The behavioral characteristics of different types of user groups are also different. Young user groups tend to shop online and browse social media, while corporate user groups tend to engage in business activities such as purchasing and sales. By analyzing the behavioral characteristics of different types of user groups, we can understand their interests, hobbies, purchasing habits, and decision-making methods. Mining interest preferences: Different types of user groups have different interest preferences. Young user groups are interested in fashion, music and movies, while corporate user groups are interested in industry dynamics, management knowledge and market trends. By mining the interest preferences of different types of user groups, we can provide accurate content recommendations and service customization. Social network analysis: Different types of user groups have different social networks. Young user groups tend to share and communicate on social media, while corporate user groups focus on brand promotion and customer relationship management on social media. By analyzing the social networks of different types of user groups, we can understand their social habits, preferences, attitudes and other information and provide personalized services. User geographic location analysis: Different types of user groups have different geographic locations. Urban users tend to shop online and order takeout, while rural users focus on agricultural product sales and agricultural production. By analyzing the geographic locations of different types of user groups, we can understand their activity ranges and travel habits, and provide accurate service positioning and promotion. These features are analyzed by neural network clustering algorithm, the RFM model is expanded to GARFM model, and the neural network clustering algorithm K-means is used to cluster the data through unsupervised learning. By clustering similar data points together, patterns and structures in the data are discovered. In terms of feature selection, these algorithms are used to identify features associated with specific clusters.

Images