EAGER: Live Reality: Sustainable and Up-to-Date Information Quality in Live Social Media through Continuous Evidence-Based Knowledge Acquisition

EAGER：实时现实：通过持续的循证知识获取，实时社交媒体中可持续且最新的信息质量

• 批准号: 2039653
• 负责人: Calton Pu
• 金额: $ 15万
• 依托单位: Georgia Tech Research Corporation
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2020
• 资助国家: 美国
• 起止时间: 2020-10-01 至 2024-09-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2039653&HistoricalAwards=false
• 关键词: EAGER; Live; Reality; Sustainable; Up

Social media have complemented traditional press with immediate reports and worldwide coverage. However, they also receive and propagate significant amounts of misinformation and disinformation such as fake news. A skillful mixture of verifiable facts and outrageous fiction, fake news aim to attract reader attention, make an immediate initial impact, and then quickly forgotten. Even as disposable novelty, fake news have had significant impact on real world events such as elections. For human readers and machine learning (ML) classifiers, distinguishing fake news from real news has been challenging due to their sophisticated construction, camouflaging fiction with facts, as well as continuously evolving by incorporating the newest and hottest topics as they mutate. The Live Reality project will track the evolution of fake news through continuous import of reliable, verified facts from authoritative sources, and separate the facts from fiction, to catch fake news in the act. The automated real-time tracking capability is a significant innovation compared to traditional ML classifiers generated from manually labeled training data, which are constrained to finding historical fake news, long after the fact.Given the short lifespan of disposable novelty (days or hours), catching fake news in the act requires significant innovation in two dimensions. First, the ML classifier must be continuously updated to recognize true novelty that have never been seen before. Second, the update must be sufficiently timely to catch disposable novelty before they expire, e.g., within hours of their initial dissemination. Continuous collection of live social media and authoritative sources will generate novel fake news and associated ground truth, which are integrated through the Evidence-Based Knowledge Acquisition (EBKA) approach, which adds reliable information from authoritative sources into a continuously adaptive teamed classifier to distinguish the verifiable facts from the fiction in fake news. As news topics evolve, fake news are expected to follow, and EBKA will generate and integrate new sub-models into the live teamed classifier to recognize the new topics. The EBKA approach will be demonstrated on live data containing fake news on a variety of topics, specifically disaster management such as the COVID-19 pandemic. Due to the disposable novelty nature of fake news, EBKA will be evaluated in two dimensions: classifier performance in terms of accuracy and precision, and timeliness of classifier identifying truly new fake news soon after their appearance in the real world.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.

社交媒体补充了传统媒体的即时报道和全球报道。然而，他们也接收和传播大量的错误信息和虚假信息，如假新闻。假新闻巧妙地混合了可证实的事实和令人发指的虚构，旨在吸引读者的注意力，立即产生初步影响，然后很快被遗忘。即使是一次性的新奇事物，假新闻也对选举等真实的世界事件产生了重大影响。对于人类读者和机器学习（ML）分类器来说，区分假新闻和真实的新闻一直是一个挑战，因为它们的结构复杂，用事实来掩盖虚构，并且随着它们的变化而不断地融入最新和最热门的话题。Live Reality项目将通过不断从权威来源导入可靠的、经过验证的事实来跟踪假新闻的演变，并将事实与虚构分开，以捕捉假新闻的行为。与传统的人工标记训练数据生成的机器学习分类器相比，自动实时跟踪功能是一项重大创新，传统的机器学习分类器只能在事实发生很久之后才能发现历史假新闻。鉴于一次性新奇事物的寿命很短（几天或几小时），在行为中捕捉假新闻需要两个维度的重大创新。首先，机器学习分类器必须不断更新，以识别以前从未见过的真正新奇。其次，更新必须足够及时，以便在过期之前捕捉一次性的新奇，例如，在最初传播的几小时内。持续收集实时社交媒体和权威来源将产生新的假新闻和相关的地面真相，这些假新闻和相关的地面真相通过基于证据的知识获取（EBKA）方法进行整合，该方法将来自权威来源的可靠信息添加到一个持续自适应的团队分类器中，以区分可验证的事实和假新闻中的虚构内容。随着新闻主题的发展，假新闻预计会随之而来，EBKA将生成新的子模型并将其集成到实时团队分类器中，以识别新的主题。EBKA方法将在包含各种主题假新闻的实时数据上进行演示，特别是COVID-19大流行等灾害管理。由于假新闻的一次性新奇性，EBKA将从两个方面进行评估：准确度和精确度方面的分类器性能，以及分类器在真实的世界中出现后不久识别真正新的假新闻的及时性。该奖项反映了NSF的法定使命，并通过使用基金会的智力价值和更广泛的影响审查标准进行评估，被认为值得支持。

项目成果

Exploring Generalizability of Fine-Tuned Models for Fake News Detection

• DOI: 10.1109/cic56439.2022.00022
• 发表时间: 2022-12
• 期刊: 2022 IEEE 8th International Conference on Collaboration and Internet Computing (CIC)
• 作者: Abhijit Suprem;Sanjyot Vaidya;C. Pu

Constructive Interpretability with CoLabel: Corroborative Integration, Complementary Features, and Collaborative Learning

• DOI: 10.1109/cogmi56440.2022.00021
• 发表时间: 2022-05
• 期刊: 2022 IEEE 4th International Conference on Cognitive Machine Intelligence (CogMI)
• 作者: Abhijit Suprem;Sanjyot Vaidya;Suma Cherkadi;Purva Singh;J. E. Ferreira;C. Pu

Evolution of Knowledge in Social Media and Their Relationship to an Evolving Real World

• DOI: 10.1109/cogmi58952.2023.00013
• 发表时间: 2023-11
• 期刊: 2023 IEEE 5th International Conference on Cognitive Machine Intelligence (CogMI)
• 作者: C. Pu;Abhijit Suprem;A. Musaev;J. Ferreira