Expansion quantization network: A micro-emotion detection and annotation framework.

Area of Science:

• Natural Language Processing
• Artificial Intelligence
• Computational Linguistics

Background:

• Current emotion detection datasets suffer from high costs, subjectivity, and label imbalance.
• Inadequate annotation of micro-emotions and lack of emotional intensity hinder AI's emotional reasoning.
• Existing methods fail to capture the full spectrum of emotions in text, impacting downstream tasks.

Purpose of the Study:

• To develop a novel framework for automatic micro-emotion detection and annotation.
• To address the limitations of manual annotation in emotion detection datasets.
• To introduce a method for representing emotional intensity and uncovering multiple emotions within text samples.

Main Methods:

• Proposed an all-labels and training-set label regression method to map label values to energy intensity levels.
• Developed the Emotion Quantization Network (EQN) framework for micro-emotion detection and annotation.
• Leveraged machine learning capabilities and interdependencies among labels to uncover multiple emotions.

Main Results:

• Validated the broad applicability of the EQN framework across five common sentiment datasets and various NLP models.
• Demonstrated high capability for automatic detection and annotation of micro-emotions on the GoEmotions dataset.
• Achieved the first automatic micro-emotion annotation with energy-level scores.

Conclusions:

• The EQN framework significantly improves micro-emotion detection and annotation accuracy.
• EQN provides strong support for further emotion detection analysis and quantitative emotion computing research.
• This approach enhances AI's ability to understand and reason about complex emotions in text.