The evolution of online ideological communities

Abstract

Online communities are virtual spaces for users to share interests, support others, and to exchange knowledge and information. Understanding user behavior is valuable to organizations and has applications from marketing to security, for instance, identifying leaders within a community or predicting future behavior. In the present research, we seek to understand the various roles that users adopt in online communities–for instance, who leads the conversation? Who are the supporters? We examine user role changes over time and the pathways that users follow. This allows us to explore the differences between users who progress to leadership positions and users who fail to develop influence. We also reflect on how user role proportions impact the overall health of the community. Here, we examine two online ideological communities, RevLeft and Islamic Awakening (N = 1631; N = 849), and provide a novel approach to identify various types of users. Finally, we study user role trajectories over time and identify community “leaders” from meta-data alone. Study One examined both communities using K-MEANS cluster analysis of behavioral meta-data, which revealed seven user roles. We then mapped these roles against Preece and Schneiderman’s (2009) Reader-to-Leader Framework (RtLF). Both communities aligned with the RtLF, where most users were “contributors”, many were “collaborators”, and few were “leaders”. Study Two looked at one community over a two-year period and found that, despite a high churn rate of users, roles were stable over time. We built a model of user role transitions over the two years. This can be used to predict user role changes in the future, which will have implications for community managers and security focused contexts (e.g., analyzing behavioral meta-data from forums and websites known to be associated with illicit activity).

Assessing the Severity of Health States based on Social Media Posts

“…Literature shows increased attention on the OHCs for computationally discovering patient health [21]. A majority of these studies follows a qualitative approach based on the manual categorization of posts by the domain experts.…”

The Personality Panorama: Conceptualizing Personality through Big Behavioural Data

“…Computational and social sciences are both deeply embedded within rich ethical traditions—traditions that are mutually informative. Drew (2016), for example, provides a brilliant overview of the ethical framework that grew from the Government Data Science Partnership, drawing from insights gained during deliberative public workshops and surveys. In her overview, Drew details principles such as data security, accountability, and articulation of public need/benefit—principles that are highly familiar to psychologists, but whose application to big data expands beyond traditional ethical research guidelines (Manzo & Brightbill, 2007; Sugiura, Wiles, & Pope, 2017).…”

“…In the world of big data, however, we find the explicit recognition that personality does not exist in a vacuum. In the digital realm, people do not merely behave, they interact with dynamic environmental factors that can also be measured-other people (Davidson, Jones, Joinson, & Hinds, 2019;Pan, Altshuler, & Pentland, 2012), games (Canossa, Badler, El-Nasr, Tignor, & Colvin, 2015), web pages (Shobeiri, Laroche, & Mazaheri, 2013;Turkyilmaz, Erdem, & Uslu, 2015), smartphone apps (Miller, 2012), and so on. Social context parameters such as network size, centrality (position or role), and local transitivity (connectedness among friends) are associated with personality (Fang et al, 2015;Gosling, Augustine, Vazire, Holtzman, & Gaddis, 2011;Staiano et al, 2012), and traits such as cooperation are at least in part emergent properties of context, dependent upon community structure (Apicella, Marlowe, Fowler, & Christakis, 2012).…”

The crossroads of digital phenotyping

“…The distinction here would be: can we reliably and accurately distinguish between a mood fluctuation and a relapse or change in symptoms? (This is an issue seen in other computational social science considering distinctions between long term behavior changes or merely a temporary fluctuation50 ). Hence, starting this research within the general population to formulate a number of reliable metrics for mood, behavior, health outcome prediction is undoubtedly a safer route, which will provide findings that other areas of health and medicine can draw and build upon.Delving deeper into the realities and practicalities of digital phenotyping, there is a question asto what data is needed and how this data will be gathered.…”

“…When obtaining data from a variety of sources, it is not uncommon to see research with >100 variables 4,55 being modelled to predict an outcome or mood severity, rather than focusing on few, but specific measures/metrics. By focusing on specific digital interactions and outcomes (e.g., behaviors, clinical outcomes of interest), this encourages 'behavioral analytics'26,50,56 , where we have underlying mechanisms or theories to understand why variable…”