<div>Computation of node proximity on graphs is a fundamental graph mining problem, which is widely explored in recommendation, web search and community discovery applications. There are basically two kinds of methods of measuring node proximity. One is content-based method, which measures nodes’ content similarity such as text similarity and set overlap. The other is structure-based method, which purely relies on graph structures to measure node proximity, such as PageRank, SimRank and PageSim, In this talk, I will introduce one commonly used structure-based proximity measure, Personalized PageRank, and detail our recent algorithms for efficiently computing single-source Personalized PageRank. Our algorithms do not need to pre-process raw graphs and significantly improve query efficiency, space cost and scalability over existing algorithms.</div>