A Survey on Hypergraph Mining: Patterns, Tools, and Generators

Geon Lee,Fanchen Bu,Tina Eliassi-Rad,Kijung Shin
2024-01-17
Abstract:Hypergraphs are a natural and powerful choice for modeling group interactions in the real world, which are often referred to as higher-order networks. For example, when modeling collaboration networks, where collaborations can involve not just two but three or more people, employing hypergraphs allows us to explore beyond pairwise (dyadic) patterns and capture groupwise (polyadic) patterns. The mathematical complexity of hypergraphs offers both opportunities and challenges for learning and mining on hypergraphs, and hypergraph mining, which seeks to enhance our understanding of underlying systems through hypergraph modeling, gained increasing attention in research. Researchers have discovered various structural patterns in real-world hypergraphs, leading to the development of mining tools. Moreover, they have designed generators with the aim of reproducing and thereby shedding light on these patterns. In this survey, we provide a comprehensive overview of the current landscape of hypergraph mining, covering patterns, tools, and generators. We provide comprehensive taxonomies for them, and we also provide in-depth discussions to provide insights into future research on hypergraph mining.
Social and Information Networks,Databases,Physics and Society
What problem does this paper attempt to address?
The problem that this paper attempts to solve is: **How to enhance the understanding of group interactions in real - world systems through hypergraph modeling?** Specifically, the paper focuses on the research progress in the field of hypergraph mining, aiming to provide a comprehensive review covering patterns, tools, and generators in hypergraphs. The paper points out that hypergraphs are a natural and powerful choice for modeling group interactions in the real world, which are often referred to as higher - order networks. For example, in a cooperation network, cooperation may involve not only two people, but also three or more people. Using hypergraphs can go beyond pairwise (binary) patterns and capture group (multivariate) patterns. ### Main problems and motivations 1. **Modeling of group interactions**: - Hypergraphs can effectively model high - order interactions among multiple people or entities, which is difficult for traditional graph models to achieve. - For example, group interactions in scenarios such as cooperation networks, protein interactions, and co - purchase of goods can be modeled by hypergraphs. 2. **Challenges and opportunities in hypergraph mining**: - The mathematical complexity of hypergraphs brings challenges to learning and mining, but also provides opportunities. - Researchers have discovered various structural patterns in real - world hypergraphs and developed corresponding mining tools. 3. **Design of generators**: - In order to reproduce and explain these structural patterns, researchers have designed generators. - These generators can generate synthetic hypergraphs, thus providing valuable insights for understanding hypergraph structures. ### Main contributions of the paper - **Comprehensive overview**: Provides a comprehensive review of the field of hypergraph mining, covering patterns, tools, and generators. - **Classification system**: Provides a detailed classification system for structural patterns and generators. - **Future research directions**: Discusses the future applications and development directions of hypergraph mining. ### Classification of structural patterns - **Static patterns vs dynamic patterns**: Patterns are divided into static patterns and dynamic patterns according to whether time evolution is considered. - **Patterns at different levels**: Patterns are divided into node - level, hyper - edge - level, sub - hypergraph - level, and hypergraph - level according to the smallest unit defined by the pattern. ### Classification of generators - **Full - hypergraph generators vs sub - hypergraph generators**: Generators are divided into two categories according to whether the entire hypergraph or sub - hypergraph is generated. - **Static generators vs dynamic generators**: Generators are further subdivided according to whether static hypergraphs or dynamic (i.e., time) hypergraphs are generated. Through these problems and motivations, the paper aims to provide a systematic framework and in - depth analysis for research in the field of hypergraph mining.