Prediction of Social Media Behavior

Why I Did This

I was fascinated by a question that seemed simple on the surface but turned out to be deeply complex: why do some social media posts go viral while others disappear? Everyone has an intuitive sense that timing, content, and audience matter, but I wanted to understand the underlying mechanics quantitatively. What patterns exist in how users interact with content, and can those patterns be predicted?

This was my first real exposure to academic research, and I was drawn to the intersection of human behavior and computational methods. Social media generates massive amounts of data about how people communicate, react, and influence each other. It felt like the perfect domain to apply mathematical modeling and machine learning to understand something genuinely interesting about human nature.

Working on this project also gave me an early introduction to collaborative research. I co-authored the paper with a research mentor, which meant learning how to structure arguments, write for an academic audience, and iterate on ideas through peer feedback. Those skills turned out to be just as valuable as the technical ones.

What We Did

The research combined two complementary approaches to predict social media behavior: Game Theory for modeling strategic interactions between users, and machine learning for pattern recognition in large datasets.

On the Game Theory side, we modeled social media users as strategic agents making decisions about what to post, when to engage, and how to respond to others. Each user’s behavior affects the behavior of others, creating a dynamic system where outcomes depend on the strategies of all participants. We used game-theoretic frameworks to analyze equilibrium states — situations where no individual user benefits from changing their strategy given what everyone else is doing.

For the machine learning component, we applied Convolutional Neural Networks to analyze content features and predict popularity metrics like engagement rates and share counts. CNNs are typically associated with image recognition, but they work well for extracting spatial patterns from structured data representations of social media posts. We also used K-Means clustering to segment users into behavioral groups based on their posting patterns, engagement history, and network characteristics. This segmentation helped us understand that different types of users follow fundamentally different behavioral models.

The combination was powerful. Game Theory gave us a theoretical framework for understanding why users behave the way they do, while machine learning gave us the tools to detect and predict those behaviors at scale. The models were trained and validated on real social media datasets, and we measured prediction accuracy against actual engagement outcomes.

Impact

The paper was published and represented my first contribution to academic research. Looking back, this project was a turning point for me. It introduced me to machine learning at a time when I was just starting to explore programming beyond basic projects, and it showed me that math and computer science could be combined to answer real questions about the world.

Working with CNNs and K-Means at that stage gave me a foundation that I have built on ever since. When I later worked on AI projects like DisasterScope’s verification system or the AI Helmet’s hazard detection, I was drawing on intuitions about model training, feature extraction, and data preprocessing that started with this research.

The academic writing experience was equally formative. Learning how to present findings clearly, support claims with evidence, and structure a logical argument is a skill that shows up in everything from technical documentation to competition presentations. The Conrad Challenge business strategy and the DisasterScope project narrative both benefited from the discipline of academic writing that I developed here.

This was the project that convinced me that I wanted to keep building at the intersection of data, algorithms, and real-world problems. Everything that came after traces back, in some way, to this early research experience.