真实世界高阶网络公开数据集

多智能体超图框架下的超图数据集整合了多个独立的真实世界超图，以支持影响力最大化问题的研究、泛化分析及优化算法的性能验证。本研究公开了八个不同来源的超图数据集，涵盖多个应用场景，包括社交网络、学术合作、电子商务和在线问答平台，以验证所提出方法在多样化超图结构中的有效性。具体描述如下：
1. HouseCommittees 超图：该数据集由 Charles Stewart 和 Jonathan Vaughan 编译，节点表示美国众议院的议员，超边表示各委员会的成员关系。整个超图包含 1290 个节点和 341 条超边，超边的平均规模为 34.8，中位数为 40，适用于研究社会关系网络中的群体结构和影响力传播模式。
2. Vegas-bars-reviews 超图：基于 Yelp Kaggle 竞赛数据构建，节点表示 Yelp 用户，超边表示在一个月内对某类场所（如酒吧）进行评价的用户集合。该数据集可用于研究用户行为模式及社交影响在消费决策中的作用。
3. Music-blues-review 超图：该超图由 Ni 等人收集的亚马逊产品评论数据生成，节点表示亚马逊评论者，超边表示评论同一类别产品的用户集合，适用于分析消费者兴趣聚类和意见领袖的影响范围。
4. Geometry-question 超图：该数据集源自 MathOverflow 平台，节点表示用户，超边表示回答某一几何类问题的用户集合。此超图为研究学术问答社区中的知识传播和专家识别提供了基础。
5. SenateCommittees 超图：该超图结构类似于 HouseCommittees，但节点表示美国参议院的议员，超边对应各委员会的成员关系。该数据集适用于分析决策群体中的影响力分布和政策传播机制。
6. Algebra-question 超图：来自 StackExchange 数据集，节点表示用户，超边表示回答同一代数问题的用户集合。该超图可用于研究在线学习社区中的知识共享模式和核心贡献者识别。
7. MAG-10 超图：基于 Microsoft Academic Graph (MAG) 数据子集构建，节点表示学术作者，超边表示作者与其发表论文之间的关系。该数据集提供了研究学术合作网络和知识传播模式的基础。
8. Mathoverflow-answers 超图：该超图基于 MathOverflow 平台，节点表示用户，超边表示回答同一问题的用户集合，适用于研究问答社区中的信息流动和影响力扩散模式。
这些数据集涵盖了多个领域的真实世界超图结构，为影响力最大化问题的研究提供了丰富的实验基准。通过在不同拓扑结构和传播机制下进行实验，该数据集有助于评估超图集体影响力（Hypergraph Collective Influence, HCI）模型在阈值传播过程中的有效性，并优化影响力传播策略。结合理论分析和数值模拟，研究人员可利用这些数据集探索超图结构对影响力扩散的作用，从而优化信息传播、社交网络推荐及关键节点识别等任务。

The hypergraph dataset under the multi-agent hypergraph framework integrates multiple independent real-world hypergraphs, supporting research on the influence maximization problem, generalization analysis, and performance verification of optimization algorithms. This study publicly releases eight hypergraph datasets from different sources, covering multiple application scenarios including social networks, academic collaborations, e-commerce, and online question-and-answer platforms, to verify the effectiveness of the proposed method in diverse hypergraph structures. The specific descriptions are as follows:
1. HouseCommittees Hypergraph: Compiled by Charles Stewart and Jonathan Vaughan, this dataset uses nodes to represent members of the U.S. House of Representatives, and hyperedges to represent the membership of each committee. The entire hypergraph contains 1290 nodes and 341 hyperedges, with an average hyperedge size of 34.8 and a median of 40. It is suitable for studying group structures and influence propagation patterns in social relationship networks.
2. Vegas-bars-reviews Hypergraph: Built based on the Yelp Kaggle Competition dataset, this hypergraph uses nodes to represent Yelp users, and hyperedges to represent the set of users who rated a certain type of venue (e.g., bars) within one month. This dataset can be used to study user behavior patterns and the role of social influence in consumption decisions.
3. Music-blues-review Hypergraph: Generated from Amazon product review data collected by Ni et al., this hypergraph uses nodes to represent Amazon reviewers, and hyperedges to represent the set of users who reviewed products of the same category. It is applicable to analyzing consumer interest clustering and the influence scope of opinion leaders.
4. Geometry-question Hypergraph: Derived from the MathOverflow platform, this hypergraph uses nodes to represent users, and hyperedges to represent the set of users who answered a certain geometric question. This hypergraph provides a basis for studying knowledge propagation and expert identification in academic Q&A communities.
5. SenateCommittees Hypergraph: Structurally similar to HouseCommittees, this hypergraph uses nodes to represent members of the U.S. Senate, and hyperedges correspond to the membership of each committee. This dataset is suitable for analyzing influence distribution and policy propagation mechanisms in decision-making groups.
6. Algebra-question Hypergraph: From the StackExchange dataset, this hypergraph uses nodes to represent users, and hyperedges to represent the set of users who answered the same algebraic question. It can be used to study knowledge sharing patterns and core contributor identification in online learning communities.
7. MAG-10 Hypergraph: Built based on a subset of the Microsoft Academic Graph (MAG) dataset, this hypergraph uses nodes to represent academic authors, and hyperedges to represent the relationship between authors and their published papers. This dataset provides a basis for studying academic collaboration networks and knowledge propagation patterns.
8. Mathoverflow-answers Hypergraph: Based on the MathOverflow platform, this hypergraph uses nodes to represent users, and hyperedges to represent the set of users who answered the same question. It is applicable to studying information flow and influence diffusion patterns in Q&A communities.

These datasets cover real-world hypergraph structures from multiple fields, providing rich experimental benchmarks for research on the influence maximization problem. Through experiments under different topological structures and propagation mechanisms, this dataset helps evaluate the effectiveness of the "Hypergraph Collective Influence (HCI)" model during threshold propagation, and optimize influence propagation strategies. Combining theoretical analysis and numerical simulations, researchers can use these datasets to explore the role of hypergraph structures in influence diffusion, thereby optimizing tasks such as information propagation, social network recommendation, and key node identification.

搜集汇总

背景与挑战

背景概述

该数据集整合了八个真实世界超图，覆盖社交网络、学术合作、电子商务和在线问答平台等多个领域，旨在为影响力最大化问题的研究提供实验基准，支持算法性能验证和泛化分析。数据包括3.83MB的39个文件，格式主要为txt和docx，适用于复杂网络和高阶网络相关的人工智能研究。