Faster Local Motif Clustering via Maximum Flows

TL;DR

This paper introduces SOCIAL, a fast and effective local motif clustering algorithm that leverages maximum flow techniques to optimize motif conductance, outperforming existing methods in both quality and speed.

Contribution

The paper presents SOCIAL, a novel algorithm that improves local motif clustering by integrating a local hypergraph model with max-flow optimization, achieving superior performance.

Findings

SOCIAL yields lower motif conductance than state-of-the-art methods.

The algorithm is significantly faster, up to multiple orders of magnitude.

Experimental results confirm the effectiveness of SOCIAL with the triangle motif.

Abstract

Local clustering aims to identify a cluster within a given graph that includes a designated seed node or a significant portion of a group of seed nodes. This cluster should be well-characterized, i.e., it has a high number of internal edges and a low number of external edges. In this work, we propose SOCIAL, a novel algorithm for local motif clustering which optimizes for motif conductance based on a local hypergraph model representation of the problem and an adapted version of the max-flow quotient-cut improvement algorithm (MQI). In our experiments with the triangle motif, SOCIAL produces local clusters with an average motif conductance lower than the state-of-the-art, while being up to multiple orders of magnitude faster.