Models and Inferences for Heterogeneous Interaction Patterns in Social Networks

社交网络平台的快速发展为识别、提取和调查用户对科学和公共利益话题的情绪带来了机遇和挑战。本研究旨在找出社会网络异质性的来源。该项目将研究回复帖子的可能性，以及通过社交网络传播情绪的可能性。这项研究的结果预计将有助于通过用户之间的互动来识别情绪交换模式；检测具有情绪集群或表现出情绪两极分化的帖子的子网络；并了解传染性情绪如何通过Twitter帖子网络传播。该项目将通过研究生咨询、现代化研究生课程内容、本科生对精心设计的小问题的研究以及对高中生的推广，纳入多层次教育。将通过与在任职人数不足的群体中促进STEM的组织建立伙伴关系，鼓励广泛参与。该项目的结果将通过数据/代码、软件包、在线教程和社交媒体的公共存储库共享。Twitter用户之间通过转发、回复和提及特定主题(例如，抑郁)来形成动态网络，这些在情绪分析中至关重要。现有的动态网络模型不能考虑异质行为模式，不能为大型真实网络提供足够的匹配。该项目旨在解决动态网络建模中的两个异构性来源：情感共享的互惠和行为特征方面的相似性聚类。研究人员计划开发两个新的模型来修改优先依恋(PA)模型。第一种方法保留了无标度特性，并允许个性化的异质倾向产生互易边。第二个模型是一个实用的空间超级明星PA模型，它允许在特征空间中测量到的更紧密社会关系的节点之间更高级别的交互。我们将通过严格的渐近分析来研究它们的理论性质。在拟合观测网络时，将通过基于似然、贝叶斯、矩和极值的方法来推断模型参数。模型和推论的实施将通过用户友好的开放源码软件包公开提供。所有正在开发的方法都将通过模拟研究进行验证，验证后的方法将应用于心理健康相关标签上的推文中的动态网络。这一奖项反映了NSF的法定使命，并已通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

The rapid growth of social network platforms brings opportunities as well as challenges to identify, extract, and investigate the users’ sentiment on subjects of scientific and public interest. This research project aims to identify the sources of social network heterogeneity. The project will study the likelihood of replies to a posting as well as the propagation of sentiments through a social network. Results of the study are expected to help identify sentiment exchange patterns through interactions between users; detect sub-networks of posts with emotional clusters or exhibiting sentiment polarization; and understand how contagious sentiments propagate through the network of Twitter posts. The project will be integrated into multilevel education through graduate student advising, modernized graduate course components, undergraduate research on carefully devised small problems, and outreach to high school students. Broad participation will be encouraged through partnerships with organizations promoting STEM among under-represented groups. Results from the project will be shared via public repositories for data/codes, software packages, online tutorials, and social media. Dynamic networks are formed among Twitter users by evolving tweets on a certain theme (e.g., depression) through retweets, replies, and mentions, which are critical in sentiment analysis. Existing dynamic network models not accounting for heterogeneous behavioral patterns cannot provide adequate fits for large, real networks. This project aims to tackle two sources of heterogeneity in dynamic network modeling: reciprocity in emotion sharing and similarity clustering in terms of behavioral features. The investigators plan to develop two novel models that modify the preferential attachment (PA) model. The first one retains the scale-free property and allows personalized heterogeneous tendencies to generate reciprocal edges. The second model is a practical spatial superstar PA model that allows a higher level of interaction among nodes of closer social ties measured in a feature space. Their theoretical properties will be studied via rigorous asymptotic analyses. Inferences about the model parameters when fitting observed networks will be developed through likelihood-based, Bayesian, moment-based, and extreme value approaches. Implementations of the models and inferences will be made publicly available via user-friendly, open-source software packages. All the methods under development will be validated through simulation studies, and the validated methods will then be applied to dynamic networks from Tweets on mental-health-related tags.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.