Collaborative Research: HCC: Medium: Fine-grained Emotion Analysis in Crises

历史上有很多这样的情况：同一件事，不同群体的人却有完全不同的解释。通过辛普森案、COVID-19危机和乔治·弗洛伊德谋杀案等事件，我们看到了社区领导人、警察部门、政策制定者和普通公民未能预料到的对事件的不同反应。这个项目的目的是开始识别危机的社会，情感和语言标记（例如，社会动荡，自然灾害等），预测人们对相同事件的不同反应方式。这是通过分析社交媒体的语言来实现的，社交媒体是一个快速增长的数据来源，我们可以从中了解大规模的情绪表达和感知，从学术研究到公共政策都有深远的潜在用途。理解情绪，这些情绪周围的环境，以及随后的行为对于那些处于危机中的人，寻求危机的信息，或帮助管理对危机的反应是非常有价值的。该项目将发现在不同社交平台上提供全面、细粒度情感分析的机制，并得出稳健可靠的预测模型。细粒度情绪分析旨在：(1)检测文本中的情绪表达并表征其强度和极性；(2)识别引起情绪的触发因素；(3)分析情绪偏差（即人们对同一触发因素表达的不同情绪）。本研究将为跨不同危机的社交媒体上表达的情绪提供带注释的数据集，并为情境化情绪分析提供具有深度语言理解的可推广模型。工业界和学术界合作伙伴将通过评估模型在不同于用于开发模型的情况和数据源上工作的能力来参与。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

History is rich with situations where the same event has been interpreted completely differently by different groups of people. Through events such as the OJ Simpson case, the COVID-19 crisis, and the murder of George Floyd, we have observed disparate reactions to events that community leaders, police departments, policymakers, and everyday citizens fail to anticipate. The purpose of this project is to begin to identify social, emotional, and linguistic markers of crises (e.g., social turmoil, natural disasters, etc.) that predict the various ways people will react to the same events. This is achieved by analyzing the language of social media, a rapidly-growing source of data from which we can understand the expression and perception of emotions at a very large scale, with far-reaching potential uses from academic research to public policy.Understanding emotions, the context surrounding these emotions, and subsequent behaviors are of great value to those in a crisis, seeking information about a crisis, or helping manage responses to a crisis. This project will discover mechanisms to provide comprehensive, fine-grained emotion analysis across different social platforms, and derive robust and reliable predictive models. Fine-grained emotion analysis aims to: (1) detect expressions of emotions in a text and characterize their intensity and polarity, (2) identify the triggers causing the emotions, and (3) analyze emotion deviation (i.e., the varied emotions that people express towards the same trigger). This research will contribute annotated datasets of emotions expressed on social media across distinct crises and generalizable models equipped with deep linguistic understanding for contextualized emotion analysis. Industry and academic partners will participate by evaluating the ability of the models to work on situations and data sources different from those used to develop the models.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.