Evaluating community structure in large network with random walks

TL;DR

This paper introduces a fast, nearly linear time algorithm based on random walks for evaluating and detecting community structures in large-scale networks, addressing the scalability issues of traditional methods.

Contribution

It presents a novel local community detection algorithm that operates efficiently on very large networks, offering a practical alternative to existing methods.

Findings

Effective in small benchmark networks with acceptable accuracy

Significantly faster than traditional algorithms on large networks

Provides a new community evaluation measure differing from Newman Modularity

Abstract

Community structure is one of the most important properties of networks. Most community algorithms are not suitable for large networks because of their time consuming. In fact there are lots of networks with millons even billons of nodes. In such case, most algorithms running in time O(n2logn) or even larger are not practical. What we need are linear or approximately linear time algorithm. Rising in response to such needs, we propose a quick methods to evaluate community structure in networks and then put forward a local community algorithm with nearly linear time based on random walks. Using our community evaluating measure, we could find some difference results from measures used before, i.e., the Newman Modularity. Our algorithm are effective in small benchmark networks with small less accuracy than more complex algorithms but a great of advantage in time consuming for large networks, especially super large networks.