Modeling and mitigating human annotation errors to design efficient stream processing systems with human-in-the-loop machine learning.