Scaling behavior of online human activity

TL;DR

This study investigates the scaling behavior and complexity of online human activity in e-commerce platforms using advanced analysis techniques, revealing self-similarity, long-range correlations, and individual behavioral differences.

Contribution

It provides a comprehensive analysis of human activity patterns in e-commerce, highlighting differences across user groups and at the individual level with novel insights into their dynamic behaviors.

Findings

Rating behaviors exhibit self-similarity and long-range correlations.

Different user activity groups show varying scaling exponents and complexities.

Some users display long-range anticorrelations linked to bimodal interevent time distributions.

Abstract

The rapid development of Internet technology enables human explore the web and record the traces of online activities. From the analysis of these large-scale data sets (i.e. traces), we can get insights about dynamic behavior of human activity. In this letter, the scaling behavior and complexity of human activity in the e-commerce, such as music, book, and movie rating, are comprehensively investigated by using detrended fluctuation analysis technique and multiscale entropy method. Firstly, the interevent time series of rating behaviors of these three type medias show the similar scaling property with exponents ranging from 0.53 to 0.58, which implies that the collective behaviors of rating media follow a process embodying self-similarity and long-range correlation. Meanwhile, by dividing the users into three groups based their activities (i.e., rating per unit time), we find that the scaling exponents of interevent time series in three groups are different. Hence, these results suggest the stronger long-range correlations exist in these collective behaviors. Furthermore, their information complexities vary from three groups. To explain the differences of the collective behaviors restricted to three groups, we study the dynamic behavior of human activity at individual level, and find that the dynamic behaviors of a few users have extremely small scaling exponents associating with long-range anticorrelations. By comparing with the interevent time distributions of four representative users, we can find that the bimodal distributions may bring the extraordinary scaling behaviors. These results of analyzing the online human activity in the e-commerce may not only provide insights to understand its dynamic behaviors but also be applied to acquire the potential economic interest.