Beyond Traditional Metrics: Graph Convolution Network for Meme Virality Prediction

View at publisher Summarise with VIRAT
Authors: Roshan Kumar R, Krithigesh J, Mirthul KV, Suraj Prakash Sahoo
Cited by: 0
References: 8

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

The explosion of meme content on social media creates real challenges in understanding what makes the content in the meme go viral. Virality Prediction is a complex task in the field of Social Media Analytics, and Computational Social Science. Traditional Machine Learning (ML) approaches focusing on simple metrics like "upvotes" and "comments" have been used in the literature, but they treat each meme as isolated entities. We propose a Graph Convolutional Network (GCN) framework that uses memes as nodes, by shared metadata like subreddit membership and template similarity. The proposed methodology uses Contrastive Language-Image Pre-training (CLIP) and Residual Networks (ResNet), for multimodal feature extraction. For the Reddit dataset of 1856 memes, our proposed GCN+CLIP architecture achieved a peak accuracy of 82.5% and Macro-F1 score of 0.735, outperforming traditional baseline models. The model predicts virality based on not only content, but also as a function of the social graph and provides pathway for future research into information diffusion.

Fields of Study & Subjects

Complex Network Analysis Techniques Misinformation and Its Impacts Advanced Graph Neural Networks

Publication details

Journal / source
International Conference on Computing, Communication, Control and Cyber-Physical Systems (I5CPS)
Pages
1–6
Publication date
26 February 2026
Language
EN