Related Experiment Videos
A novel hybrid model for identifying the most informative instances for improving text data classification
Amira Abdelwahab1, Mohamed Salama2
1Department of Information Systems, College of Computer Science and Information Technology, King Faisal University, Al-Ahsa, Saudi Arabia.
Plos One
|August 10, 2026
Summary
This study introduces a hybrid active learning framework for text classification, significantly reducing the need for manual data labeling. The model efficiently classifies unlabeled text, offering a cost-effective solution for various domains.
Area of Science:
- Natural Language Processing
- Machine Learning
- Data Science
Background:
- The proliferation of online text data necessitates efficient classification methods.
- Supervised learning is hindered by the high cost of manual data annotation.
- Scalable and accurate text classification is crucial for managing large datasets.
Purpose of the Study:
- To develop a novel hybrid active learning framework for efficient text classification.
- To address the limitations of manual data annotation in supervised learning.
- To evaluate the effectiveness of integrating multiple classifiers with active learning.
Main Methods:
- A hybrid ensemble architecture combining Support Vector Machines, Logistic Regression, Naive Bayes, and Random Forest.
- A pool-based active learning strategy for selecting informative unlabeled instances.
- Feature extraction using Bag-of-Words, TF-IDF, Word2Vec, and BERT embeddings.
Main Results:
- The hybrid active learning model consistently outperformed traditional ensemble classifiers across diverse datasets.
- TF-IDF based hybrid ensembles demonstrated the highest improvements in accuracy, precision, recall, and F1 score.
- The framework showed robustness in imbalanced datasets, enhancing minority class detection.
Conclusions:
- Combining hybrid ensemble learning with active learning provides an effective and cost-efficient text classification solution.
- This approach is suitable for real-world applications with limited labeled data.
- The proposed framework offers a lightweight alternative to purely transformer-based methods.