Fine-grained emotion detection in suicide notes: a thresholding approach to multi-label classification.

Area of Science:

• Computational Linguistics
• Natural Language Processing
• Psychological Informatics

Background:

• Suicide notes contain complex emotional expressions.
• Accurate identification of fine-grained emotions in text is challenging.
• Previous methods often assign only a single emotion label.

Purpose of the Study:

• To develop and evaluate an automated system for identifying emotion-carrying sentences in suicide notes.
• To detect specific, fine-grained emotions conveyed within these notes.
• To improve upon existing emotion detection systems by allowing multiple emotion labels per sentence.

Main Methods:

• Utilized a Support Vector Machine (SVM) classifier to estimate emotion probabilities.
• Implemented a thresholding approach to assign multiple emotion labels to sentences.
• Experimentally determined optimal probability thresholds for emotion assignment.

Main Results:

• The thresholding approach successfully assigned multiple emotion labels to single instances.
• Demonstrated advantages over a system assigning only the most probable label.
• Showcased improved performance compared to a system trained only on emotion-carrying sentences.

Conclusions:

• The developed system effectively identifies and categorizes fine-grained emotions in suicide notes.
• The thresholding method enhances the accuracy of multi-label emotion detection.
• This approach offers a more nuanced understanding of emotional content in sensitive texts.