Deep active learning for multi label text classification.

Area of Science:

• Natural Language Processing
• Machine Learning
• Artificial Intelligence

Background:

• Multi-label text classification (MLTC) assigns multiple labels to text.
• Deep learning models show promise in MLTC but require extensive labeled data.
• Labeling multi-label data is more costly and time-consuming than single-label data.

Purpose of the Study:

• To develop an efficient active learning strategy for deep MLTC models.
• To reduce the dependency on large-scale labeled datasets for training deep MLTC models.

Main Methods:

• Proposing BEAL (Bayesian deep learning and Expected confidence) for deep MLTC.
• Utilizing Bayesian deep learning to obtain the posterior predictive distribution.
• Introducing an expected confidence-based acquisition function for uncertain sample selection.

Main Results:

• BEAL demonstrated efficient model training for deep MLTC.
• The proposed method achieved model convergence with significantly fewer labeled samples.
• Experiments were conducted using a BERT-based MLTC model on benchmark datasets.

Conclusions:

• BEAL effectively enhances the efficiency of deep MLTC model training.
• The method reduces the need for extensive labeled data, making MLTC more accessible.
• BEAL represents a significant advancement in active learning for deep multi-label classification.