Master Thesis : Human-centered machine learning for peer-to-peer exchange

Abstract

The Shareish platform project has been created to encourage mutual sharing of goods in the context of gift economy and generalized exchange. One of the critical aspect of the platform is to localize items on a map and to provide them with a detailed description. This however is time consuming and ungrateful to do by hand, to address this problem the use of artificial intelligence (AI) and more specifically computer vision has been explored and integrated in the back-end of the prototype platform. Previously, an item recognition network has been already deployed and the Tesseract optical character recognition engine was used to provide experimental results. This work will expand and refine the integration of machine learning in the platform by introducing metrics, defining tasks, and comparing different networks. Previous work has been done relative to AI integration for the website : an automatic image tagger trained on ImageNet, where some preliminary metric have been computed, fully integrated in the website. Tesseract was also tested as text detection and reading model, the formal comparison between this model and the other new models proposed is one of the subject of this document.