Fake News Detection and Fact-Checking System

In the digital age, the rapid proliferation of information has transformed the way societies consume news and make decisions.However, this democratization of content distribution has also enabled the widespread dissemination of misleading or false information-Commonly referred to as "fake news."Such misinformation can influence public opinion, undermine trust in reputable institutions, and even affect electoral outComes.As a result, developing robust, automated methods for verifying the authenticity of news content has beCome a critical challenge for researchers, technologists, and policymakers alike.This project addresses the urgent need for an AI-driven solution that not only detects fake news but also provides Comprehensive fact-checking support.By Combining advanced Natural Language Processing (NLP) techniques with powerful Machine Learning (ML) algorithms, our system analyzes textual features-such as word usage patterns, sentiment, and linguistic style-to classify news articles as real or fake.At its core, the detection engine leverages XGBoost, a gradient boosting framework known for its efficiency and high accuracy in classification tasks.Through rigorous text preprocessing (including tokenization, stop-word removal, and stemming) and feature engineering (such as TF-IDF vectorization and domain reputation scoring), the model captures both surfacelevel and semantic nuances that distinguish factual reporting from deceptive content.Beyond classification, the platform integrates multiple AI-powered fact-checking tools-namely Microsoft Copilot, OpenAI's ChatGPT, and Google's Gemini-to cross-verify claims in real time.This multi-tool Approach enhances reliability by synthesizing perspectives from diverse AI models.Additionally, live Google search snippets and references to established fact-checking databases offer users immediate access to corroborating evidence and original source material.A user-centric dashboard, built with Django and Bootstrap, presents results in an intuitive format, allowing for seamless interaction, feedback submission, and model retraining.By uniting detection and verification, the Fake News Detection and Fact-Checking System empowers users to navigate the information landscape with confidence, fosters greater media literacy, and contributes to a more informed, resilient public discourse. II. RELATED WORKNumerous studies have investigated machine learning Approaches for fake news detection, highlighting the importance of both traditional and deep learning methods.A Comprehensive survey of fake news detection techniques classified existing methods into content-based, context-based, and hybrid models, emphasizing that Combining multiple feature categories yields more robust performance cite turn0search2.Iceland (2023) evaluated the generalization capabilities of classical ML models versus state-of-theart deep learning models, finding that traditional algorithms like XGBoost and Random Forest often generalize better to out-ofdistribution data than large language models cite turn0search1.In Comparative experiments on benchmark datasets, XGBoost achieved up to 99.67% accuracy in classifying fake and real news, outperforming other gradient boosting and ensemble methods cite turn0 search0.Deep learning architectures have also been explored extensively: one study Compared LSTM, CNN-GRU, and hybrid CNN-LSTM models across multiple datasets, reporting accuracy rates exceeding 99.9% for certain configurations on the FakeNewsNet dataset cite turn0search3.Another work demonstrated that Combining TF-IDF weighted vectors with Word2Vec embeddings in an ensemble framework improves detection accuracy by capturing both global and contextual word semantics cite turn0search6.Hybrid Approaches integrating feature engineering with deep learning have been shown to offer the best of both worlds, balancing interpretability and performance cite turn0search9.Knowledge graph-based fact verification systems augment textual classifiers by mApping claims onto structured entity relations, enabling multi-hop reasoning for more reliable verification, though coverage limitations remain a challenge cite turn0search4.Recent research has revisited existing datasets by augmenting them with ChatGPT-generated fact-checks, revealing that large language models can serve as preliminary screening tools and help mitigate human bias in truth assessments cite turn1academia10.A "News Verifiers Showdown" study Compared the performance of ChatGPT 3.5 and 4.0, Bing AI, and Bard, finding that GPT-4.0 achieved the highest accuracy (71/100) but still lagged behind human fact-checkers in contextual understanding cite turn1academia11.