PROJECT TITLE:

A Ranking Approach on Large-Scale Graph With Multidimensional Heterogeneous Information - 2015

ABSTRACT:

Graph-primarily based ranking has been extensively studied and often applied in many applications, like webpage ranking. It aims at mining probably valuable data from the raw graph-structured information. Recently, with the proliferation of made heterogeneous information (e.g., node/edge features and previous information) offered in several real-world graphs, a way to effectively and efficiently leverage all data to boost the ranking performance becomes a brand new difficult downside. Previous methods solely utilize half of such info and attempt to rank graph nodes consistent with link-based strategies, of which the ranking performances are severely stricken by several well-known issues, e.g., over-fitting or high computational complexity, particularly when the scale of graph is terribly large. In this paper, we have a tendency to address the massive-scale graph-primarily based ranking drawback and specialize in how to effectively exploit made heterogeneous info of the graph to boost the ranking performance. Specifically, we propose an innovative and effective semi-supervised PageRank (SSP) approach to parameterize the derived information within a unified semi-supervised learning framework (SSLF-GR), then simultaneously optimize the parameters and also the ranking several graph nodes. Experiments on the real-world large-scale graphs demonstrate that our methodology significantly outperforms the algorithms that consider such graph data solely partially.