Core Concepts
FakeNewsGPT4 utilizes LVLMs and forgery-specific knowledge to improve cross-domain performance in detecting multimodal fake news.
Abstract
FakeNewsGPT4 proposes a framework to enhance fake news detection by augmenting LVLMs with forgery-specific knowledge.
The framework involves multi-level cross-modal reasoning and dual-branch fine-grained verification to extract semantic correlations and artifact traces.
Candidate answer heuristics and soft prompts are used to enhance input informativeness.
Extensive experiments show FakeNewsGPT4 outperforms previous methods in cross-domain performance.
Quotes
"FakeNewsGPT4 achieves superior cross-domain performance compared to previous methods."