Date of Award

Summer 2014

Degree Type

Thesis

Degree Name

Master of Science (MS)

Department

Communication

First Advisor

Seungyoon Lee

Committee Member 1

Lorraine Kisselburgh

Committee Member 2

Sorin Matei

Abstract

Web-based peer production communities, like Wikipedia and open source software, have created digital artifacts of growing cultural, financial, and technological importance. Understanding how and why people choose to join these communities, and why they eventually leave them, is therefore an important topic.

We take all of the edit data from six years of activity on the online genealogy wiki WeRelate, and create monthly snapshots of behavior and interaction networks for all 9,570 users who edited the site. We use machine learning to cluster these behavioral snapshots into four "behavioral roles". We identify one of these roles as being indicative of a community of practice, and we investigate how users move from role to role. As in many other online, peer production projects, the vast majority of users are only active for a short time, and contribute very little while a small number of users contribute a great deal.

Figuring out how to recruit and encourage these users is very important to the success of peer production projects. We use visualizations, regression analysis, and stochastic actor-oriented modeling of four different types of interaction networks to study whether these very active users represent a community of practice that new users can learn from and join. We also study how people leave the community, and whether there are signals that someone is starting to disengage.

We do not find much evidence that these users go through a period of legitimate peripheral participation or acculturation. Rather, those who will become core members show behavior that is similar to long-term core members from their first few months on the site. We find that these core members show a clear trend of disengaging from the community over a few months before leaving completely, indicating a period where intervention may be effective. We also find a potentially effective intervention, as those who are actively interacting with others who are core members are less likely to disengage.

Our findings provide implications for understanding how online communities function, how interaction networks influence user activity, and how those who are members of these communities might make them more effective. The study also provides a new methodological framework for studying the influence of communicative interactions in online communities.

Share

COinS