KEEN - Knowledge-driven Explainable Misinformation Detection for Trustworthy Computational Social Systems

KEEN - 知识驱动的可解释错误信息检测，用于可信赖的计算社会系统

• 批准号: EP/Y015894/1
• 负责人: Geyong Min
• 金额: $ 25.55万
• 依托单位: UNIVERSITY OF EXETER
• 依托单位国家: 英国
• 项目类别: Fellowship
• 财政年份: 2024
• 资助国家: 英国
• 起止时间: 2024 至 无数据
• 项目状态: 未结题
• 来源: https://gtr.ukri.org/projects?ref=EP%2FY015894%2F1
• 关键词: KEEN; Knowledge; driven; Explainable; Misinformation

With the prosperity of social media platforms like Facebook and Twitter, misinformation can be disseminated widely among the general public, causing a severe threat to the trustworthiness of computational social systems. To address this critical issue, various misinformation detection models have been proposed recently. However, the existing methods either use black-box deep learning (DL) models which cannot provide explainability of detection results, or leverage shallow experience-based explainable models which leads to low detection accuracy.This project aims to create a novel knowledge-driven approach to build both accurate and explainable misinformation detection models for trustworthy computational social systems. To this end, I will first establish a novel knowledge-driven integration mechanism to seamlessly integrate social psychological theories with DL models based on multi-modal social media data. Secondly, a novel explanation scheme will be developed to effectively convey social psychological theories into reliable model explainability through knowledge extraction. Thirdly, an accurate and explainable DL framework will be constructed base on hybrid DL models and hierarchical attention-based explanation. Finally, a prototype system will be developed to implement the proposed solutions and evaluate their performance. The scientific breakthroughs to be made in this project will contribute to provide the effective design of accurate and explainable misinformation detection models. The originality of this project lies in its interdisciplinary research on how to establish an innovative explainable DL approach for trustworthy computational social systems. A series of well-arranged research, training, knowledge transfer, and open science activities are planned to accomplish the ambitious aim of this project, facilitate knowledge transfer and dissemination, and enhance my creative and innovative potential and careerprospects with new skills and competences.

随着Facebook和Twitter等社交媒体平台的繁荣，错误信息可以在公众中广泛传播，对计算社交系统的可信度造成严重威胁。为了解决这一关键问题，最近提出了各种错误信息检测模型。然而，现有的方法要么使用黑盒深度学习（DL）模型，无法提供检测结果的可解释性，要么利用基于经验的浅层可解释模型，导致检测准确率低。本项目旨在创建一种新的知识驱动方法，为可信的计算社会系统构建准确且可解释的错误信息检测模型。为此，我将首先建立一个新的知识驱动的整合机制，无缝整合社会心理学理论与基于多模态社会媒体数据的深度学习模型。其次，一个新的解释方案将开发有效地传达社会心理学理论到可靠的模型解释能力，通过知识提取。第三，基于混合深度学习模型和基于注意力的层次化解释，构建一个精确的、可解释的深度学习框架。最后，将开发一个原型系统来实现所提出的解决方案，并评估其性能。该项目取得的科学突破将有助于提供准确和可解释的错误信息检测模型的有效设计。该项目的独创性在于其跨学科研究如何建立一个创新的可解释的DL方法，值得信赖的计算社会系统。一系列精心安排的研究，培训，知识转移和开放科学活动计划完成这个项目的宏伟目标，促进知识转移和传播，并提高我的创造力和创新潜力和职业前景与新的技能和能力。