Network Analysis of CampusRoost

Network Analysis of CampusRoost

Is it possible to create a social network that allows neighbors to connect with each other? Will geography determine community, or do people connect regardless of where they live? We conducted a network analysis of CampusRoost, a social system created to help students connect with others in their neighborhood, to answer these questions.

The Facts

What: A network analysis of CampusRoost, a social network created by students at The University of Michigan
When: Fall 2008
Who: Myself and Gaurav Bhatnagar
Where: SI508 – Networks: Theory and Application with Lada Adamic

The Story

Gaurav and I were given the opportunity to examine the data from a budding social network called CampusRoost. This is a network created by computer science students at The University of Michigan with the goal of better connecting student communities in neighborhoods around campus. The system centers around the idea of “chirps” which are broadcasts metaphorically analogous to standing on one’s porch with a megaphone and yelling your message. The goal was to encourage neighbors to gather for social events like parties, going out, study sessions, and playing games. The system also included aspects of traditional social networks, such as establishing friend connections and writing on walls.

We began our project by extracting the data from the CampusRoost database using MySQL and Perl. We loaded the graph data in Pajek to begin our network analysis, calculating such network factors as centrality by degree, closeness, and betweenness as well as average shortest path and clustering coefficient for comparison to an equivalent, random network. From this, we got a sense of the architecture of the network, including the range of influence that nodes have. We then used FANMOD to perform a motif analysis and establish common patterns within the network. By performing additional data analysis, we were able to incorporate geographic data in our network. From this, we used the R statistical package to look at correlations between the physical distance between houses and network metrics like edge weight and betweenness. Finally, we used GUESS both to visualize the network and run the Girvan–Newman algorithm for community detection. One of the most interesting and informative visualizations we created is of the network communities overlayed on top of a geographic map, which demonstrated the relationship (or lack there of) between physical closeness and community.

Based on this analysis, we were able to conclusively determine how well CampusRoost was accomplishing it’s goal of connecting neighbors and strengthening communities. Combining these insights with our interaction design sensibilities led us to make several suggestions to the creators of CampusRoost on how they might improve the strength and health of their online community.

My Role

Gaurav and I worked closely throughout the duration of this project; however, we each had our unique strengths that we brought to the project. In my case, my technical background allowed me to focus more on the data extraction and analysis using Perl as well as the creation of more complicated analysis and visualization. Gaurav had done previous research on this system, including interviewing users, so he was able to represent the voice of the user when interpreting our results. Together, we shared responsibility in crafting recommendations for the CampusRoost team, and we both learned a lot about the various ways of quantifying network metrics and characterizing the nature of an online community.

Artifacts