HAPI: An efficient Hybrid Feature Engineering-based Approach for Propaganda Identification in social media

Area of Science:

• Computational Linguistics
• Social Computing
• Information Science

Background:

• Social media platforms facilitate rapid information sharing, often irrespective of accuracy.
• Propaganda, including fake news and conspiracy theories, is disseminated on these platforms to influence public opinion.
• Effective detection of propaganda is crucial for maintaining information integrity.

Purpose of the Study:

• To introduce a Hybrid Feature Engineering Approach for Propaganda Identification (HAPI) for text-based content.
• To develop and evaluate a machine learning methodology for classifying propaganda and non-propaganda tweets.
• To enhance the accuracy of propaganda detection systems.

Main Methods:

• Collected data from Twitter via API and proposed an annotation scheme for binary classification (propaganda/non-propaganda).
• Employed hybrid feature engineering, combining TF-IDF, Bag of Words, sentimental features, and tweet length.
• Trained and evaluated multiple machine learning classifiers, including SVM, MNB, DT, and LR, using 40 selected features.

Main Results:

• The Support Vector Machine-based HAPI (SVM-HAPI) achieved superior performance with 69.2% overall accuracy, 0.69 precision, 0.69 recall, and 0.69 F-Measure.
• The proposed HAPI approach outperformed most existing methods on several evaluation metrics.
• All evaluated algorithms demonstrated promising results in propaganda detection.

Conclusions:

• The developed HAPI system offers a robust solution for identifying propaganda in textual content.
• The study highlights the effectiveness of combining conventional and machine learning features for propaganda detection.
• Future research should explore deep learning for multimodal propaganda detection, incorporating text, images, and video.