Fine grain emotion analysis in Spanish using linguistic features and transformers

Area of Science:

• Computational linguistics
• Mental health informatics
• Social media analytics

Background:

• Mental health issues, particularly depression, are a growing global concern.
• Social media platforms are increasingly used for sharing health information and seeking support.
• A link exists between emotions expressed online and mental health status, offering potential for early detection.

Purpose of the Study:

• To explore emotion analysis in Spanish for detecting mental health disorders using social media data.
• To create and evaluate a novel, comprehensive Spanish emotion dataset.
• To assess the performance of state-of-the-art transformer models on this dataset.

Main Methods:

• Compilation, translation, and evaluation of a novel dataset with 16 emotions, focusing on negative ones.
• In-depth evaluation of transformer-based models (encoder-only and encoder-decoder).
• Analysis included monolingual, multilingual, and distilled models, alongside feature integration techniques.

Main Results:

• The encoder-only MarIA model achieved the highest performance.
• A macro-average F1 score of 60.4771% was obtained with the MarIA model.
• The study demonstrates the effectiveness of specific transformer architectures for Spanish emotion analysis in mental health contexts.

Conclusions:

• Automated emotion analysis of social media data holds promise for early detection of mental health issues like depression.
• The developed dataset and evaluated models provide a valuable resource for research in this area.
• Further advancements in AI can aid in monitoring and supporting public mental health through social media.