Active Learning with Visualization

Abstract

Labeled datasets are limited even though data nowadays are produced with an incredible speed. This affects automatic machine learning methods, especially supervised learning, which requires labels to generate valuable information. However, active learning algorithms help to achieve good analytic models with the least labeling efforts, by querying data points to an oracle, which can provide labels. However, users have no control on the instances to be labeled, and for text data, the annotation interface is usually document only. Both of them decrease user experience, and thus, affect the performance of a learning model. Visualization techniques, particularly interactive ones, help to solve this problem. In this thesis, we study the role of visualization in active learning for text classification with an interactive labelling interface. We compare the results of three experiments to show that visualization accelerates high-performance machine learning model constructions with an active learning algorithm.