Research On Mobile Social E-commerce Classification Model Based On Machine Learning

With the rapid development of the mobile Internet and the widespread use of mobile devices and various social software,mobile social e-commerce has become an emerging e-commerce model.This emerging e-commerce model combines the mobility and convenience of mobile devices,as well as the social and commercial nature of mobile devices.Under this emerging social e-commerce model,e-commerce is not only affected by users' personal interests and hobbies,but also by social media.Therefore,through social media and e-commerce,it is of great significance to study the interests and hobbies of users and analyze the behavior of social media users.At present,people have some preliminary research results on social e-commerce,but there are still some problems that need to be researched and solved.The first is the collection of mobile social e-commerce data.It is necessary to focus on the data flow of mobile social application data in the applications and how to collect data from third-party applications on the mobile side.Secondly,when analyzing the collected data,due to the huge number of samples required for data analysis,and there is no clear boundary between ordinary social users and e-commerce users in the data samples,which makes it more difficult to classify social e-commerce users.Moreover,social e-commerce users have not been registered uniformly,and the types of goods sold are different.How to classify social e-commerce users based on their available attributes in social apps faces unprecedented challenges.This paper conducts intensive study on the classification models of mobile social e-commerce based on the analysis of social e-commerce user behavior,and the main innovations achieved are as follows:(1)A container-based data collection scheme for social applications is proposed.The solution consists of two parts:static analysis and dynamic collection.It obtains user behavior data collection points of potential user in social applications through static analysis,the dynamic collection part is based on the data collection strategy file to collect the data generated and used in the application when it conforms to the data collection strategy.A dynamic data collection and tracking algorithm based on a lightweight virtual container combining static analysis and dynamic collection is proposed,which realizes privacy leakage detection and data collection of social data in Android applications.A data collection container is designed and developed for social applications and applied in practice.the application results show that the proposed social application data collection scheme can better solve the problem of mobile social e-commerce data collection and the distinction between ordinary social users and e-commerce users in the data samples.(2)A clustering method of social e-commerce users based on K-means++algorithm is proposed.This method is based on the research of traditional e-commerce user classification methods,and according to the characteristics of social e-commerce users,through improved data preprocessing and parameter tuning methods,the classification of social e-commerce business data and software usage data is realized,and ordinary users and active users of the software can be screened out.Experiments are conducted on actual data of a social e-commerce user.The experimental results show that there are significant differences in the retention rate of users of various classes.The proposed method can accurately classify social e-commerce users.(3)A method for dividing user attributes in a single dimension is proposed.With the help of polynomial curve fitting method,the user attribute division is modeled as a polynomial function to find the turning point.A method to find the turning point of the curve based on the degree of local dispersion is proposed,and users of different levels are divided by finding the turning point of the polynomial curve.In order to obtain the optimal user division,an interval merging algorithm based on interval retention is proposed to determine the number of user division layers.Using Davies-Bouldin Index.Silhouette Coefficient and Calinski-Harabaz Index and other indicators for comparative analysis,the experimental results show that the proposed method is better than the K-means++algorithm.(4)A method based on the NLP classification model is proposed.This method analyzes the social content data of social e-commerce and establishes a deep learning model based on BERT to achieve accurate classification of the commercial attributes of social e-commerce.The experimental results show that the measured accuracy of the model exceeds 90%.The NLP classification model is actually deployed based on the TensorFlow framework and edge computing.Experimental data shows that the accuracy of the analytical running model is basically the same as the model deployed on the server side.A digital integrated service platform for Wechat Business is designed and developed,integrating functions such as data collection,transmission and big data analysis.Based on virtualization and RPA process automation technology,the classification model we proposed has been practically applied,which has improved the marketing efficiency and accuracy of mobile social e-commerce.