Emotion recognition of social media users based on deep learning

Area of Science:

• Natural Language Processing
• Social Media Analytics
• Computational Linguistics

Background:

• Social media sentiment analysis faces challenges with long-distance semantic links and effective feature word capture.
• Current methods often rely heavily on manual annotation, limiting scalability.

Purpose of the Study:

• To develop a user emotion recognition model for analyzing public opinion events on microblogs.
• To enhance the accuracy of emotion categorization in social media text.

Main Methods:

• Utilized linear discriminant analysis (LDA), an emotion dictionary, and manual annotation for feature word extraction.
• Employed Word2vec for word vector conversion and Bidirectional Long Short-Term Memories (BiLSTM) with Convolutional Neural Networks (CNN) for semantic data gathering and feature extraction.
• Focused on three core emotions: joy, anger, and sadness.

Main Results:

• Achieved an average F1 score increase of 3.66% for machine learning models and 1.84% for deep learning models.
• The proposed model demonstrated superior performance in identifying user emotions compared to existing approaches.

Conclusions:

• The developed model effectively addresses limitations in current social media sentiment analysis.
• This research offers a more robust method for microblog public opinion event emotion analysis.