Effective fashion labelling via image-based recommendation with deep learning

Tremendous increase in the volume of e-commerce data directs recommender systems to be an efficient approach to overcome this overload of information. Recently, deep learning has been applied in varied business fields such as image processing and natural language processing for higher performance. In particular, fashion organisations have started applying deep learning training methods to their online business. Classification of objects/images is the most significant backbone of these applications. Generally, existing techniques depend on traditional features/attributes to characterise an image, for example, the visual properties retrieved by convolutional neural systems. We have developed a two-tier deep learning system that recommends apparel images based on various apparel images passed as input. To accomplish this, a neural network classification system is employed as a visually aware, data-driven extractor of features. The extracted features and the ranking matrix are taken as the input for similarity-based recommendations employing a significant nearest neighbour algorithm.