Emoji multimodal microblog sentiment analysis based on mutual attention mechanism

Area of Science:

• Natural Language Processing
• Computer Vision
• Social Media Analytics

Background:

• Emojis are vital for conveying emotions and opinions on social media platforms like Sina Weibo.
• Current sentiment analysis methods often overlook the rich visual data within emojis, treating them as mere symbols or text.
• This limitation hinders accurate emotion detection and analysis in microblogging contexts.

Purpose of the Study:

• To propose a novel multimodal information integration model for enhanced emoji microblog sentiment analysis.
• To effectively utilize the visual information inherent in emojis for more nuanced sentiment detection.
• To address the limitations of existing approaches that neglect emoji visual data.

Main Methods:

• Developed a multimodal information integration model specifically for emoji sentiment analysis.
• Incorporated a text-emoji visual mutual attention mechanism to leverage emoji visual features.
• Conducted experiments on a manually annotated microblog dataset.

Main Results:

• The proposed model demonstrated improved performance compared to baseline models that do not use emoji visual information.
• Achieved a 1.37% increase in macro F1 score and a 2.30% increase in accuracy.
• The findings highlight the significance of integrating visual emoji data for sentiment analysis.

Conclusions:

• The novel multimodal model effectively captures and utilizes emoji visual information for sentiment analysis.
• Integrating visual emoji data significantly enhances sentiment analysis accuracy and effectiveness on microblogs.
• The study provides a valuable resource for future research in multimodal sentiment analysis.