Publication Type

PhD Dissertation

Version

publishedVersion

Publication Date

8-2025

Abstract

The rapid dissemination of information through online platforms has sparked widespread concern about the propagation of misinformation. Manual fact-checking by pro- fessional fact-checkers is time-consuming and lacks scalability to address the vast volume of daily information. Consequently, computational fact-checking, driven by automated techniques in natural language processing (NLP), has garnered interest as
a potential solution. However, computational fact-checking faces critical challenges limited resources, particularly due to the issues of data scarcity and computing resource constraints. One key challenge is data scarcity, which arises from the constant generation of new information and emerging events on social media. This scarcity manifests in two key ways: the limited availability of early-stage data during misinformation propagation and the lack of high-quality, labeled training data for new events. While Large Language Models (LLMs) have demonstrated impressive performance on various NLP tasks with limited data, but their substantial computa- tional requirements create another significant barrier. This highlights the pressing need for cost-effective solutions that can perform fact-checking tasks within limited computational resources while ensuring robust performance. These challenges em- phasize the importance of innovative approaches to effectively address fact-checking tasks with limited resources—a critical topic that remains largely unexplored.

To enhance computational fact-checking with limited resources, this thesis pro- poses methodologies that can make the most of the available data and computing resources, striking a balance between performance and resource availability. This thesis introduces the stabilization process of early rumor detection to automatically determine when to make a timely, accurate and stable predictions with limited early- stage data. Further, it proposes the first few-shot early rumor detection method to handle scenarios with limited labeled training instances. For fact verification and multimodal misinformation detection, this thesis proposes enabling medium-sized pre-trained models to effectively compete with LLMs and multimodal LLMs, in both few-shot and zero-shot settings. Furthermore, the thesis presents the first solution for justification generation with evidence retrieval in the few-shot setting, using a medium-sized Pre-trained Language Model (PLM). Extensive experiments show that the proposed methodologies achieve more effective performance as compared to strong baselines, including LLMs. Lastly, this thesis discusses potential future research directions to advance computational fact-checking with limited resources.

Keywords

Fact Checking, NLP, LLM

Degree Awarded

PhD in Computer Science

Discipline

Programming Languages and Compilers

Supervisor(s)

GAO, Wei

First Page

1

Last Page

203

Publisher

Singapore Management University

City or Country

Singapore

Copyright Owner and License

Author

Download

Share

COinS