In Search of State and Trait Emotion Markers in Mobile-Sensed Language: Field Study.

Area of Science:

• Psychology
• Computer Science
• Human-Computer Interaction

Background:

• Emotions and mood are crucial for well-being, driving research into continuous emotion prediction.
• Mobile sensing offers a promising avenue for effortless emotion detection through language analysis.

Purpose of the Study:

• To assess the predictive value of mobile-sensed language data for momentary emotional experiences.
• To investigate the utility of language data in predicting individual differences in emotional traits and depression.

Main Methods:

• A 2-week experience sampling study collected emotion ratings, voice recordings, and keyboard activity.
• Correlations were analyzed between emotions, depression, and language features (speech/writing content and form).
• Cross-validation and hold-out sets were used to evaluate predictive performance.

Main Results:

• Weak correlations (up to 0.25) were observed between mobile-sensed language and reported emotions.
• Speech content and form showed the strongest predictive power for state emotions, particularly valence and happiness.
• No significant correlations were found for trait emotions after correction.

Conclusions:

• Mobile-sensed language shows limited promise as a direct emotion marker.
• Current correlations and predictive values (R² ) are low, indicating a need for further research.