A Multimodel-Based Deep Learning Framework for Short Text Multiclass Classification with the Imbalanced and Extremely...

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Text classification is crucial for applications, but small, imbalanced datasets pose challenges.
• Existing pretrained models are often too large for mobile devices and may not capture deep semantic nuances.
• Efficient and effective text classification methods are needed for real-world, resource-constrained scenarios.

Purpose of the Study:

• To propose a novel multimodel-based deep learning framework for short-text multiclass classification.
• To address the limitations of existing methods concerning dataset size, model deployment, and semantic information extraction.
• To develop a compressed and efficient framework suitable for mobile devices.

Main Methods:

• A five-layer framework incorporating DistilBERT for context-sensitive word vectors.
• Hierarchical feature extraction using word-level and sentence-level bidirectional LSTM networks.
• Max-pooling and a SoftMax layer for dimensionality reduction and classification.

Main Results:

• The proposed framework demonstrates effectiveness on extremely small, imbalanced datasets.
• It achieves state-of-the-art performance in precision, recall, accuracy, and F1 score.
• The model exhibits reduced size, faster training times, and quicker convergence compared to baselines.

Conclusions:

• The developed framework is highly effective for short-text multiclass classification with limited data.
• Its compressed nature and efficient performance make it suitable for deployment on mobile devices.
• This approach offers a practical solution for real-world text classification challenges with scarce data.