Hybrid geo-information processing :crowdsourced supervision of geo-spatial machine learning tasks

Abstract

This paper introduces an approach to crowdsource the supervision of machine learning classification and regression tasks in order to process geo-social media streams. It builds on a review and comparison of four existing approaches to process geo-social media streams in order to identify specific opportunities and challenges. An original conceptual framework situates the machine learning tasks within a geoinformation processing workflow. The paper presents and discusses concrete techniques and software solutions for implementing it. Keywords: crowdsourcing, supervised machine learning, geo-social media streams, user-generated geographic content, volunteered geographic information.