Integrating Multiclass Light Weighted BiLSTM Model for Classifying Negative Emotions

Area of Science:

• Social Media Analysis
• Natural Language Processing
• Machine Learning

Background:

• Social networks like Weibo are crucial for public opinion. Analyzing user sentiment has applications in public opinion control, surveys, and recommendations.
• Traditional deep learning models require extensive data for new tasks, limiting their adaptability.
• Accurate classification of negative sentiment on social media is challenging yet valuable.

Purpose of the Study:

• To propose a multiclassification method for microblog negative sentiment detection using MAML and BiLSTM.
• To improve the efficiency and accuracy of sentiment analysis for new tasks with limited data.
• To develop a robust model for understanding user emotions on Weibo.

Main Methods:

• Word vectorization of microblog text.
• Integration of Model-Agnostic Metalearning (MAML) with Bidirectional Extended Short-Term Memory (BiLSTM) networks.
• Parameter updates via machine gradient descent for BiLSTM and meta-learner parameters via second gradient descent in MAML.

Main Results:

• The proposed MAML-BiLSTM model demonstrated improved performance on a Weibo negative sentiment dataset.
• Compared to existing models, precision increased by 1.68%, recall by 2.86%, and F1 score by 2.27%.
• The metalearner enabled rapid iteration for new classification tasks.

Conclusions:

• The MAML-BiLSTM approach offers a more efficient and accurate solution for microblog negative sentiment classification.
• This method effectively addresses the data limitation challenge in traditional deep learning models.
• The findings suggest significant potential for applications in social media monitoring and analysis.