Robert Axelrod; William D. Hamilton http://citeseer.ist.psu.edu/viewdoc/download?doi=10.1.1.147.9644&rep=rep1&type=pdf
Great paper on Iterated Prisoners Delimma. Many citations of biological research. Shows that Tit4Tat is dominant strategy and gives parameters in which t4t players become dominant agaisnt allD.
Page, Brin http://citeseer.ist.psu.edu/viewdoc/download?doi=10.1.1.31.1768&rep=rep1&type=pdf
Google PageRank paper. The thing I really like about this paper, is how it's model is a just a mathmatical abstraction of a simple human activity. The page rank is the probability that a random web sufer would land on that page. The model is very easy to graps intuitively. Search and spam is very closly related to the sybil problem. indeed most papers on sybil attacks use some sort of graph theory, though I am yet to see a sybil attack paper as good as this.
Lik Mui, Mojdeh Mohtashemi http://citeseer.ist.psu.edu/viewdoc/download?doi=10.1.1.116.6001&rep=rep1&type=pdf
Probably most useful as a collection of citations to works in computer science, psychology, economics. mentions game theory (and Evolution of Cooperation) and proposes a statistical trust model, but dosesn't demonstrate it in a simulation. Also, this paper proposes a way of calculating reputation across trust relations but again doesn't demonstrate any of it's implications.
Wenyu Wang, Li Zhao, Ruixi Yuan http://www.cecs.uci.edu/~papers/ipdps06/pdfs/1568976582-HOTP2P-paper-1.pdf
Gnutella era paper that outlines a incentive scheme for exchanging files. This seems obvious post-bittorrent. Instead of using a bittorrent like tracker there is just a mesh of nodes. They conduct simulations with various freeloaders and sybils. the "social networks" are networks of friendly computers, not humans.
Kamvar, Schlosser, Garcia-Molina http://nlp.stanford.edu/pubs/eigentrust.pdf
a distributed implementation of something similar to pagerank. They discuss a distributed implementaion where every node is responsible for calculating/reporting their own trust value, and then another method that is "secured" via DHT-hammer. They admit that there system is vunerable to a sybil attack, and suggest using captchas (!) (they don't cite anything for secure decentralized capchas...) Also, they do not get a statistical deviation on the feedback, so a new peer looks like one with balance of good and bad feedback. I think this system would not work in practice.
Aditya Kurve and George Kesidis http://citeseer.ist.psu.edu/viewdoc/download?doi=10.1.1.220.8236&rep=rep1&type=pdf
Detect sybils clusters by searching for weakly attached portions of the total graph. I'm not convinced that attackers could not interfer with the calculation, which involves electing trusted leaders "police nodes". I'm not satisified there is a strong model of what "sybil" means.
#TODO
These look promising but I havn't read them yet
Audun Josang
Patrick Roos, Michele Gelfand, Dana Nau, Ryan Carr
http://www.cs.umd.edu/~nau/papers/roos2014high.pdf
Stefania Ottone http://polis.unipmn.it/pubbl/RePEc/uca/ucapdv/ottone57.pdf