Active Learning for Arabic Text Classification

Abstract

Active Learning explores the use of minimal human intervention to improve the efficiency of supervised machine learning algorithms during the learning/training phase. Active learning improves machine learning algorithms performance, especially for ambiguous or unknown cases that are not clearly defined in the classification criteria applied to data. In machine learning, the quality of used data greatly determines the quality of the classification task outcomes. Especially with the current abundance of data resources, the data labeling process represents a major hurdle to data classification. In this paper, we share our results of using active learning approach for Arabic text classification. We demonstrate in this work how active learning approach greatly improves the efficiency of machine learning systems when compared to traditional passive learning approaches. This work introduces our preliminary results of using active learning approach to help annotate the ever-growing Arabic data corpora using state-of-the-art learning techniques.