IBM computer scientist and University of Illinois colleagues take 2012 best paper award for presentation at BigMine-2012 in association with KDD’2012.
Cao, who works in the area of large-scale visual recognition, received the award for "Delta-SimRank Compuing on MapReduce." He and his co-authors presented the paper at the 1st International Workshop on Big Data, Streams and Heterogeneous Source Mining: Algorithms, Systems, Programming Models and Applications (BigMine-12) on August 12, 2012 in Beijing.
SimRank is a measure asserting that two objects are similar if they are related to similar objects. It has been used for many applications in social networks, including citation networks and student course networks.
In the paper, Cao et al argue that SimRank is expensive to use in at least two senses:
(1) Time complexity: It can take 46 hours to compute the SimRank measures in a synthetic network with 10K nodes on a single machine.
(2) Size of network: A huge amount of memory is necessary to compute SimRank -- a feat beyond the ability of a single computer.
The authors show that "the use of a distributed system makes it possible to compute SimRank in large networks." Their key contribution lies in a new algorithm that computes the change in SimRank scores instead of the original SimRank scores.
The BigMine 2012 award represents Liangliang Cao's first best paper award since coming to IBM Research in June 2011.
Download and read "Delta-SimRank Computing on MapReduce." (pdf)
Last updated on September 10, 2012