Formulation
PR(v) = (1-d) · r_v + d · sum_{u→v} PR(u)/L(u). Where r_v = preference for v (nonzero on 'seed' set).
Advertisement
Random walk with restart
Alternative name. Walk from seed set with probability (1-d) of restarting.
Advertisement
Locality
Push algorithm: local computation focused on high-PR region near seeds. Sub-linear time for approximate answers.
Applications
Twitter's Who to Follow. Pinterest recommendations. Community detection. Knowledge graph queries.