Technologies Used
- Perl
- SQL
- GUESS
- Pajek
- SPSS
- Matlab
"The presentation is clear, informative and extremely interesting. I find your suggestions and conclusion very relevant and enjoyed the graphs along the way. This will definitively help us guide the future of this safety feature as the community continues to grow and evolve."
- Member of CouchSurfing Leadership Team
Home > Portfolio > CouchSurfing Network
Network Analysis of the CouchSurfing Online Community
Using network analysis techniques to evaluate the effectiveness of the 'vouching' trust mechanism on CouchSurfing.com.
Date: August 2008 - August 2009
Collaborators: Hung Truong, Tanuj Shah, Professor Lada Adamic
Client: CouchSurfing.com Leadership Team
Course: SI 508: Networks, Independent study with Lada Adamic
Recognition: Accepted to IEEE Social Com 2009 Conference, held August 29-31, 2009 in Vancouver.
My role: I did all the data munging work and most of the visualizations with GUESS.
The Challenge
The goal of this project was to use network analysis techniques to find out if the 'vouch' network of CouchSurfing.com - a crucial part of their reputation system - is working well at determining which members of the community are trustworthy. Specifically, we wanted to determine things like:
- Which of their friends are members vouching for?
- Do members who are vouched differ significantly from members who are not vouched?
- Are certain members (i.e., those from cities with many CouchSurfers) more likely to get vouched?
- Is the vouching system working well as a safety feature for CouchSurfing?
Process
For our semester-long project, my teammate and I studied a dataset of over 660,000 members of CouchSurfing and the friendship connections between them that was provided to us by CouchSurfing. We began by writing Perl scripts to create formatted data files. Then, we used standard network analysis techniques and tools including GUESS and iGraph to analyze the properties and structure of the network. We chose to narrow our focus to only a subset of the network, the members from France, because it made visualization easier. We created many visualizations in GUESS to show different aspects of the France vouch network.
Result
We found that the friendship network in France is largely geographically-based, as seen in the image above. This image was created by running a community-finding algorithm which found that there were 4 distinct 'communities' in France. We further found that many of the vouches are between friends with lesser friendship levels (CouchSurfing has a way for users to publicly rate the strength of each of their friendships), which was worrying, as vouches should only be between those who know each other well. Another finding was many members already have many vouches. Therefore, it becomes important to display not only the total number of vouches one has, but perhaps also an overall reputation score that takes into account how "high-quality" the vouches are that one has, based on the underlying social network.
We created a report for the CouchSurfing Leadership Team outlining our recommendations, which was well-received (see quote in sidebar).