An AI-Driven Framework for Political Misinformation Detection Using DistilRoBERTa: Strengthening Democratic Communication in the United States

Abstract

The rapid proliferation of political misinformation across digital media platforms poses a serious threat to democratic communication, public trust, and informed decision-making. Although traditional machine learning approaches have demonstrated promising performance in fake news detection, they often struggle to capture the contextual and semantic characteristics of political news. This study proposes an AI-driven media framework for political misinformation detection using the PolitiFact subset of the FakeNewsNet dataset. The proposed framework employs a comprehensive text preprocessing pipeline, including text normalization, URL removal, punctuation elimination, and feature selection, followed by a comparative evaluation of a TF–IDF with Logistic Regression baseline and a fine-tuned DistilRoBERTa transformer model. The dataset was partitioned into training, validation, and testing sets using a 70:15:15 ratio, with the test set comprising 212 news articles (125 real and 87 fake). Experimental results demonstrate that the proposed framework outperforms the conventional baseline across all evaluation metrics. Specifically, the fine-tuned DistilRoBERTa model achieved an accuracy of 88.21%, precision of 82.98%, recall of 89.66%, F1-score of 86.19%, and an ROC-AUC of 95.92%, compared with the Logistic Regression baseline, which obtained an accuracy of 85.38%, precision of 87.84%, recall of 74.71%, F1-score of 80.75%, and an ROC-AUC of 93.26%. Furthermore, the proposed framework correctly classified 109 real and 78 fake news articles, demonstrating its effectiveness in identifying political misinformation. These findings highlight the potential of lightweight transformer-based language models for developing accurate, scalable, and computationally efficient misinformation detection systems that can support journalists, fact-checking organizations, policymakers, and digital media platforms in strengthening trustworthy democratic communication.