Learning Emotion Category Representation to Detect Emotion Relations Across Languages.

Area of Science:

• Computational linguistics
• Affective computing
• Psycholinguistics

Background:

• Traditional emotion detection in NLP uses one-hot vectors, failing to represent nuanced relationships between emotion categories.
• Existing methods do not capture the complex, often blurred boundaries between distinct emotion categories.
• Representing emotions as mutually exclusive limits understanding their interconnected nature.

Purpose of the Study:

• To challenge the assumption of mutually exclusive emotion categories in NLP.
• To develop a novel framework and algorithms for learning distributed representations of emotion categories.
• To enable cross-linguistic emotion relation detection using Natural Language Processing (NLP).

Main Methods:

• Developed an innovative framework leveraging soft labels from neural network models.
• Introduced two algorithms for learning distributed representations of emotion categories.
• Utilized computational models to analyze interconnections between emotions across languages.

Main Results:

• Demonstrated the ability of distributed representations to articulate complex emotional connections.
• Achieved the first detection of emotion relations across different languages via NLP.
• Validation experiments confirmed the superiority of the proposed algorithms.

Conclusions:

• The proposed framework and algorithms offer a breakthrough in emotion detection by capturing nuanced inter-category relations.
• This approach facilitates cross-linguistic analysis of emotion, aligning computational findings with psychology and linguistics.
• The work bridges the gap between computational emotion modeling and humanistic understanding of emotions.