VeriNews: An Explainable AI Framework for Fake News Detection

Abstract

The widespread use of digital news services, social media, and instant communication channels has made information dissemination faster and more accessible, but it has also increased the spread of misleading and fabricated content. This paper introduces VeriNews, an explainable artificial intelligence framework for assisting with the analysis of potentially deceptive news. The framework combines textual processing, contextual language representations, linguistic characteristics, source-oriented signals, external evidence retrieval, and prediction-confidence estimation within a unified verification pipeline. In addition to generating a reliability classification, VeriNews identifies evidence and explanatory factors that allow users to better understand the basis of a model prediction. The proposed workflow includes data collection, text preprocessing, representation learning, classification, evidence retrieval and comparison, confidence assessment, and explanation generation. Benchmark resources such as the LIAR and FakeNewsNet datasets may be employed for model development and evaluation, while transformer architectures including BERT and RoBERTa can be investigated for contextual text representation. The study also considers robustness, dataset-related bias, privacy, interpretability, and the uncertainty inherent in automated information verification. VeriNews is therefore positioned as a human-oriented decision-support framework rather than an autonomous authority on factual truth, particularly in cases where available evidence is limited or inconclusive.