Nepotistic relationships in Twitter and their impact on rank prestige algorithms
Compared 5 "prestige algorithms": PageRank, HITS, NodeRanking, TunkRank, TwitterRank on a large dataset: 28 million English tweets from 5 million users. TunkRank (description) (implementation) (API) (slides about), which discounts reciprocal follows, is best.
Theoretical and practical relevance:
Spam is an ongoing problem in social networks. Evaluating algorithms for filtering it is useful! Presented at CERI2010 as "Overcoming Spammers in Twitter" http://slidesha.re/a2RKs8 This is a paper with both extensive related work/perspective on the literature and significant quantitative analysis -- a rare combination! Methodology for identifying abusive users in Twitter may be useful elsewhere.
This was published in an open access journal.