Lollapalooza Chatbot

A data-driven analysis of Lollapalooza's 20-year lineup history (2005–2025), paired with an interactive chatbot that answers natural-language questions about the findings.

Course: TE 440, Spring 2026 · Author: Prisha Singh

What this is

I wanted to know whether Lollapalooza — one of the largest annual music festivals in the country — has become more "mainstream" over its two-decade run, and whether that shift shows up anywhere in Chicago's local economy. This repo has the full analysis plus a chatbot that lets you ask about it directly instead of digging through a report.

Key findings

• Popularity is up. Average Spotify popularity of performers rose roughly 13–15 points from the 2005–2010 era to 2018–2025, with the steepest growth after 2017 — right as Spotify's user base quadrupled globally.
• Genres diversified, but didn't get replaced. Bedroom pop, rap, and electronic acts have grown in the lineup. Rock and indie didn't disappear — they're just sharing the stage with streaming-native genres now.
• Stage and genre predict popularity better than most lineup features. A linear regression (R² = 0.21, 70 features) found the SBC West stage carries the largest single effect (+32.4), with rap close behind (+28.4).
• The festival barely moves the needle on taxi demand. A Bayesian structural time series model of Chicago taxi trips 2013–2023 found Lollapalooza's effect on rides is close to zero — attendees are walking, transit-ing, or rideshare-ing instead.
• Lollapalooza is structurally unique in Chicago's event calendar. Of 593 permitted city events, Block Parties dominate at 38%. Lollapalooza is the only one combining multi-day duration, 150+ performers, and a multi-stage Grant Park footprint.

Data sources

• Lollapalooza Performers 2005–2025 (Kaggle, scraped from Wikipedia)
• City of Chicago Special Events permits (Chicago Data Portal)
• City of Chicago Building Permits (833K rows)
• Chicago taxi trip effect model based on methodology from Lea Wu (2024), Towards Data Science

Methodology

• Cleaning: dropped columns exceeding 50% missingness for regression; imputed remaining numeric nulls with column medians
• Trend analysis: average Spotify popularity by festival year
• Genre analysis: frequency counts on the 1,427 rows with non-null genre data
• Regression: year, stage, and primary genre as predictors (one-hot encoded), 80/20 train-test split, random_state=42
• Clustering: K-Means (k=4, selected via elbow/second-difference method) on Chicago Special Events by date and start time

Built with Python (pandas, scikit-learn, statsmodels, Prophet, matplotlib).

chatbot-app/

An interactive React app powered by the Anthropic Claude API. Rather than querying raw data at runtime, the key findings above — era-level averages, genre frequencies, regression coefficients, cluster breakdowns — are embedded directly in the system prompt, so answers are fast, consistent, and grounded in the actual analysis rather than hallucinated.

Try asking it things like:

• "Which stage maximizes artist popularity?"
• "What's Chicago's most common event type?"
• "Did rock get pushed out of the lineup?"

Repo structure

chatbot-app/         React app + Claude API integration
project milestone 1/ Early data cleaning & exploration
project milestone 2/ Modeling & analysis