AscentX: A System for Crowdsourcing and Annotating with Active Supervision

Abstract

This paper presents a system, AscentX, for labeled data creation and training with active supervision. The aim of system is that it makes collecting data and training models easier. We integrate semi-supervised learning and active learning as well as crowdsourcing learning paradigms. We design a web-based interface for labeling examples based on Python-Django and incorporate our proposed algorithms into it, including semi-supervised classification algorithms, semi-supervised clustering algorithms and with application to data filtering. Moreover, we provide the interfaces for future development including standard datasets attaching, algorithms and active learning models plunging. The system supports the visualization of labeling the examples and analyzing the results of algorithms, which is user-friendly for human annotator. We experimentally verify that AscentX achieves comparable performance to the state-of-the-art systems, using real-life (crawling web data) and benchmark datasets.