A novel sentiment analysis method based on multi-scale deep learning.

Area of Science:

• Computational Social Science
• Artificial Intelligence
• Natural Language Processing

Background:

• College students are active social media users, making sentiment analysis crucial for understanding public opinion.
• Effective sentiment analysis aids in university ideological management by capturing student community viewpoints.
• Current sentiment analysis models, often based on recurrent neural networks, primarily focus on vertical semantic characteristics, neglecting horizontal ones.

Purpose of the Study:

• To develop a novel sentiment analysis method tailored for college public opinions.
• To address the limitations of existing models by incorporating both vertical and horizontal semantic characteristics.
• To enhance the balance of sentiment analysis models by considering bidirectional semantic features.

Main Methods:

• A multi-scale deep learning approach is proposed for sentiment analysis.
• A sequential neural network with two propagation paths serves as the backbone architecture.
• The model is extended with additional layers in the horizontal direction to capture bidirectional semantic characteristics.

Main Results:

• The proposed method effectively analyzes sentiment in college public opinions.
• Experiments on a real-world social media dataset demonstrate the model's efficiency.
• The multi-scale deep learning approach successfully balances model depth and breadth.

Conclusions:

• The novel sentiment analysis method offers an improved approach for understanding college public opinion.
• The model's ability to capture bidirectional semantic characteristics enhances sentiment analysis accuracy.
• This research contributes to more effective ideological management in universities through better public opinion analysis.