📄 Paper: mdpi.com/2078-2489/16/7…

💾 Source Code: github.com/altuncu/AKE

🎧 Podcasts:
🇹🇷 Sade-ce Bilim: open.spotify.com/episode/2gVX6Z���
🇬🇧 Research Talks: open.spotify.com/episode/5Hf#AutomaticKeywordExtractiontr#NaturalLanguageProcessingo#AcademicSkyd#OpenAccessenAccess

It presents a universal post-processing approach to boost AKE methods. Our experiments on 17 benchmark datasets showed significant improvements across 5 AKE methods.

Thanks all the co-authors (@jasonnurse.bsky.social,Yang Xu,Jie Guo, @hooklee.bsky.social) for their effort and collaboration.

Our paper is a dataset paper with example applications to show its usefulness. So, we used an existing SOTA political bias classifier (based on previous research and used in other relevant studies) for simplicity. How to validate/improve the performance of this model was beyond our scope.

The predictions of the model based on previously published fact-checking reports says so. But, it's important to note that the reported findings are limited by the accuracy of the model.

We used a publicly available BERT-based pretrained model for political bias classification.

Those bars refer to the standard deviation of predicted bias scores. If a bar is shown on both sides of the centre line, it means the corresponding organisation published both left- and right-leaning reports.

📄 Preprint: arxiv.org/abs/2505.09414
💾 Dataset & code: github.com/altuncu/FACT...

#FactChecking #Dataset #AcademicChatter #AcademicResearch #InformationRetrieval #SIGIR2025

🛠️ Example applications enabled by the dataset:

• Detailed stats for 39 orgs + ~2k authors (activity span, avg. word counts, % unique claims)
• Predicted political bias scores (via a pre-trained model)
• Credibility assessment based on historical signals

✅ ~118k English fact-checking reports from 39 organisations:

✔️ All are active signatories of @factchecknet.bsky.social and/or EFCSN
🗓️ Spanning 1995–2025
💬 Includes 7k+ claims investigated by multiple organisations

Many thanks to the co-authors (Can Başkent, Sanjay Bhattacherjee, Shujun Li, and Dwaipayan Roy) for their effort and collaboration.

Hi, just added!