Two papers on Online Community Analysis accepted at Social Computing Conference 2012

In the former work, Role Analysis was performed over data from the SAP Community Network, one of the partners of ROBUST, to examine the role compositions present in the platform's different communities. Such compositions describe the different proportions of behavioural roles that are present in different communities and allow community managers to interpret and understand what their community looks like. The authors defined four community health indicators: Churn Rate, i.e. the proportion of users who leave the community; User Count, the total number of active contributors to the community; Seeds to Non-seeds Proportion, the proportion of posts that get a reply to those that do not; and the Clustering Coefficient of the community's social network, gauging the social capital of the community. By using the role composition information related to individual communities the authors were able to predict the health indicators with high accuracy and detect changes in community health. Furthermore, by using logistic regression models the authors examined the relation between increases in the health of the communities and role compositions. Findings indicated that individual communities exhibit unique health patterns and that the use of a single, global pattern for predicting community health is limited.

The second paper, authored with Claudia Wagner and Markus Strohmaier from Joanneum Research, Graz, Austria and the Knowledge Management Institute and Know Center, Graz, Austria, respectively, also uncovered, in a similar vein, differences between communities, this time in terms of attention patterns. Attention patterns relate to the factors that are associated with heightened attention to content shared within online communities, where attention in this context is measured using the number of replies to a given thread starter post. The paper examined: a) what factors are associated with initiating a discussion, and how these patterns differ between communities; and b) what factors are associated with heightened attention levels, and again how these patterns differed. The work examined these differences across 10 randomly selected communities from Boards.ie - the largest Irish community message board. The findings demonstrate the divergence in attention patterns between communities: for instance, in supportive-communities an increase in the number of hyperlinks in a thread starter is likely to produce no attention, while for content-sharing communities an increase in the number of hyperlinks is more likely to yield attention. The subject-specificity of the community was also found to be an indicator of differing patterns: a post in a community that is focussed on a specific topic will generate attention only if it matches the topic of the community, otherwise it will be ignored.

Links

• SocialCom Conference: http://www.asesite.org/conferences/socialcom/2012/