Comparison of Different Algorithms for Sentiment Analysis: Psychological Stress Notes.

Area of Science:

• Computational linguistics
• Psychology
• Digital health

Background:

• Self-report stress notes collected via smartphone offer a novel data source for mental health research.
• Text analysis algorithms can quantify emotional content within subjective experiences.

Purpose of the Study:

• To compare the performance of three distinct sentiment analysis algorithms: AFINN, Bing, and Syuzhet.
• To evaluate the utility of these algorithms for analyzing stress levels from ecological momentary assessment (EMA) data.
• To provide insights into the nuances of digital stress measurement and management.

Main Methods:

• Utilized a dataset of 1549 self-report stress notes obtained through smartphone collection.
• Applied and compared sentiment analysis scores derived from AFINN, Bing, and Syuzhet algorithms.
• Visualized and statistically compared the outputs of the three algorithms.

Main Results:

• Significant variations were observed in sentiment scores generated by AFINN, Bing, and Syuzhet for the same stress notes.
• The algorithms differed in their sensitivity to specific emotional nuances within the stress narratives.
• Visualizations highlighted distinct patterns of stress expression captured by each algorithm.

Conclusions:

• The choice of sentiment analysis algorithm impacts the interpretation of stress levels derived from text data.
• Understanding algorithm-specific biases is crucial for accurate digital stress measurement.
• Further research is needed to refine algorithms for ecological stress assessment and personalized management strategies.