We are excited to share our findings and specific supplementary material through the Verifee Research initiative. Fellow researchers, builders, and enthusiasts, this one’s for you! Below, you will find our so-far outputs, including a dataset and a set of machine learning models. Verifee’s research activities are conducted in close cooperation with our partners, The Faculty of Social Sciences at Charles University and Central European Digital Media Observatory academic hub.
Verifee is constantly learning and adapting, though its backbone leverages a robust news article dataset with trustworthiness annotations. Thanks to proactive journalism students and great support from Charles University, we collected a unique resource counting thousands of reports and texts complemented by information about present manipulative techniques. Are you seeking the right tools to study the Czech media landscape or train various specialized models? Look no further! We published a research paper about these findings at the AAAI'23 conference.
Can we assess a given article’s credibility without projecting our personal preference? This and many other complex questions were the subjects of our expert consortium debates. Assembled to lay a theoretical background for our project, this group contained an interdisciplinary scope of academics and journalists. After many of its iterations and discussions, we answer those questions in our methodology. We also describe quantifiable criteria of trustworthiness and offer its classification framework.
Verifee doesn’t just measure the overall trustworthiness – it shines light at specific manipulative techniques deep down the text, however subtle they might be. Their detectors will surely come in handy in many other applications as well. Thus, we open-source some of them to the research community.
Verifee @ 2022. All rights reserved.