Undergraduate @ Chung-Ang University — triple major in Psychology, English Literature, and Technology Arts.
I came to behavioral data research through an unlikely parallel: studying cognitive psychology and operating systems side by side, I noticed that human cognition and data processing systems share the same fundamental constraints — limited working memory mirrors limited cache, attention bottlenecks mirror I/O bottlenecks, chunking mirrors compression. That connection made me want to understand user behavior the way a systems engineer understands a pipeline: not just what fails, but why, and where in the process.
My research focuses on moments of hesitation, drop-off, and unexpected behavior — because that's where the system is telling you something the happy-path metrics don't capture.
- Behavioral diagnosis — residual analysis, counterfactual audits, drop-off mapping
- Recommendation systems — popularity bias, long-tail exposure, collaborative filtering
- Consumer insights — variance decomposition, momentum effects, discovery gaps
- UX research — usability testing, clickstream analysis, interaction flow design
| Project | What I asked | Key finding |
|---|---|---|
| 🎧 What Drives Music Popularity? | When popularity predictions fail — why? | Residual analysis revealed ~20% of tracks behave unexpectedly; niche genres systematically underrated. Best model R²=0.198, leaving 80% unexplained — that gap is the signal. |
| 🎬 MovieLens Behavioral Recommender | Is popularity bias caused by power users — or is it structural? | Counterfactual removal of top 1–5% users barely shifted head/tail exposure. Built MF from scratch (RMSE 0.9062, NDCG@10 0.9502); designed offline A/B experiment with popularity penalty. |
| 🎪 University Festival App — Behavioral Analytics | Why do our most engaged users leave so abruptly? | Discovered a post-task "completion cliff" in 110K+ clickstream events. Sole designer and UX researcher; shipped design interventions validated with engineering. |
| 🎵 Sonic Analytics Hit Predictor | What actually predicts a song's success? | Artist fame explains only 11% of variance; recent momentum explains 34%. Built and deployed a live Random Forest app with genre-specific "Doctor's Orders" recommendations. · Live app → |
Analysis & Modeling
Python Pandas NumPy Scikit-learn Librosa SQL Statistical Modeling
Visualization & Reporting
Matplotlib Seaborn Tableau Streamlit Jupyter
UX & Research
Figma Usability Testing Interaction Logs A/B Test Design
Workflow
Git GitHub
Roles where understanding why behavior happens matters more than optimizing a metric.
I'm drawn to problems at the intersection of behavioral science and data systems — recommendation research, content ecosystem analysis, consumer insights, and human-AI interaction — where the most important findings live in the residuals, not the predictions.
"The interesting signal is almost never in the average."