Analysis and modeling of online service user behavior trajectories using the RETENTIONEERING platform

Abstract

The active development and widespread distribution of online stores and sales sites sets the task for marketers and IT specialists to analyze the results of sites and customer behavior to maximize store profits and predict the development of online sales. The article provides an analysis of the purchasing activity of Internet users of various categories in various periods, the postulates of online trading are formulated. To comprehensively analyze the impact of a variety of factors, predict the demand for goods, form additional recommendations and special support for purchases, the authors offer an integrated approach based on taking into account traditional components, and using the latest tools and software products for analyzing the behavior of online buyers. As a system of analysis, the Retentioneering platform is considered as the most representative and meets the modern requirements for business intelligence systems. The use of platform tools simplifies the processing and analysis of event flows, user behavior trajectories, user classification, allows you to create logical connections and functions for machine learning when predicting a user's category and behavior, as well as the likelihood of a target event - making a purchase based on previously collected data on user behavior. Based on a set of statistics on the behavior of online store buyers, the article discusses such tools for analyzing the Retentioneering platform as an interactive graph of visualization of the behavior trajectory, the matrix of steps and transitions, conversion funnels, clustering the behavior of vectoring user trajectories. Clustering methods use multivariate space convolution algorithms. The UMAP and t-SNE algorithms are considered as dimensionality reduction methods. The main stages and formulas of implementation of convolution algorithms are given, their advantages and disadvantages are considered. These algorithms simplify the process of finding global minima, and improve the quality of rendering. The described algorithms and methods allow you to analyze the behavior of visitors to the online store, combine users into clusters with a similar behavior strategy according to various target features, identify the most pressing problems and bottlenecks of the network platform.