Using Cassovary for "Who to Follow"?

[Afton Geil]

My research group has recently implemented Twitter's Who to Follow recommendation algorithm (http://stanford.edu/~rezab/papers/wtf_overview.pdf) on the GPU. We would like to measure the run time of this algorithm on a CPU system, and we are considering using the Cassovary library to do this. Since Cassovary used to be the framework Twitter used for this calculation, it seems like a good place to start.

The algorithm is essentially three steps: a Personalized PageRank calculation, then the creation of a bipartite graph based on this ranking, followed by another ranking calculation similar to SALSA (http://en.wikipedia.org/wiki/SALSA_algorithm) or HITS (http://en.wikipedia.org/wiki/HITS_algorithm).

If anyone has advice on how to do this with Cassovary, or knows if there are implementations of PageRank/SALSA/HITS using Cassovary available anywhere, we would really appreciate it. Thanks!

Afton

[Pankaj Gupta]

Hi Afton,

Yes, Cassovary has a personalized pagerank implementation as well as a Salsa implementation. 

PPR:

https://github.com/twitter/cassovary/blob/master/cassovary-core/src/main/scala/com/twitter/cassovary/graph/GraphUtils.scala#L136

Salsa:

https://github.com/twitter/cassovary/blob/master/cassovary-core/src/main/scala/com/twitter/cassovary/graph/bipartite/IterativeLinkAnalyzer.scala

--
You received this message because you are subscribed to the Google Groups "Cassovary" group.
To unsubscribe from this group and stop receiving emails from it, send an email to twitter-cassov...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.