{"id":137,"date":"2026-04-13T11:35:47","date_gmt":"2026-04-13T11:35:47","guid":{"rendered":"https:\/\/biasbreak.com\/?page_id=137"},"modified":"2026-04-13T11:35:47","modified_gmt":"2026-04-13T11:35:47","slug":"research","status":"publish","type":"page","link":"https:\/\/biasbreak.com\/ar\/research\/","title":{"rendered":"Our Research"},"content":{"rendered":"\n

Academically Validated. Peer-Reviewed. Built for Truth.<\/h2>\n\n\n\n
\n

BiasBreak is not just another AI tool it is built on a foundation of rigorous academic research. The technology powering BiasBreak was developed by Arif Wali, founder of BiasBreak and a Computer Science graduate of Middlesex University London, in collaboration with two independent academics. The research was formally peer-reviewed and published in an internationally recognized scientific journal, giving BiasBreak a level of scientific credibility that most AI tools simply do not have.<\/p>\n<\/blockquote>\n\n\n\n

\n

Published Research<\/h2>\n\n\n\n

Title:<\/strong> Towards a News Authenticity Predictor (NAP AI) <\/p>\n\n\n\n

Authors:<\/strong> Arif Wali, Stelios Kapetanakis, Giacomo Nalli Published in:<\/strong> Engineering Proceedings, MDPI \u2014 2026 DOI:<\/strong> 10.3390\/engproc2026124089 <\/p>\n\n\n\n

Presented at:<\/strong> 6th International Electronic Conference on Applied Sciences, December 2025<\/p>\n\n\n\n

This paper introduces the core AI system behind BiasBreak \u2014 a News Authenticity Predictor that uses Large Language Models and Natural Language Processing to detect misinformation, bias, and unverified claims in online content.<\/p>\n\n\n\n

\ud83d\udcc4 Read the Full Paper on MDPI \u2192<\/a><\/p>\n<\/blockquote>\n\n\n\n

\n\n\n\n
\n

What the Research Found<\/h2>\n\n\n\n

The model was rigorously tested on a dataset of 1,118 real and fake news articles. Here is what the results showed:<\/p>\n\n\n\n

98.03%<\/strong> \u2014 Overall Accuracy 98.15%<\/strong> \u2014 Precision (fake news detection) 98.15%<\/strong> \u2014 Recall 98.15%<\/strong> \u2014 F1-Score<\/p>\n\n\n\n

Out of 1,118 articles tested, only 22 were misclassified \u2014 and those errors were evenly balanced between false positives and false negatives, meaning the model does not unfairly lean toward labeling content as fake or real.<\/p>\n<\/blockquote>\n\n\n\n

\n

How the AI Works<\/h2>\n\n\n\n

The research behind BiasBreak combines two powerful approaches:<\/p>\n\n\n\n

1. Fine-Tuned BERT Model<\/strong> BiasBreak uses BERT (Bidirectional Encoder Representations from Transformers), a state-of-the-art language model developed by Google. It was fine-tuned on a diverse dataset including authentic news from outlets like BBC, Reuters, and The Guardian, as well as fake news articles verified by Snopes and FactCheck.org. BERT’s bidirectional architecture allows it to understand the full context of words \u2014 not just their surface meaning \u2014 making it exceptionally effective at spotting subtle linguistic patterns common in misinformation.<\/p>\n\n\n\n

2. Real-Time External Verification<\/strong> For claims requiring external validation, BiasBreak uses a retrieval-augmentation mechanism \u2014 querying live search results and cross-referencing claims against reputable sources in real time. This two-step process combines deep language understanding with live fact-checking for a more reliable and dynamic result.<\/p>\n<\/blockquote>\n\n\n\n

What We Trained On<\/h2>\n\n\n\n
\n

The model was trained on a carefully curated dataset combining multiple sources:<\/p>\n\n\n\n