Centaurs - a Component Based Framework to Mine Large Graphs

Abstract

The increase of the amount of data represented as a graph, like
complex networks, motivated the creation of a new research area called graph mining.
This work proposes a new framework based on components, called Centaurs, to mine data represented as a graph. The main idea of Centaurs is to couple community detection and link prediction algorithms to mine missing edges that were missed during the graph building process.
Graph preprocessing and storage algorithms are also explored in this proposal, given that large graphs cannot always be storage in main memory only.
The main Centaurs's case study is the Read the Web project that aims to build a graph to represent knowledge extract from the Web based on a never ending learning algorithm.