Peng QI      

I'm currently a postdoctoral research fellow in the Centre for Trusted Internet and Community, National University of Singapore, under the supervision of Prof. Wynne Hsu and Prof. Mong Li Lee. I received my Ph.D. degree at last June in the Institute of Computing Technology, Chinese Academy of Sciences, supervised by Prof. Juan Cao. I also had a one-year visiting study in NExT++, NUS, supervised by Prof. Tat-Seng Chua.

My research interests mainly lie in combating misinformation, multimedia content analysis, and social media mining. Recently, I focus more on misinformation detection and intervene in the era of LLM.

Open for collaboration on interesting topics! :)

SNIFFER: Multimodal Large Language Model for Explainable Out-of-Context Misinformation Detection

Focusing on the innovative research perspective of explainable out-of-context misinformation detection, this paper proposes a new multimodal large language model, SNIFFER, designed to offer both accurate detection and persuasive explanations simultaneously. Enhanced by two-stage instruction tuning and retrieval-enhancement techniques, SNIFFER effectively models both internal image-text inconsistency and external claim-evidence relationships.

This paper proposes that current LLMs may not substitute fine-tuned SLMs in fake news detection but can be a good advisor for SLMs by providing instructive multi-perspective rationales. Based on this assumption, we design an adaptive rationale guidance network ARG in which SLMs selectively acquire insights from LLM's rationales and further derive a rationale-free version by distillation.

This paper proposes a model-agnostic cross-sample framework NEED for detecting fake news videos, which utilizes the neighbor relationship between the related news videos in the same event and thus can largely improve the performance of existing single-sample detectors. This paper is a successful attempt to combine automatic fake news detection and fact-checking, and firstly proves the effectiveness of debunking videos in detection.

Last updated: 9 March 2024

Stolen from Jon Barron