Automatic emotion recognition in healthcare data using supervised machine learning.

Area of Science:

• Computational linguistics
• Psychological informatics
• Machine learning in healthcare

Background:

• Human emotions are crucial for understanding individual behavior and well-being.
• Emotional health significantly impacts overall life satisfaction.
• Detecting emotions in online healthcare data offers valuable insights into patient states.

Purpose of the Study:

• To propose a supervised machine learning method for automatic detection of patient emotions in healthcare data.
• To address the challenge of recognizing patient emotions related to specific diseases from online text.
• To establish a link between negative emotions and psychological health issues.

Main Methods:

• Development of a new dataset, EmoHD, containing 4,202 text samples across eight disease and six emotion classes from online resources.
• Application and comparison of six supervised machine learning models utilizing diverse feature engineering techniques.
• Evaluation of model performance using various feature vectors on the EmoHD dataset.

Main Results:

• The MultiLayer Perceptron model achieved the highest accuracy of 87% among the evaluated state-of-the-art models.
• A detailed comparison highlighted the performance variations of the six machine learning algorithms on the EmoHD dataset.
• The emotional guidance scale confirmed a correlation between negative emotions and psychological health problems.

Conclusions:

• The proposed method effectively automates the detection of patient emotions within healthcare contexts.
• Early detection of negative emotions can help mitigate risks such as suicide, mental disorders, and psychological distress.
• Publicly available implementation details facilitate further research and application in digital mental health.