Agent-Based Modeling to Inform Online Community Design: Impact of Topical Breadth, Message Volume, and Discussion Moderation on Member Commitment and Contribution

AbstractThe design of complex social systems, such as online communities, requires the consideration of many parameters, a practice at odds with social science research that focuses on the effects of a small set of variables. In this article, we show how synthesizing insights from multiple, narrowly focused social science theories in an agent-based model helps us understand factors that lead to the success of online communities. The agent-based model combines insights from theories related to collective effort, information overload, social identity, and interpersonal attraction to predict motivations for online community participation. We conducted virtual experiments to develop hypotheses around three design decisions about how to orchestrate an online community—topical breadth, message volume, and discussion moderation—and the trade-offs involved in making these decisions. The simulation experiments suggest that broad topics and high message volume can lead to higher member commitment. Personalized moderation outperforms other types of moderation in increasing members' commitment and contribution, especially in topically broad communities and those with high message volume. In comparison, community-level moderation increases commitment but not contribution, and only in topically narrow communities. These simulation results suggest a critical trade-off between informational and relational benefits. This research illustrates that there are many interactions among the design decisions that are important to consider; the particulars of the community's goals often determine the effectiveness of some decisions. It also demonstrates the value of agent-based modeling in synthesizing simple social science theories to describe and prescribe behaviors in a complex system, generating novel insights that inform the design of online communities. NOTESNotes1 An online community can be created on various technological platforms (e.g., listservers, Usenet news, chats, bulletin boards, web forums, and social networking sites) around various purposes (e.g., interest, health support, technical support, education, e-commerce; CitationPreece 2000). In this article, we focus on conversation-based interest communities such as newsgroups or web forums created to host online discussion of shared interests.2 We examine personalized moderation at the conceptual level. Its implementation is beyond the scope of this article.3 Note that in this article we simulate behaviors in an interest community, like a movie discussion group, and do not vary community type. Doing so would involve varying these weights. For example, the relative weights in a technical support group, in which people typically care less about interpersonal bonds and more about information, identity, and reputation, the weights may be 0.5, 0.25, 0.1, and 0.15, respectively. In contrast, the weights for a cancer support group, where one's disease helps defines one's identity, may be 0.33, 0.33, 0.33, and 0.1.4 The Usenet groups in the data set had no moderation, which is the condition we used for model validation.5 Because previous analyses revealed no significant difference between medium and broad topical breadth and a linear effect of message volume, we omitted medium topical breadth in Figure 11 and medium message volume in Figure 12 to make the figures more readable.Acknowledgments. Earlier versions of this article have been presented at the research colloquiums of Carnegie Mellon University, McGill University, University of Maryland, and University of Minnesota and workshops at Academy of Management Annual Meeting and ACM SIGCHI Conference on Human Factors in Computing Systems. We thank our participants for their valuable feedback. We also thank Shawn Curley for his helpful comments and Sam Hashemi for research assistance.Funding. This work was supported by Carlson School of Management Dean's Small Research Grant, National Science Foundation Grant IIS-0325049 (Designing Online Communities to Enhance Participation), IIS-0729286 (Solving Critical Problems in Online Groups), and IIS-0808692 (Understanding Online Volunteer Communities). The views and conclusions contained in this document are those of the authors and should not be interpreted as representing the official policies of the National Science Foundation.Supplemental Data. Supplemental data are available for this article. RenKraut-2013-SimulateOnlineCommunity.nlogo is the simulation model and RenKraut-2013-SupplementaryMaterials.pdf provides instructions for downloading the simulation platform, running the model, and interpreting the results. Free access to these materials is available at the online edition of this article on the Human–Computer Interaction publisher's website (www.tandfonline.com/hhci).HCI Editorial Record. First manuscript received August 22, 2012. Revision received June 26, 2013. Accepted by Peter Pirolli. Final manuscript received June 26, 2013. — EditorAdditional informationNotes on contributorsYuqing RenYuqing Ren (chingren@umn.edu, www.tc.umn.edu/~chingren) is a social scientist with an interest in online community design, social media, distributed collaboration, knowledge management, and computational modeling of social and organizational systems; she is an Assistant Professor of Information and Decision Sciences at the Carlson School of Management at the University of Minnesota.Robert E. KrautRobert E. Kraut (robert.kraut@cmu.edu, kraut.hciresearch.org) is a social psychologist with an interest in the design and social impact of computing, analysis, and design of online communities, everyday use of the Internet, technology and conversation, collaboration in small work groups, and computing in organizations; he is the Herbert A. Simon Professor in the School of Computer Science and the Tepper School of Business at Carnegie Mellon University.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.