Adaptive Dense Ensemble Model for Text Classification.

Area of Science:

• Natural Language Processing
• Machine Learning
• Deep Learning

Background:

• Text classification is a fundamental task in natural language processing.
• Existing methods often struggle with robustness and effective information flow.

Purpose of the Study:

• To propose a novel adaptive dense ensemble model (AdaDEM) for enhanced text classification.
• To improve classification ability, robustness, and generality across diverse datasets.

Main Methods:

• Developed a selective ensemble model using enhanced attention convolutional neural networks (EnCNNs).
• Generated diverse EnCNNs through different sample subsets and kernel granularities.
• Introduced an evaluation criterion in the local ensemble stage (LES) balancing accuracy and diversity.
• Designed an adaptive dense ensemble structure in the global dense ensemble stage (GDES) to optimize information flow and mitigate redundancy.

Main Results:

• AdaDEM demonstrated superior performance against state-of-the-art methods on multiple real-world datasets.
• The model proved effective for both long and short text classification tasks.
• Extensive experiments verified the effectiveness and generality of the proposed AdaDEM approach.

Conclusions:

• The proposed AdaDEM model offers a significant advancement in text classification.
• Its adaptive and ensemble nature provides robust and generalizable performance.
• AdaDEM effectively addresses limitations in current text classification techniques.