Crowd-Labeling Fashion Reviews with Quality Control
This addresses the challenge of cost-effective and reliable labeling for fashion domain sentiment analysis, though it appears incremental as it builds on existing crowd-labeling and filtering techniques.
The paper tackled the problem of obtaining high-quality aspect-based sentiment analysis labels for fashion reviews using crowd workers instead of experts, by filtering inaccurate inputs while preserving opinion variability, and demonstrated its effectiveness by outperforming Facebook's FastText baseline.
We present a new methodology for high-quality labeling in the fashion domain with crowd workers instead of experts. We focus on the Aspect-Based Sentiment Analysis task. Our methods filter out inaccurate input from crowd workers but we preserve different worker labeling to capture the inherent high variability of the opinions. We demonstrate the quality of labeled data based on Facebook's FastText framework as a baseline.