Platforms such as Youtube that post user-generated videos have received significant negative feedback for posting inappropriate and, at times, offensive content. While machine learning methods can eventually become good enough to identify the content of videos, they may still be challenged when it comes to identifying the emotions they evoke among consumers of such content. Harder still, understanding the nuances of brand management requires not only understanding what may be offensive to certain consumers, but also what may fly in the face of a carefully curated brand image with a particular consumer base (the problem of “misaligned content”). The other alternative is to hire human workers to watch each new video in entirety and identify offensive and inappropriate content. However, this is a time-consuming and expensive operation. In this proposal, we aim to introduce scalable crowdsourcing methods for quickly and efficiently identifying a broad class of subjective reactions (e.g., offensive or misaligned content, humor, horror etc.) by taking into account the diversity of perspectives among viewers and the viewing context itself. We will focus on video-based domains with interconnected events in video and aim to understand the connections/relationships between these events, advertising content, and subjective user judgement. To make these approaches more efficient, we will explore data mining and sampling techniques to minimize the amount of human effort required to make accurate classifications in real time. In essence, the idea is to combine human judgment based on small snippets of each video (and hence avoid having to watch the whole video) with data mining methods to solve the problem. To address the challenges outlined here, our team draws on experts in Crowdsourcing, Marketing, and Data Mining.