III: Small: Collaborative Research: Modeling and Managing Extremist Group Influence in Massive Social Media Networks

III：小型：协作研究：在大规模社交媒体网络中建模和管理极端主义团体的影响力

• 批准号: 1909252
• 负责人: James Caverlee
• 金额: $ 25万
• 依托单位: Texas A&M Engineering Experiment Station
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2019
• 资助国家: 美国
• 起止时间: 2019-10-01 至 2022-09-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1909252&HistoricalAwards=false
• 关键词: III; Small; Collaborative; Research; Modeling

Social media enables an unprecedented positive shift in how people communicate and consume content. In the other direction, extremist groups can embrace social media to spread negative/violent ideologies. This project aims to build mathematical and data-driven models to understand the dynamics of extremist groups at scale, the patterns of their influence, and integrated micro (individual-level) and macro (group-level or system-level) data-driven models that can guide future interventions. This project provides a greater understanding of users' behavioral patterns and social dynamics related to online extremism. This will be coupled with a set of technical solutions to detect and counter episodes of self-isolation and anti-social behavior. This project may lead to new results eventually impacting social sciences, by means of new or refined cyber-behavior theories, and in exciting, novel mathematical models and natural language processing methods for online discourse. This project will promote awareness and normative behavior to both K-12 students and college students, and will train and educate students in mathematics and computer science.Given the exceptional increase in social network use and the critical role it is now playing among extremist groups, this project focuses on developing methods and algorithms for modeling and managing anti-normative behaviors. The goal is to mitigate the effects of negative behavior on user social interactions while simultaneously managing the ideological self-isolating effect social networks can encourage. The approaches will involve both macro and micro level statistical behavior models incorporated with innovative natural language processing and deep learning methods. One approach to be investigated is behavioral models of extremist influence, which are creating mathematically rigorous behavioral models of social capital at the individual level that can capture imitative behaviors. A second approach is data-driven discourse-level contagiousness, use to identify extremist discourse structures representing ideological payloads, and uncover overall patterns of extremist ideology infectiousness. The third approach, nudging antinormative behaviors, builds on the others to design and validate new interventions that could lessen the impact of extremist group influence through simulation and off-line experiments. Experimental results, datasets, and project software will be made accessible to the research community via public project websites and code sharing platforms (e.g., GitHub).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.

社交媒体使人们在交流和消费内容的方式上实现了前所未有的积极转变。在另一个方向上，极端主义团体可以拥抱社交媒体来传播负面/暴力的意识形态。该项目旨在构建数学和数据驱动的模型，以大规模了解极端主义团体的动态，其影响的模式以及集成的微观（个人级别）和宏（组级或系统级别的数据驱动）模型，这些模型可以指导未来的干预措施。该项目对用户的行为模式和与在线极端主义相关的社会动态提供了更深入的了解。这将与一组技术解决方案相结合，以检测和反对自我隔离和反社会行为的发作。该项目可能会通过新的或精致的网络行为理论，以及令人兴奋的新型数学模型和自然语言处理方法来影响社会科学的新结果，从而影响社会科学。该项目将促进K-12学生和大学生的认识和规范行为，并将培训和教育数学和计算机科学的学生。赋予社交网络使用的显着增长，以及它现在在极端主义团体之间发挥的关键作用，该项目着重于开发方法和算法来建模和管理反阶级行为。 目的是减轻负面行为对用户社交互动的影响，同时管理社交网络可以鼓励的意识形态自我溶解效果。这些方法将涉及宏观和微级别统计行为模型，这些模型与创新的自然语言处理和深度学习方法结合在一起。 一种要调查的方法是极端主义影响的行为模型，这些模型正在在个人层面上创建数学上严格的社会资本行为模型，以捕获模仿行为。第二种方法是数据驱动的话语级传染性，用于识别代表意识形态有效载荷的极端主义话语结构，并揭示了极端主义意识形态感染性的整体模式。第三种方法是推动反式行为的方法，它建立在其他方法上，以设计和验证新的干预措施，这些干预措施可以通过模拟和离线实验来减少极端主义群体影响的影响。实验结果，数据集和项目软件将通过公共项目网站和代码共享平台（例如Github）访问。该奖项反映了NSF的法定任务，并被认为是值得通过基金会的知识分子优点和更广泛的影响审查标准通过评估来进行评估的。

项目成果

One Classifier for All Ambiguous Words: Overcoming Data Sparsity by Utilizing Sense Correlations Across Words

• 发表时间: 2020-05
• 期刊: Proceedings of the 8th International Conference on Computing and Artificial Intelligence
• 作者: Prafulla Kumar Choubey;Ruihong Huang

PARADE: A New Dataset for Paraphrase Identification Requiring Computer Science Domain Knowledge

• DOI: 10.18653/v1/2020.emnlp-main.611
• 发表时间: 2020-10
• 期刊: ArXiv
• 作者: Yun He;Zhuoer Wang;Yin Zhang;Ruihong Huang;James Caverlee

Explicitly Capturing Relations between Entity Mentions via Graph Neural Networks for Domain-specific Named Entity Recognition

• DOI: 10.18653/v1/2021.acl-short.93
• 发表时间: 2021
• 作者: Pei Chen;Haibo Ding;J. Araki;Ruihong Huang