电子商务中人类活动的标度行为实证研究

Empirical study on scaling of human behaviors in e-commerce

应用去趋势波动分析法,对电子商务中人类网上购物行为进行研究,首次探讨了人类浏览及购买行为时间序列(数量波动)标度律.首先,研究发现人类网上购物行为呈现出明显的周期性,其时间序列的概率密度函数具有显著的双模态特征.其次,利用傅里叶变换方法分析浏览以及购买行为时间序列的功率谱,发现其演化过程不同于无关联的泊松过程.最后,基于功率谱过滤周期性趋势的影响,对去除周期趋势后的浏览和购买行为时间序列进行去趋势波动分析,发现其标度行为表明其具有较强的长程关联特性,且平均标度值近似为1,表明其具有自组织临界性.实证研究结果与其他领域如因特网交通流和金融市场价格波动的标度行为相似,有助于理解人类活动如何影响电子商务系统演化和提高在线商务活动效率,对分析电子商务中人类行为活动的机制和预测其波动趋势具有重要的启示作用.

The detrended fluctuation analysis （DFA） method is used to analyze the human on-line activities in e-commerce. We compre- hensively investigate the scale laws of browse and purchase behaviors, which have received little attention before. The time series of browse and purchase behaviors each obviously show a periodical character, and their probability density distributions each have a significant bimodal form. Based on the Fourier transform method, the power spectra of time series indicate that each of them obeys a stochastic process with a long^range self-similar feature （i.e., deviation far from the Poisson process）. After identifying and filtering the influence of periodic trend based on power spectra, the detrended fluctuation analysis is used to study the scaling law of time series. Several interesting results can be found that their scaling behaviors on small and large scales show similar values that confirm the long-range correlations rooting in the time series of human on-line activities, and their average scaling exponent approximately equaling 1 suggests that the human online activity may be associated with a self-organized criticality. Although the empirical results are only the observed phenomena like those found in the Internet traffic and stock price fluctuation of financial market, we still think that they may provide an important insight to deeply understand the mechanism of human dynamic behaviors in e-commerce and predict their fluctuation trend for the potential business application.