Textual emotion classification using MPNet and cascading broad learning.

Area of Science:

• Natural Language Processing (NLP)
• Machine Learning
• Artificial Intelligence

Background:

• Textual emotion classification (TEC) is crucial for understanding human sentiment in text.
• Existing deep learning methods for TEC often face challenges with long training times and convergence issues.
• There is a need for more efficient and effective approaches to emotion detection in text.

Purpose of the Study:

• To propose a novel solution, CBLMP, for textual emotion classification (TEC).
• To address the limitations of traditional deep learning methods in TEC.
• To enhance feature extraction and model performance in emotion detection tasks.

Main Methods:

• Utilized MPNet (masked and permuted pre-trained language model) for generating sentence embeddings with emotional semantic information.
• Employed Cascading Broad Learning (CBL) to improve feature extraction capabilities and model performance.
• Implemented the L-curve model for optimizing regularization parameters to balance under- and over-regularization.

Main Results:

• The proposed CBLMP method demonstrated superior performance in textual emotion classification.
• Experiments conducted on SMP2020-EWECT and SemEval-2019 Task 3 datasets confirmed the effectiveness of CBLMP.
• CBLMP outperformed existing baseline methods in recognizing and detecting emotions within texts.

Conclusions:

• CBLMP offers a promising and efficient alternative for textual emotion classification.
• The integration of MPNet sentence embeddings and CBL significantly enhances emotion detection capabilities.
• The study highlights the potential of novel machine learning approaches in advancing NLP tasks.