Dynamics and diversity of online community activities

Web sites where users create and rate content as well as form networks with other users display long-tailed distributions of user activity. Using data from one such community site, Essembly, we propose and evaluate mechanisms for these distributions that rely only on information actually available to users. For Essembly, we describe the scale-free degree distribution of the social network as a result of users’ shared interests, manifested by their content rating activity. We find the long tails in network properties arise from user activity rates that are broadly distributed, as well as the extensive variability in the time users devote to the site. Introduction. – Participatory web sites facilitate their users creating, rating, and sharing content. Examples include Digg[.com] for news stories, Flickr[.com] for photos, and Wikipedia[.org] for encyclopedia articles. To aid users in finding content, many such sites employ collaborative filtering [1] to allow users to specify links to other users whose content or ratings are particularly relevant. These links can involve either people who already know each other, or people who discover their common interests through participating in the web site. In addition to helping identify relevant content, the resulting networks enable users to find others with similar interests and establish trust in recommendations [2]. The availability of activity records from these sites has led to numerous studies of user behavior and the networks they create. Observed commonalities in these systems suggest general generative processes leading to these observations. Examples include preferential attachment in forming networks and multiplicative processes leading to wide variation in user activity. Moreover, observed behavior can arise from a variety of mechanisms [3, 4]. Identifying information readily available to users on a participatory web site can suggest plausible causal mechanisms. The simplest such approach considers average behavior of users on a site [5]. Such models can indicate how system behavior relates to the average decisions of many users. By design, such models do not address a prominent aspect of observed online networks: the long tails in their distributions of links and activity. Models including this diversity could be useful to improve effectiveness of the web sites by allowing to focus on significantly active users or especially interesting content, and enhancing user experience by leveraging the long tail in niche demand [6]. A key question with respect to the observed diversity is whether users and the networks are reasonably viewed as behaviors arising from a statistically homogeneous population, and hence well-characterized by a mean and variance. Or is diversity of intrinsic characteristics among participants the dominant cause of the observed wide variation in behaviors? Moreover, to the extent user diversity is important, what is a minimal characterization of this user variation sufficient to produce the observed long-tail distributions? This paper considers these questions in the context of a politically-oriented web community, Essembly. Unlike most such sites, Essembly provides multiple networks with differing nominal semantics, which is useful for distinguishing among some models. We consider plausible mechanisms users could be following to produce the observed long-tail behaviors both in their online activities and network characteristics. In the remainder of this paper, we first describe Essembly and our data set, then separately examine the network formation of users and then their highly variable behaviors, respectively. We suggest models to describe the observed characteristics of the network and the users. Essembly. – Essembly is an online service helping users engage in political discussion through creating and voting on resolves reflecting controversial issues. User Essembly LLC at http://www.essembly.com