Korea Academic Institute of Science and Technology

We present a new methodology for high-quality labeling in the fashion domainwith crowd workers instead of experts. We focus on the Aspect-Based SentimentAnalysis task. Our methods filter out inaccurate input from crowd workers butwe preserve different worker labeling to capture the inherent high variabilityof the opinions. We demonstrate the quality of labeled data based on Facebook'sFastText framework as a baseline.