Modelling and Mining Complex Networks

复杂网络的建模和挖掘

• 批准号: RGPIN-2022-03804
• 负责人: Pralat, Pawel
• 金额: $ 2.99万
• 依托单位: Ryerson University
• 依托单位国家: 加拿大
• 项目类别: Discovery Grants Program - Individual
• 财政年份: 2022
• 资助国家: 加拿大
• 起止时间: 2022-01-01 至 2023-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=758564
• 关键词: Modelling; Mining; Complex; Networks

Currently, we experience a rapid growth of research done in the intersection of mining and modelling of social networks. This research proposal concentrates on problems from this intersection. There are two main reasons to include random graph models in mining complex networks: - - Synthetic models. Many important algorithms (such as community detection algorithms) are unsupervised in nature. Moreover, despite the fact that the research community gets better with exchanging datasets (see, for example, Stanford Large Network Dataset Collection) there are still very few publicly available networks with known underlying structure, the so-called ground truth. Hence, to test, benchmark, and properly tune unsupervised algorithms, one may use random graphs to produce synthetic "playground": graphs with known ground truth (such as the community structure in the context of community detection algorithms). - - Null -models. Null--model is a random object that matches one specific property P of a given object but is otherwise taken, unbiasedly, at random from the family of objects that have property P. As a result, the null-models can be used to test whether a given object exhibits some "surprising" property that is not expected on the basis of chance alone or as an implication of the fact that the object has property P. It is expected that both applications of random graphs will continue to gain their importance. My experience with industrial projects allows me to identify missing tools and algorithms that are of interest to the practitioners. On the other hand, my pure research background allows me to better understand processes that shape self-organizing complex networks and, as a result, to be better prepared to design efficient algorithms that work on data collected from real-world applications. In the proposal, we will discuss in detail the following objectives: - Unsupervised Framework for Evaluating of Node Embedding Algorithms, - Community Detection in Networks Modelled as Hypergraphs, - Generating Synthetic Networks. In each of these sub-projects, rigorous definitions, theorems, and proofs (that will be published in research papers) are needed to design the tool, and the tool will be implemented in Julia language (that will be made publicly available on GitHub repository) and carefully tested on both synthetic and real- world networks.

目前，我们经历了在社交网络挖掘和建模的交叉领域进行的研究的快速增长。本文的研究方案集中在这一交叉点的问题上。在挖掘复杂网络中包含随机图模型有两个主要原因：--合成模型。许多重要的算法(如社区检测算法)本质上是无监督的。此外，尽管研究界在交换数据集方面做得更好(例如，参见斯坦福大型网络数据集收集)，但具有已知底层结构的公开可用网络仍然很少，即所谓的地面事实。因此，为了测试、基准测试和适当调整非监督算法，可以使用随机图来生成合成的“游乐场”：具有已知基本事实的图(例如社区检测算法上下文中的社区结构)。--空-模特。空-模型是与给定对象的一个特定属性P匹配的随机对象，但是以其他方式无偏见地从具有属性P的对象族中随机获取。因此，空模型可以用于测试给定对象是否表现出一些仅基于偶然或者作为该对象具有属性P的事实的暗示而没有被期望的“令人惊讶的”属性。预计随机图的两个应用将继续获得它们的重要性。我在工业项目中的经验使我能够识别从业者感兴趣的缺失的工具和算法。另一方面，我纯粹的研究背景使我能够更好地理解形成自组织复杂网络的过程，从而更好地准备设计有效的算法，这些算法处理从现实世界应用程序收集的数据。在该方案中，我们将详细讨论以下目标：-节点嵌入算法的无监督评估框架，-超图网络中的社区检测，-生成合成网络。在每个子项目中，都需要严格的定义、定理和证明(将在研究论文中发表)来设计工具，该工具将以Julia语言实现(将在GitHub存储库公开提供)，并在合成和真实世界的网络上进行仔细测试。