[Paper Review] A Survey on the Densest Subgraph Problem and Its Variants

Tommaso Lanciano, Atsushi Miyauchi+2 more|arXiv (Cornell University)|Mar 25, 2023

Advanced Graph Theory Research199 references5 citations

TL;DR

This survey provides a comprehensive overview of the Densest Subgraph Problem (DSP) and its many variants, covering exact and approximation algorithms, constrained and generalized formulations, and applications across diverse graph types and computational settings. It highlights recent breakthroughs, including improved approximation ratios and convergence guarantees for iterative peeling methods, and identifies key open problems in fairness, negatively weighted graphs, and metric space graphs.

ABSTRACT

The Densest Subgraph Problem requires to find, in a given graph, a subset of vertices whose induced subgraph maximizes a measure of density. The problem has received a great deal of attention in the algorithmic literature since the early 1970s, with many variants proposed and many applications built on top of this basic definition. Recent years have witnessed a revival of research interest in this problem with several important contributions, including some groundbreaking results, published in 2022 and 2023. This survey provides a deep overview of the fundamental results and an exhaustive coverage of the many variants proposed in the literature, with a special attention to the most recent results. The survey also presents a comprehensive overview of applications and discusses some interesting open problems for this evergreen research topic.

Motivation & Objective

• To provide a deep, exhaustive review of fundamental results and recent advances in the Densest Subgraph Problem (DSP) and its variants.
• To analyze exact and approximation algorithms for DSP, including connections to submodular function minimization and core decomposition.
• To investigate constrained variants (e.g., size, seed set, connectivity) and generalized objective functions (e.g., distance-h density, edge/vertex labeling).
• To explore DSP in richer graph models such as directed, temporal, uncertain, multilayer, and metric space graphs.
• To identify and discuss open problems, including convergence of iterative peeling, fairness, and approximation in negatively weighted graphs.

Proposed method

• Uses a systematic taxonomy to classify DSP variants based on constraints, objective function changes, and graph types.
• Reviews exact algorithms based on maximum flow (Goldberg) and submodular function minimization.
• Analyzes approximation algorithms, especially the greedy peeling and iterative peeling methods, with theoretical guarantees.
• Introduces and discusses the iterative peeling method's convergence as a key open problem.
• Covers computational settings such as streaming, distributed, and MapReduce for scalable DSP computation.
• Applies DSP to real-world applications including community detection, motif discovery, and fraud detection.

Experimental results

Research questions

• RQ1 What are the theoretical and algorithmic foundations of the Densest Subgraph Problem, and how do exact and approximation algorithms compare?
• RQ2 How do constrained variants—such as the Densest $k$-subgraph or seed-constrained subgraphs—impact solution quality and computational complexity?
• RQ3 What are the limitations and open challenges in approximating distance-$h$ dense subgraphs and negatively weighted graphs?
• RQ4 How can fairness be integrated into dense subgraph mining, and what are the current gaps in this emerging area?
• RQ5 What are the key open problems in metric space graphs and the convergence of iterative peeling algorithms?

Key findings

• The iterative peeling method for DSP has shown strong empirical performance but lacks a proven convergence guarantee, representing a major open problem.
• Recent work has improved the approximation ratio for the Densest $k$-subgraph problem to $O(n^{1/4+\epsilon})$, though better bounds remain elusive.
• The densest subgraph with a seed set constraint has a 2-approximation, but improving this ratio is an open challenge.
• For distance-$h$ dense subgraphs, current approximation relies on generalized core decomposition, which is a first step but not yet optimal.
• In negatively weighted graphs, the best-known approximation ratio depends on the input, indicating a need for improved bounds in this important setting.
• Fairness in dense subgraph mining is underexplored, with only one existing work addressing the issue, highlighting a critical research gap.