SAFE
Structured Argumentation for Fact-checking with Explanations
- Generate argument-structured summaries based on a fact-checking article or a list of evidence.
- Retrieve evidence and generate argument-structured summaries when the fact-checking article is not available.
- Assign a truthfulness label to a news claim which is enriched with a summary.
Module demonstration
Argument-structured Summarization
This module analyzes a fact-checking article or a list of pieces of evidence. Then it produces an argument-structured summary indicating claims, premises, and argument relations. An argument graph is also provided.
Retrieval-based Summarization
This module retrieves relevant pieces of evidence to fact-check claims then formulates the evidence in an argument-structured summary. It includes datasets such as LIAR-PLUS, FNC-1, Check-Covid, and ExClaim. Users can also provide their own custom corpora tailored to their specific interests.
Fake News Classification
This module uses a claim enriched with the argument-structured summary as input and produces a truthfulness label using the state-of-the-art FNC method.
Pipeline demonstration
The complete pipeline of SAFE
This interface provides an overview of the entire SAFE pipeline, detailing how evidence retrieval, argument-structured summarization, and fake news classification work together to provide a comprehensive and explainable fact-checking solution.