An Efficient Text Labeling Framework Using Active Learning Model

Abstract

Electronic medical discharge summaries provide a wealth of information. Extracting useful structured information from such unstructured text is challenging. However, supervised machine learning (ML) algorithms can achieve good performance in extracting useful relations between different entities. To use supervised ML techniques, huge annotated datasets are required. Annotating manually is very expensive and time taking due to the requirement of domain experts for annotation. Active learning (AL), a sample selection approach integrated with supervised ML, aims to minimize the annotation cost while maximizing the performance of ML-based models. Active learning leverages the advantage of training the classifier with a limited number of samples but achieving maximum performance. This strategy not only saves time but also decreases the annotation cost involved. Active learning works well with datasets where annotation cost is high, and training a decent classifier with the available annotated dataset is a requirement. The key factor for an active learning model’s success is its selection of samples that needs annotation. The more informative the samples are, the less time it takes to train the supervised model with high accuracy. Thus, the query strategy in sample selection plays a vital role in the AL process. In this study, we aim to develop a novel query strategy to select the most informative samples from the dataset that can eventually accelerate the supervised model’s performance. The query strategy is designed using deep reinforcement learning techniques like actor-critic. The performance of the sample selection strategy is determined by finding the accuracy of the model after a predefined number of iterations.