Uncertainty-based Active Learning with Ensemble Technique for Enhancing the Performance of Natural Language Classification with Limited Data

Recently, advances in artificial intelligence have been rapidly driven by the development of large-scale language models, such as GPT-4. These models, trained on more extensive datasets, show remarkable performance across diverse natural language tasks. However, leveraging these models to create effective services can be resource-intensive. Particularly, in addition to the cost of refining and preprocessing data, getting a large amount of data and training them is very challenging. In this paper, we propose an uncertainty-based active learning approach with ensemble technique to enhance the performance of a natural language classification model using limited data. We achieve higher performance with less data regardless of data characteristics and the number of classes.