Rumour detection using deep learning and filter-wrapper feature selection in benchmark twitter dataset.

Area of Science:

• Natural Language Processing
• Computational Social Science
• Artificial Intelligence

Background:

• Microblogs are a primary news source, but 'readfakes' (synthetic text) fuel online disinformation.
• Unsubstantiated information on social media can mislead the public, necessitating effective rumor detection.
• Early detection of rumors during initial propagation is crucial for assessing veracity and stance.

Purpose of the Study:

• To develop an automatic rumor detection model.
• To propose a hybrid classification model combining deep learning and optimized machine learning.
• To improve the performance of rumor classification compared to existing methods.

Main Methods:

• A hybrid model integrating a Convolutional Neural Network (CNN) for textual feature learning.
• Utilizing an Information Gain-Ant Colony Optimization (IG-ACO) filter-wrapper technique for feature selection.
• Training a Naive Bayes classifier with CNN-learnt features and the IG-ACO optimized feature vector on the PHEME dataset.

Main Results:

• The hybrid model effectively learns textual features using CNN.
• The IG-ACO technique optimizes the feature vector for improved classifier performance.
• The proposed classifier demonstrated superior performance over existing approaches in rumor detection.

Conclusions:

• The hybrid deep learning and optimized Naive Bayes model offers an effective solution for automatic rumor detection.
• Combining CNN for feature extraction with IG-ACO optimized Naive Bayes enhances classification accuracy.
• This research contributes to combating the spread of online disinformation by improving rumor detection capabilities.