Rumor Detection Project

Project Overview

The Rumor Detection Using Machine Learning project focused on identifying and classifying rumors on social media. Rumors spread quickly, especially during uncertain situations, and can influence public perception. Since manually detecting them is difficult, this project used machine learning and deep learning models to automate the process.

6

80%

Average Accuracy

Models

Research Motivation

Rumor detection is crucial for maintaining social trust and combating misinformation in digital communication, impacting public perception and decision-making in various contexts.

Overview of Methodology

The study used the PHEME dataset, which contains real tweets labeled as rumors or non-rumors, and tested models like Support Vector Classifier (SVC), Decision Tree (DT), Random Forest (RF), XGBoost, Multi-Layered Perceptron (MLP), and Convolutional Neural Networks (CNN) to find the best one for rumor detection.

It also introduced new ways to analyze tweets, including sentiment analysis, a time delay feature measuring how fast people respond to tweets, and user-based factors like verification status and follower count. By comparing model performance through 10-fold cross-validation, the research provided useful insights into how rumors spread and how AI can help detect misinformation more effectively.

Key findings revealed that sentiment and time delay play a role in rumor spread, and some models performed better than others in distinguishing between rumors and non-rumors. The study contributed to misinformation research by demonstrating how AI can be leveraged to improve the accuracy and efficiency of rumor detection on social media platforms.

Overall, the Rumor Detection Using Machine Learning project successfully explored how AI can be used to detect rumors on social media. The project introduced new feature engineering techniques, such as sentiment analysis using VADER, a time delay feature (measuring how quickly a tweet gets its first response), and user-based factors like verification status and follower count. Through 10-fold cross-validation, the study compared the models' performances to determine which approach worked best for rumor detection.