Golden questions of the day

At systems seminar today, we had a discussion about this Eurosys 2009 paper: User Interactions in Social Networks and their Implications
Christo Wilson (UCSB), Bryce Boe (UCSB), Alessandra Sala (UCSB), Krishna Puttaswamy (UCSB), Ben Y. Zhao (UCSB)


Some thoughts and questions:

#1 Can we extract something useful from Facebook wall posts? Data mining? What do people talk about? See Google's flu paper (Nature).

#2 Why study user interactions at all? To look for invariant trends like Dunbar's number? Why is this important in systems research?

#3 There seems a clear distinction between "knowledge" and "interaction" when it comes to measuring tie strength, e.g., "I know him well, but haven't talked for a while". Knowledge part is not pronounced in OSNs. Can we infer tie strength or the level of trust without doing user studies?

#4 Why SybilGuard performs poorly in interaction graphs? It is not intuitive why SybilGuard should work better in interaction graph. In SybilGuard, information about social graph might be enough, because it only matters that you know the other friend is a real user, not a fake user. Fax-mixing properties SybilGuard exploits do not align well with a community structure that is embedded in OSNs.

#5 Links need to be tagged with a purpose. Simply knowing the social graph or the interaction graph may not be enough for many of the socially-enhanced applications.

#6 Low rate conversation made in OSNs could sum up to be quite valuable. See Google's flu paper.

#7 Need to be careful in drawing conclusions. Results could be extrapolated. Extra care needed in methodology.

#8 Related to "trust" in social ties.
Here is a link to the CHI2009 paper on measuring "tie strength" based on information available in online social networks (e.g., number of messages exchanged, word count, education level, mutual friends). Authors do surveys to get ground truth data and some of the survey questions are highly related to trust (e.g., would you lend this friend $100?).