Opinion Mining at Scale: A Case Study of the First Self-driving Car Fatality

摘要

We present a comprehensive pipeline for large-scale opinion mining via a case study of the first self-driving car fatality, in an effort to qualitatively and quantitatively evaluate trending techniques in web searching as well as sentiment analysis. We first perform a scalable and fault-resilient web scraping with a partially-stateful data model. We then apply recent advances in deep learning comparing with a commercial software for sentiment detection. Not only do we measure the performances of the models by numerical metrics, we subsequently align the prediction results with amid economic indices and impactful social events. We further discuss trade-offs of above models from perspectives of both performance improvements of computer systems and accuracy enhancements of machine learning models, and provide deeper insights for stakeholders in the autonomous vehicle industry and the computational social science community.