An Ensemble ResNet-XGBoost Model for Assessing User-Generated Content Trends
S. Seethalakshmi, U. Marimuthu, Anju Mohan, K. S. Alakumarimuthu · 2025
The rapid proliferation of user-generated content (UGC) on social media has catalyzed new forms of political expression, notably through parody, memes, and satire. This study proposes a novel Ensemble ResNet-XGBoost model to decode political parody trends in India from 2019 to 2024 by analyzing multimodal data from platforms such as YouTube, Instagram Reels, X (formerly Twitter), and Facebook. Our dataset includes over$\mathbf{1 0, 0 0 0}$annotated samples of image-based and caption-driven satirical content. Visual features were extracted using a pre-trained ResNet-50 model, while textual embeddings were generated using TF-IDF and BERT. These features were fused to train the XGBoost classifier to identify parody categories (e.g., Mockery, Sarcasm, Satirical Praise) and sentiment orientations (Pro-Government, Anti-Government, Neutral). The model achieved an average classification accuracy of 89.6 %, with an AUC-ROC score peaking at 0.97 across sentiment analysis tasks. Among parody categories, Dark Humor and Satirical Praise were the most accurately classified, with$\mathbf{F 1}$-scores of$\mathbf{0. 9 2}$and$\mathbf{0. 9 0}$respectively. This framework presents a robust mechanism for decoding political satire at scale, offering insights into public discourse and digital activism trends in Indian politics.