人类行为一直是社会科学的重点研究对象,而人类个体之间的行为如何影响人类群体 的行为是人类行为研究的一个重要问题。随着信息技术的发展,越来越多的人们通过网络 进行交流和获取信息,而在线社区是其中重要的一种网络媒介。网络论坛集中了大量人们 的日常行为,并且可以通过一定的手段将这些行为记录下来。这也为我们研究人类群体行 为模式提供了便利。科学定量的分析人们在论坛上的行为模式在应用上也有非常重要意义。 一方面,对人们在论坛上的行为模式研究可以帮助优化网络资源配置,有十分重要的经济 意义;另一方面,这也有助于改进网络服务技术,使网络媒体更好的服务人们的信息交流 和沟通。通过对 tianya 网络社区论坛的大量实证数据分析,我们发现在论坛上人们群体行为涌 现出了很强的复杂性。在用户行为的时间特征上,我们发现论坛当中所有帖子的发生时间 间隔服从幂律分布。这意味着论坛当中人们行为有高度的集聚性,有时人们会一起大量的 发表新话题和回复,而有时候人们会长时间不参与任何讨论。另一方面论坛中不同的话题 之间也表现了很大的差异性,所有版面的话题收到回复的数目同样服从幂律分布。这表明 有些的话题会吸引人们进行非常热烈的讨论,而其他大部分话题仅仅只能得到少量的回复。为了解释用户行为在网络论坛上体现出的幂律特征,我们构建了一个用户─论坛相互 影响的动力学模型。我们假设论坛当中出现新帖子的几率由两方面因素决定:一是人们在 一定时间内观察到的新帖子的数目;另一个是人们在日常生活中积累的可以用于讨论的话 题的数目。通过数值模拟我们可以利用该模型重现 tianya 论坛上八个不同版面的实证结果。 这也表明我们的模型解释了论坛用户行为模式的产生机制。此外,我们发现模型模拟的最优参数具有非常直观的实际意义。通过对拟合参数和实 证数据的分析我们发现,不同讨论区当中的用户行为模式有很大的差别。话题比较具有专 业性的版面当中用户倾向于积极参与讨论,而在话题非常宽泛的版面当中用户更倾向与浏 览版面内容。

On-line forum is the most important mass media in modern society. The scientific attention to human behaviors on forums is motivated by clear economic and technological purposes since the possibility to mathematically describe human behavior may have important implications in resource allocation and service management.We focus on timing of posts and structures of different threads which separately describe temporal and organization characteristic of human activities on the web. The main source of data is from forums of Tianya Community and captured using web crawler program. Our results reveal that complexities emerge from human behaviors on forums: both the temporal intervals between successive posts and the number of replies to different topic threads on forums surprisingly deviate from Poisson distribution and show obvious power-law characteristic. Further more, time series of posting have short memory, which indicates auto-correlation of human behaviors on internet. The unevenness of human activities on the web possibly suggests that herd instinct and divergent interest of human nature contribute equally to opinion formation and information spreading.To illustrate empirical findings, we propose a dynamic model based on people-forum interacting. In this model, we assume that the probability there comes a new post at time t is determined by two factors: the amount of new posts observed by forum users and the potential willingness of forum users to post. The potential posting willingness of web users increases at constant speed. When there comes a new post, the amount of new posts observed by forum users increases and their potential willingness to post decreases. Results of numerical simulation using Matlab reveal that the amount of people who focus on different themes and the level of their participation in discussions are key factors to determine the distributions of successive temporal intervals.