Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Early Multimodal Prediction of Cross-Lingual Meme Virality on Reddit: A Time-Window Analysis

Sedat Dogan, Nina Dethlefs, Debarati Chakraborty

TL;DR

This paper tackles early cross-lingual meme virality prediction on Reddit with incomplete diffusion data. It introduces a data-driven virality definition based on a hybrid score $HS_{j, final}$ and a leak-free threshold $\tau_{train}$ learned from training data, and evaluates a multimodal feature set across time windows $W = 30$ to $420$ minutes. It compares Logistic Regression, XGBoost, and an MLP, with XGBoost delivering the strongest performance and revealing an evidentiary transition from static context to dynamic engagement as memes unfold. The work provides a large cross-lingual meme dataset and a practical benchmark for early virality prediction where diffusion cascades are not fully observed, supporting proactive moderation, recommendation, and cultural diffusion studies.

Abstract

Predicting the virality of online content remains challenging, especially for culturally complex, fast-evolving memes. This study investigates the feasibility of early prediction of meme virality using a large-scale, cross-lingual dataset from 25 diverse Reddit communities. We propose a robust, data-driven method to define virality based on a hybrid engagement score, learning a percentile-based threshold from a chronologically held-out training set to prevent data leakage. We evaluated a suite of models, including Logistic Regression, XGBoost, and a Multi-layer Perceptron (MLP), with a comprehensive, multimodal feature set across increasing time windows (30-420 min). Crucially, useful signals emerge quickly: our best-performing model, XGBoost, achieves a PR-AUC $>$ 0.52 in just 30 minutes. Our analysis reveals a clear "evidentiary transition," in which the importance of the feature dynamically shifts from the static context to the temporal dynamics as a meme gains traction. This work establishes a robust, interpretable, and practical benchmark for early virality prediction in scenarios where full diffusion cascade data is unavailable, contributing a novel cross-lingual dataset and a methodologically sound definition of virality. To our knowledge, this study is the first to combine time series data with static content and network features to predict early meme virality.

Early Multimodal Prediction of Cross-Lingual Meme Virality on Reddit: A Time-Window Analysis

TL;DR

This paper tackles early cross-lingual meme virality prediction on Reddit with incomplete diffusion data. It introduces a data-driven virality definition based on a hybrid score and a leak-free threshold learned from training data, and evaluates a multimodal feature set across time windows to minutes. It compares Logistic Regression, XGBoost, and an MLP, with XGBoost delivering the strongest performance and revealing an evidentiary transition from static context to dynamic engagement as memes unfold. The work provides a large cross-lingual meme dataset and a practical benchmark for early virality prediction where diffusion cascades are not fully observed, supporting proactive moderation, recommendation, and cultural diffusion studies.

Abstract

Predicting the virality of online content remains challenging, especially for culturally complex, fast-evolving memes. This study investigates the feasibility of early prediction of meme virality using a large-scale, cross-lingual dataset from 25 diverse Reddit communities. We propose a robust, data-driven method to define virality based on a hybrid engagement score, learning a percentile-based threshold from a chronologically held-out training set to prevent data leakage. We evaluated a suite of models, including Logistic Regression, XGBoost, and a Multi-layer Perceptron (MLP), with a comprehensive, multimodal feature set across increasing time windows (30-420 min). Crucially, useful signals emerge quickly: our best-performing model, XGBoost, achieves a PR-AUC 0.52 in just 30 minutes. Our analysis reveals a clear "evidentiary transition," in which the importance of the feature dynamically shifts from the static context to the temporal dynamics as a meme gains traction. This work establishes a robust, interpretable, and practical benchmark for early virality prediction in scenarios where full diffusion cascade data is unavailable, contributing a novel cross-lingual dataset and a methodologically sound definition of virality. To our knowledge, this study is the first to combine time series data with static content and network features to predict early meme virality.

Paper Structure

This paper contains 25 sections, 15 figures, 12 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Definitions and Operationalization of Virality
  4. Early Prediction of Online Content Virality
  5. Feature Engineering and Predictive Modeling
  6. Research Gaps and Our Contribution
  7. Data and Methodology
  8. Dataset Construction and Scope
  9. Target Variable Definition: A Data-Driven Approach
  10. Exploratory Analysis Findings
  11. Feature Engineering
  12. Experimental Setup
  13. Evaluation Metrics
  14. Experiments and Results
  15. Main Results: Performance Across Time Windows
  16. ...and 10 more sections

Figures (15)

  • Figure 1: Examples of memes classified as viral (left, Drake meme template) and non-viral (right, multi-panel comic meme) based on our data-driven definition.
  • Figure 2: Total engagement metrics of all the collected memes over the tracking time.
  • Figure 4: Viral memes lifespan trajectories. Top-left: Individual engagement curves for all viral posts. Top-right: Average viral growth curve. Bottom-left: Distribution of viral take-off times. Bottom-right: Distribution of time to reach peak engagement velocity.
  • Figure 5: Average normalized engagement trajectories (Score, Comments, Crossposts) for posts classified as Viral vs. Non-Viral over the first 500 minutes. Viral posts show distinctly higher and faster-rising engagement across all metrics.
  • Figure 6: Evolution of feature modality importance (count in Top 30 features) for XGBoost models trained at different time windows.
  • ...and 10 more figures