Analyzing fashion images and trends in online social networks offers a wide range of commercial opportunities, most of which are still completely unexplored. However, despite the enthusiastic approach typical of social network analysis, the fashion domain presents peculiarities that strongly limit automated solutions. Indeed, due to its intrinsic culture and context-dependent subjective character, the perceptions around fashion can significantly vary from one person to another, making it hard to define an “objective” view or perspective.
This work proposes a crowd-powered system for fashion similarity search and trend analysis in Twitter aimed at professionals where the similarity criteria are defined according to human models. This system is specifically designed to support highly subjective environments, allowing the simultaneous coexistence of multiple correct solutions. This approach differs from most others because, instead of using the crowd output to verify an automatic analysis, it considers this as the expected outcome. The problem space is designed according to a flexible approach that assumes different notions of quality and assessment to better support human judgment and personalized criteria.
This work is definitely interesting. The problem is well presented, and the authors’ contribution is clearly explained. Personally, I enjoyed reading this paper for the answers it provides, but also for the ones it does not: open research issues in the field outline exciting possibilities in terms of future work.