Understanding Information Virality on Social Networks through Predictive Models

Abstract

The rapid adoption of social media has greatly amplified the speed at which information is shared, causing certain events to go viral in ways previously unseen. Through predictive modeling, this paper outlines the processes behind the virality of information on social networks. In particular, we hope to build a model capable of predicting virality by analyzing user interaction data, content characteristics, and time factors. Using social media datasets from Twitter and Facebook, we tested the predictive power of machine learning algorithms such as Random Forest, Support Vector Machines (SVM), and Long Short-Term Memory (LSTM) networks. Our approach includes social media post preprocessing, feature extraction, sentiment analysis, and applying various labeled data classifiers. Results show that information that “goes viral” tends to already have significant emotional engagement from users, is highly relevant, and garners substantial initial interaction. Furthermore, LSTM models surpassed traditional algorithms in terms of temporal parameter sensitivity with a prediction accuracy rate of 87.2%. This work provides insight into the relationship between content and user engagement in virality spread – a relationship that is critically underexplored. For marketers and brand managers, as well as social media platform developers, this study sheds light on how to better target content to optimize exposure. More importantly, it highlights the need for enhanced virality prediction algorithms, especially in the context of misinformation and crisis management communications.