Abstract by Joshua Smith
Sentiment Extraction from Crowd Noise
3M Many scientists have encountered the paramount obstacle of modeling crowd behavior. In this project, we annex part of the greater issue by predicting crowd sentiment from acoustical signals using machine learning methods. The anticipated outcome of the project will provide a proof-of-concept for crowd noise classification. Our data is acoustic and video recordings of crowd response at divers sporting events. This approach entails a number of technical challenges. I discuss our approach to data collection and analysis and explain some of the challenges we face. These challenges include creating a unique regimen for data collection and labeling, identifying isolated events in large data files, extracting relevant acoustical features for each event, and selecting auspicious machine learning blends. The project's ultimate goal is to better understand the acoustical signatures that characterize shifts in a crowd's mood that could accompany abrupt changes in behavior.