baogaotimu：effective network embedding for massive graphs
报告内容简介：Network embedding is a common task in graph analytics that maps the local graph structure of a node to a fixed-length vector, which can then be used in downstream machine learning operations. This talk focuses on network embedding for massive graphs, often with billions of edges. On this scale, most existing approaches fail, as they incur either prohibitively high costs, or severely compromised result utility. The solution covered in this talk, called Node-Reweighted PageRank, is based on a classic idea of deriving embedding vectors from pairwise personalized PageRank values. Specifically, the talk covers two main topics: first, how personalized PageRank can be computed efficiently for every node in a massive graph; second, why personalized PageRank in its original form is actuallynotsuitable for network embedding, and how we fix this issue through an iterative reweighting algorithm. The resulting solution obtains more accurate results than 18 existing methods on 7 real graphs, on 3 different graph analytics tasks.