Explainable hyper-local demand sensing for fashion retail

Abstract

Stock allocation based on hyper-local market demand is essential to prevent dead inventory in fashion industry. For this purpose, we deployed an AI system which predicts the market demand of a product using sentiment extracted from customer reviews of visually similar products. Our system incorporates deep image and text understanding to auto-extract visually similar products & capture the complex semantics in the noisy reviews. We further map this market sentiment to demographic features of a reviewer's location, thus allowing us to predict the hyper-local market sentiment for any location.