Integrating Deep Learning and NLP for Effective Identification of Fake News in Social Media Content Analysis

Authors

  • Karthik Sathyanarayana Bysani

Abstract

Social media is becoming the primary news source globally, but the quick dissemination of false news on these platforms is a major global issue that affects social, political, and economic institutions as well as public perception by evoking feelings like contempt, surprise, and terror. Therefore, identifying and thwarting false news has become essential to maintaining the credibility of information found online. This study draws on Framing Theory and the Diffusion of Misinformation Theory to better understand how deceptive content spreads and is perceived on social platforms. A major limitation of prior research lies in its struggle to accurately capture deep contextual nuances, which this study addresses using the Roberta sequence classifier, known for its rich contextual embeddings. The research utilizes two widely recognized datasets—PolitiFact and Gossip Cop—and implements a robust preprocessing pipeline involving text normalization, tokenization, and class imbalance handling via random oversampling. The purpose of exploratory data analysis (EDA) is to uncover underlying patterns. Standard measures like as accuracy, precision, recall, F1-score, AUC, and confusion matrix are used to assess the refined Roberta model. Results show superior performance, achieving 93.55% accuracy on PolitiFact and 94.19% on Gossip Cop, outperforming models. This work presents an accurate, scalable solution for fake news detection, grounded in theoretical insights and demonstrating enhanced context comprehension and classification performance.
Keywords: Social media, Fake news, Machine learning, Roberta, PolitiFact, and Gossip Cop.

Downloads

Published

2025-09-12

How to Cite

Bysani, K. S. (2025). Integrating Deep Learning and NLP for Effective Identification of Fake News in Social Media Content Analysis. Digital Repository of Theses - SSBM Geneva. Retrieved from https://repository.e-ssbm.com/index.php/rps/article/view/965