Text classification is a fundamental task in natural language processing (NLP) with applications ranging from sentiment analysis to spam detection. Traditional supervised learning methods for text classification often require large labelled datasets, which can be costly and time-consuming to acquire. Active learning, a subfield of machine learning, aims to address this issue by iteratively selecting the most informative data samples for annotation, thus reducing the labelling effort. Additionally, transfer learning has emerged as a powerful technique in NLP, leveraging pre-trained models to enhance performance on downstream tasks with limited labelled data.

In this study, we propose an active learning framework based on transfer learning techniques for text classification, specifically focusing on sentiment analysis of IMDb movie reviews. The IMDb dataset contains a large collection of movie reviews labelled with sentiment polarity (positive or negative), making it a suitable candidate for our investigation. Our approach utilizes a pre-trained language model, such as BERT, fine-tuned on a large corpus of text data. We then employ active learning strategies, such as uncertainty sampling or query by committee, to select the most informative instances for manual annotation, thereby improving the classification performance with minimal labelling effort.

We evaluate the effectiveness of our proposed framework through extensive experiments on the IMDb dataset, comparing the performance of different active learning strategies and transfer learning models. Our results demonstrate that combining active learning with transfer learning significantly reduces the annotation cost while achieving competitive classification accuracy compared to fully supervised approaches. Furthermore, we provide insights into the impact of various factors such as model architecture, selection criteria, and dataset size on the performance of our framework. Overall, our study contributes to the advancement of text classification methodologies by leveraging the synergies between active learning and transfer learning techniques.

Keywords: Text data, text classification, movie reviews, movie sentiment.